JP7515096B2 - Learning method, program, and learning system - Google Patents

Description

The present disclosure generally relates to a learning method, a program, and a learning system. More specifically, the present disclosure relates to a learning method, a program, and a learning system for a classifier that classifies whether an article with an electronic tag is a target for reading.

Patent Document 1 describes the use of an RF (Radio Frequency) tag (electronic tag) reader in a POS (Point Of Sales) device for processing the payment for at least the product (goods) that a user wishes to purchase. The electronic tag is attached to the product and stores price information or identification information of the product to which the electronic tag is attached.

In Patent Document 1, the reading device reads information from an electronic tag when a product is stored in a box-shaped storage chamber made up of six sides of a housing. An opening leading to the inside of the storage chamber is formed in the housing. The opening is opened and closed by a lid. In other words, when the lid is open, a user can place a product into the storage chamber through the opening. To prevent the reading device from reading information from an electronic tag outside the storage chamber, the housing and the lid are made of a material that reflects or absorbs radio waves.

JP 2016-162177 A

The present disclosure aims to provide a learning method, program, and learning system that can easily improve the accuracy of classifying whether an object is a target for reading, regardless of the environment in which the object is placed.

A learning method according to one aspect of the present disclosure includes a first acquisition step, a second acquisition step, a labeling step, a learning step, a first step, a second step, an input receiving step, and a third step. The first acquisition step is a step of acquiring article information related to the article from the first electronic tag of the article by performing wireless communication using radio waves as a medium with a first reading unit and a first electronic tag attached to the article. The second acquisition step is a step of acquiring the article information of the article from the first electronic tag of the article by performing wireless communication using radio waves as a medium with the first electronic tag of the article by a second reading unit different from the first reading unit. The labeling step is a step of assigning a label indicating whether the article is a reading target to the read information of the first electronic tag of the article based on a comparison result between the read information of the first electronic tag of the article obtained by acquiring the article information of the article in the first acquisition step and the read information of the first electronic tag of the article obtained by acquiring the article information of the article in the second acquisition step. The learning step is a step of calculating a judgment criterion of the classifier by machine learning using training data. The classifier classifies whether or not the read target article is the read target. The training data is data including one or more parameters included in the read information of the first electronic tag of the article acquired in the first acquisition step, one or more parameters included in the read information of the first electronic tag of the article acquired in the second acquisition step, and the label assigned in the label assignment step. The first step is a step of acquiring article information on the read target article from the first electronic tag of the read target article by performing wireless communication using radio waves with the first electronic tag of the read target article in the first reading unit. The second step is a step of acquiring the article information of the read target article from the first electronic tag of the read target article by performing wireless communication using radio waves with the first electronic tag of the read target article in the second reading unit. The input receiving step is a step of receiving an input for evaluating a classification result of the classifier after the classifier classifies whether the article to be read is the read target based on a comparison between the read information of the first electronic tag of the article to be read obtained by acquiring the article information of the article to be read in the first step and the read information of the first electronic tag of the article to be read obtained by acquiring the article information of the article to be read in the second step and the judgment criterion. The third step is a step of assigning a label indicating whether the article to be read is the read target to the read information of the first electronic tag of the article to be read based on the input received in the input receiving step. The learning method is executed by one or more processors.
A learning method according to an aspect of the present disclosure includes a first acquisition step, a labeling step, a learning step, a first step, an input receiving step, a second step, and a state acquisition step. The first acquisition step is a step of acquiring article information related to the article from the first electronic tag of the article by performing wireless communication using radio waves as a medium with the first electronic tag attached to the article by a first reading unit. The labeling step is a step of assigning a label indicating whether the article is a read target or not to the read information of the first electronic tag of the article obtained by acquiring the article information of the article in the first acquisition step. The learning step is a step of calculating a judgment criterion of a classifier by machine learning using training data. The classifier classifies whether the read target article is the read target or not. The training data is data including one or more parameters included in the read information of the first electronic tag of the article acquired in the first acquisition step and the label assigned in the labeling step. The first step is a step of acquiring article information on the read target article from the first electronic tag of the read target article by performing wireless communication using radio waves as a medium with the first reading unit and the first electronic tag of the read target article. The input receiving step is a step of receiving an input for evaluating the classification result of the classifier after the classifier classifies whether the read target article is the read target or not based on a comparison between the read information of the first electronic tag of the read target article obtained by acquiring the article information of the read target article in the first step and the judgment criterion. The second step is a step of assigning a label indicating whether the read target article is the read target or not to the read information of the first electronic tag of the read target article based on the input received in the input receiving step. The status acquiring step is a step of acquiring status information indicating the status of the article. The status of the article includes at least the volume of the article or the weight of the article. The explanatory variables input to the classifier include the status information. In the learning step, the judgment criteria of the classifier are learned so as to classify whether the article is the reading target or not based on the state information. The learning method is executed by one or more processors.
A learning method according to an aspect of the present disclosure includes a first acquisition step, a second acquisition step, a labeling step, a learning step, a first step, a second step, an input receiving step, and a third step. The first acquisition step is a step of acquiring article information related to the article from the first electronic tag of the article by performing wireless communication using radio waves with the first electronic tag attached to the article by the first reading unit. The second acquisition step is a step of acquiring read information of the second electronic tag of the moving object by performing wireless communication using radio waves with the second electronic tag attached to the moving object moving together with the article by the first reading unit. The labeling step is a step of assigning a label indicating whether the article is a read target or not based on a comparison result between the read information of the first electronic tag of the article obtained by acquiring the article information of the article in the first acquisition step and the read information of the second electronic tag of the moving object acquired in the second acquisition step. The learning step is a step of calculating a judgment criterion of a classifier by machine learning using training data. The classifier classifies whether the object to be read is the object to be read. The training data is data including one or more parameters included in the read information of the first electronic tag of the object acquired in the first acquisition step, one or more parameters included in the read information of the second electronic tag of the moving object acquired in the second acquisition step, and the label assigned in the label assignment step. The first step is a step of acquiring object information related to the object to be read from the first electronic tag of the object to be read by performing wireless communication using radio waves with the first electronic tag of the object to be read by the first reading unit. The second step is a step of acquiring read information of the second electronic tag of the second moving object by performing wireless communication using radio waves with the second electronic tag attached to the second moving object that moves together with the object to be read by the first reading unit. The input receiving step is a step of receiving an input for evaluating a classification result of the classifier after the classifier classifies whether the article to be read is the read target based on a comparison of the read information of the first electronic tag of the article to be read and the read information of the second electronic tag of the second moving body with the judgment criterion. The third step is a step of assigning a label indicating whether the article to be read is the read target to the read information of the first electronic tag of the article to be read based on the input received in the input receiving step. The learning method is executed by one or more processors.

A program according to one aspect of the present disclosure causes one or more processors to execute the above learning method.

A learning system according to an aspect of the present disclosure includes a first acquisition unit, a second acquisition unit, a label assignment unit, a learning unit, and an input acceptance unit. The first acquisition unit acquires item information related to the item from the first electronic tag of the item by performing wireless communication using radio waves with a first electronic tag attached to the item by a first reading unit. The second acquisition unit acquires the item information of the item from the first electronic tag of the item by performing wireless communication using radio waves with the first electronic tag of the item by a second reading unit different from the first reading unit. The label assignment unit assigns a label indicating whether the item is a read target to the read information of the first electronic tag of the item based on a comparison result between the read information of the first electronic tag of the item obtained by acquiring the item information of the item by the first acquisition unit and the read information of the first electronic tag of the item obtained by acquiring the item information of the item by the second acquisition unit. The learning unit calculates a judgment criterion of a classifier by machine learning using training data. The classifier classifies whether the object to be read is the object to be read. The training data is data including one or more parameters included in the read information of the first electronic tag of the object acquired by the first acquisition unit, one or more parameters included in the read information of the first electronic tag of the object acquired by the second acquisition unit, and the label assigned by the label assignment unit. The first acquisition unit acquires the object information on the object to be read from the first electronic tag of the object to be read by performing wireless communication using radio waves with the first electronic tag of the object to be read by the first reading unit. The second acquisition unit acquires the object information on the object to be read from the first electronic tag of the object to be read by performing wireless communication using radio waves with the first electronic tag of the object to be read by the second reading unit. The input receiving unit receives an input for evaluating a classification result of the classifier after the classifier classifies whether or not the article to be read is the article to be read based on a comparison between the read information of the first electronic tag of the article to be read, which is obtained by acquiring the article information of the article to be read by the first acquiring unit, and the read information of the first electronic tag of the article to be read, which is obtained by acquiring the article information of the article to be read by the second acquiring unit, and the judgment criterion. The labeling unit assigns a label indicating whether or not the article to be read is the article to be read, to the read information of the first electronic tag of the article to be read, based on the input received by the input receiving unit.
A learning system according to an aspect of the present disclosure includes a first acquisition unit, a labeling unit, a learning unit, an input receiving unit, and a state acquisition unit. The first acquisition unit acquires article information related to the article from the first electronic tag of the article by performing wireless communication using radio waves as a medium with a first electronic tag attached to the article by a first reading unit. The labeling unit assigns a label indicating whether the article is a read target to read information of the first electronic tag of the article obtained by acquiring the article information of the article by the first acquisition unit. The learning unit calculates a judgment criterion for a classifier by machine learning using training data. The classifier classifies whether an article to be read is the read target. The training data is data including one or more parameters included in the read information of the first electronic tag of the article acquired by the first acquisition unit and the label assigned by the labeling unit. The first acquisition unit acquires article information on the read target article from the first electronic tag of the read target article by performing wireless communication using radio waves as a medium with the first electronic tag of the read target article in the first reading unit. The input reception unit receives an input for evaluating the classification result of the classifier after the classifier classifies whether the read target article is the read target or not based on a comparison between the read information of the first electronic tag of the read target article obtained by acquiring the article information of the read target article in the first acquisition unit and the judgment criterion. The label assignment unit assigns a label indicating whether the read target article is the read target or not to the read information of the first electronic tag of the read target article based on the input received by the input reception unit. The status acquisition unit acquires status information indicating the status of the article. The status of the article includes at least the volume or weight of the article. The explanatory variables input to the classifier include the status information. The learning unit learns the judgment criteria of the classifier so as to classify whether the article is the reading target or not based on the status information.
A learning system according to an aspect of the present disclosure includes a first acquisition unit, a second acquisition unit, a label assignment unit, a learning unit, and an input acceptance unit. The first acquisition unit acquires article information related to the article from the first electronic tag of the article by performing wireless communication using radio waves with a first electronic tag attached to the article by the first reading unit. The second acquisition unit acquires read information of the second electronic tag of the moving body by performing wireless communication using radio waves with a second electronic tag attached to a moving body that moves together with the article by the first reading unit. The label assignment unit assigns a label indicating whether or not the article is a read target based on a comparison result between the read information of the first electronic tag of the article obtained by acquiring the article information of the article by the first acquisition unit and the read information of the second electronic tag of the moving body acquired by the second acquisition unit. The learning unit calculates a judgment criterion for a classifier by machine learning using training data. The classifier classifies whether or not the read target article is the read target. The training data is data including one or more parameters included in the read information of the first electronic tag of the item acquired by the first acquisition unit, one or more parameters included in the read information of the second electronic tag of the moving body acquired by the second acquisition unit, and the label assigned by the label assignment unit. The first acquisition unit acquires item information related to the item to be read from the first electronic tag of the item to be read by performing wireless communication using radio waves as a medium with the first electronic tag of the item to be read by the first reading unit. The second acquisition unit acquires read information of the second electronic tag of the second moving body by performing wireless communication using radio waves as a medium with a second electronic tag attached to a second moving body that moves together with the item to be read by the first reading unit. The input receiving unit receives the input for evaluating the classification result of the classifier after the classifier classifies whether the article to be read is the article to be read or not based on the read information of the first electronic tag of the article to be read and the read information of the second electronic tag of the second moving body and a comparison with the judgment criterion. The labeling unit, based on the input received by the input receiving unit, assigns a label indicating whether the article to be read is the article to be read or not to the read information of the first electronic tag of the article to be read.

The present disclosure has the advantage of making it easier to improve the accuracy of classifying whether an item is a target for reading or not, regardless of the environment in which the item is placed.

FIG. 1 is a block diagram showing a configuration of a shopping support system including a learning system according to a first embodiment of the present disclosure. FIG. 2 is an explanatory diagram that illustrates a shopping action of a user using the shopping support system. FIG. 3 is a flowchart showing the operation of the learning system in the learning phase. FIG. 4 is a flowchart showing the operation of the learning system in the inference phase. FIG. 5 is an explanatory diagram that illustrates the operation of the shopping support system of the comparative example before changing the layout of the store. FIG. 6 is an explanatory diagram that illustrates the operation of the shopping support system of the comparative example after the layout of the store is changed. FIG. 7 is an explanatory diagram that illustrates a shopping action of a customer using a shopping support system including a learning system according to a modification of the first embodiment of the present disclosure. FIG. 8 is a block diagram showing a configuration of a shopping support system including a learning system according to the second embodiment of the present disclosure. FIG. 9 is an explanatory diagram that shows a model of actions related to shopping by a customer using the shopping support system. FIG. 10 is an explanatory diagram that shows a model of actions related to shopping by a customer using the shopping support system. FIG. 11 is a waveform diagram showing an example of the change over time in the strength of radio waves received from a first electronic tag attached to an article to be read in the learning system. FIG. 12 is a waveform diagram showing an example of the change over time in the strength of radio waves received from a first electronic tag attached to an article that is not the target of reading in the learning system. Figure 13 is a waveform diagram showing an example of the change over time in the strength of radio waves received from the first electronic tags when the first electronic tags attached to the items to be read overlap in the above-mentioned learning system. FIG. 14 is a waveform diagram showing an example of the change over time in the strength of radio waves received from a first electronic tag attached to an item to be read in the learning system according to embodiment 1 of the present disclosure. Figure 15 is a waveform diagram showing an example of the change over time in the strength of radio waves received from the first electronic tags when the first electronic tags attached to the items to be read overlap in the above-mentioned learning system.

(Embodiment 1)
(1) Overview The learning method according to this embodiment is a method for learning the judgment criteria of a classifier 14 (see FIG. 1 ) that classifies whether an article 9 having a first electronic tag 91 attached thereto is a target for reading or not.

In the present disclosure, the "reading target" refers to the object 9 whose item information is to be read, and among the objects 9 whose item information is read by the first reading unit 2A (see FIG. 1), zero or more objects 9 are "reading targets". In other words, not all of the objects 9 whose item information is read by the first reading unit 2A immediately become "reading targets", and the objects 9 whose item information is read by the first reading unit 2A do not necessarily match the objects 9 that are "reading targets". In other words, the first reading unit 2A reads the item information on the object 9 by performing wireless communication with the first electronic tag 91 using radio waves as a medium, so there is a possibility that the first reading unit 2A reads the item information of the object 9 that is not a "reading target". For example, when an object 9 that is not a "reading target" is present in the vicinity of the object 9 that is a "reading target", the first reading unit 2A may read the item information not only of the object 9 that is a "reading target" but also of the object 9 that is not a "reading target". In other words, out of the 10 items 9 whose item information has been read by the first reading unit 2A, none may be the "reading target," and between one and ten items 9 may be the "reading target."

In this embodiment, the item 9 that is the "target to be read" is an item (goods) that is placed by the user 90 in the placement area A11 (described later) on the cash register counter 7 of the store, as shown in FIG. 2. That is, in this embodiment, the item 9 that is the "target to be read" is an item that the user (here, a customer) 90 has picked up from among multiple items 9 displayed in the store and that the user 90 plans to purchase. On the other hand, an item 9 that is not the "target to be read" is an item that is outside the placement area A11 on the cash register counter 7. As an example, an item 9 that is not the "target to be read" is an item that is still displayed on a shelf in the store and has not been picked up by the user 90.

Here, the "user" is not limited to a customer, but may be a store clerk who is trying to calculate the judgment criteria of the classifier 14 through machine learning. In the following, unless otherwise specified, the "user" will be described as a customer.

In this disclosure, "item information" refers to information for identifying an item 9, such as a product identification code, such as the JAN (Japanese Article Number) code used in Japan. In addition to the JAN code, this type of product identification code includes the EAN (European Article Number) code used in Europe and the UPC (Universal Product Code) used in the United States. Furthermore, the item information is not limited to information for identifying the type (category) of the item 9, but may also include information such as serial information for individually identifying items 9 of the same type. In this way, even if there are multiple items 9 of the same type, each of these multiple items 9 of the same type can be specified by the item information. An example of this type of code is the SGTIN (Serialized Global Trade item Number) included in the EPC (Electronic Product/Code) code system.

The learning method of this embodiment includes a first acquisition step, a labeling step, and a learning step.

The first acquisition step is a step in which the first reading unit 2A acquires item information about the item 9 from the first electronic tag 91 by performing wireless communication using radio waves as a medium with the first electronic tag 91 attached to the item 9. Here, the first reading unit 2A is installed in a specified area, such as the cash register counter 7 (see FIG. 2), and in principle performs a process of reading the item information of the item 9 by performing wireless communication using radio waves as a medium with the first electronic tag 91 located within the specified area. In other words, the first reading unit 2A does not read the item information directly from the item 9, but reads the item information contactlessly from the first electronic tag 91 attached to the item 9.

The labeling step is a step in which a label indicating whether or not the item 9 is to be read is added to the read information of the first electronic tag 91 obtained by acquiring the item information.

The "read information" in this disclosure may include, as an example, at least one of the number of reads, time intervals, frequency, strength of received radio waves, and phase of received radio waves by the reading unit (first reading unit 2A) of the electronic tag (first electronic tag 91).

In this disclosure, the "number of reads" refers to the number of times that the reading unit (first reading unit 2A) reads the item information within a certain period of time, and more specifically, the number of times that the reading unit successfully reads the item information. In this disclosure, the "time interval between reads" refers to the time interval between the reading unit reading the item information within a certain period of time. More specifically, it is the time from the time when the first reading of the item information is performed to the time when the second reading of the item information is performed when the reading unit successfully reads the item information twice. In this disclosure, the "frequency of reading" refers to the frequency at which the reading unit reads the item information within a certain period of time, and more specifically, the frequency at which the reading unit successfully reads the item information. In this disclosure, the "strength of the received radio wave" refers to the strength of the radio wave transmitted from the electronic tag (first electronic tag 91) received by the antenna 21 (described later), and is expressed as an example by the RSSI (Received Signal Strength Indication) value. In this disclosure, the "phase of the received radio waves" refers to the phase of the radio waves transmitted from the electronic tag and received by the antenna 21.

The learning step is a step in which the judgment criteria of the classifier 14 are calculated by machine learning using training data. The training data is data including one or more parameters included in the read information of the first electronic tag 91 acquired in the first acquisition step and the label assigned in the label assignment step. The classifier 14 classifies whether or not the item 9 is a target for reading based on a comparison between the read information of the first electronic tag 91 and the judgment criteria. Specifically, when one or more parameters are input from the read information of the first electronic tag 91 as explanatory variables, the classifier 14 compares the input one or more parameters with the judgment criteria and outputs binary information indicating whether or not the item 9 to which the first electronic tag 91 is attached is a target for reading as a target variable.

Here, it is assumed that the item 9 that is the "target to be read" is placed in a position where the item information can be easily read by the first reading unit 2A (e.g., placement area A11). The item information of an item 9 placed in an easy-to-read position is easier to read by the first reading unit 2A than that of other items 9. For this reason, there may be a significant difference between the read information of the item 9 that is the "target to be read" and the read information of the other items 9 that distinguishes them from each other. Therefore, the classifier 14 is able to distinguish between the item 9 that is the "target to be read" and the other items 9 based on the read information of the first electronic tag 91.

As described above, in this embodiment, it is possible to learn the judgment criteria of the classifier 14 using training data based on the read information of the first electronic tag 91. Therefore, in this embodiment, even if the environment in which the item 9 is placed is changed, it is possible to classify whether or not the item 9 is a target for reading even in the changed environment by readjusting the judgment criteria of the classifier 14 using the training data. In other words, this embodiment has the advantage that it is easy to improve the accuracy of classifying whether or not the item 9 is a target for reading, regardless of the environment in which the item 9 is placed.

(2) Details A learning system 100 for implementing the learning method of the present embodiment, and a shopping support system 200 including the learning system 100 will be described in detail below with reference to Figures 1 and 2. In this embodiment, the learning system 100 will be described as being incorporated into the shopping support system 200, but it may be configured as a system separate from the shopping support system 200.

(2.1) Shopping Support System First, an overview of the shopping support system 200 will be described. As shown in Fig. 1, the shopping support system 200 includes a reading device 5 and a payment system 6. The reading device 5 is used to obtain item information of an item 9. The payment system 6 performs a payment process for the item 9 using the item information obtained by the reading device 5.

The shopping support system 200 is a system that is introduced to, for example, a convenience store, supermarket, department store, drug store, electronics retailer, or home center, and supports the purchase of goods (i.e., "shopping") by the user 90. In this disclosure, "purchase" refers to an action on the part of the buyer (user) in an act (sales transaction) in which the ownership of an item 9 is transferred from the seller (store) to the buyer (user) and the buyer pays the seller a price (money) for the item. Therefore, when a reading device 5 is used in the shopping support system 200 introduced in a store, the item 9 from which item information is obtained using the reading device 5 is a "good" sold in the store, and the item information is "item information." In this embodiment, a convenience store is used as an example of a store in which the shopping support system 200 is introduced.

In this type of store, multiple items 9 are displayed in the store and sold. Therefore, a user 90 picks out a desired item 9 from the multiple items 9 displayed in the store and purchases the desired item 9 by making a payment (settlement) for the picked item 9. In this embodiment, a staffed store with at least one store clerk is assumed. However, this is not the only example, and the shopping support system 200 can also be used in situations such as an unmanned store with no store clerk.

When the reading device 5 is used in the shopping support system 200, the item information acquired by the reading device 5 is used for payment processing in the payment system 6. That is, the payment system 6 performs payment processing for the item 9 about which the reading device 5 has acquired the item information, using the item information. Therefore, according to the shopping support system 200, the item information is read by the reading device 5 installed in the store, and payment processing for the item 9 becomes possible.

In this embodiment, the reading device 5 is housed in the cash register counter 7 of the store. The upper surface of the cash register counter 7 constitutes a placement area A11 on which the item 9 to be purchased by the user 90 (i.e., the item to be read) is placed. The reading device 5 acquires item information from the item 9 in the placement area A11 without contact. Therefore, the user 90 picks up the item 9 to be purchased from among the multiple items 9 displayed in the store and places it on the placement area A11 to cause the reading device 5 to acquire the item information. In this embodiment, the user 90 places a shopping basket (mobile body) 8 containing the item 9 to be purchased on the placement area A11 to cause the reading device 5 to acquire the item information. Then, the payment process is completed in the payment system 6 using the item information on the item 9 to be read out of the item information acquired by the reading device 5, and the ownership is transferred from the seller (store) to the buyer (user), and the purchase of the product by the user 90 is completed. The user 90 will take home the purchased item 9, i.e., the item 9 for which the payment process has been completed.

The payment system 6 is configured to be able to communicate with the reading device 5. An appropriate communication method such as wireless communication or wired communication is adopted as the communication method between the payment system 6 and the reading device 5. Furthermore, the payment system 6 is configured to be able to communicate directly or indirectly via a repeater or the like with a store terminal such as a POS (Point Of Sales) terminal. In this embodiment, the payment system 6 is housed in the cash register counter 7, just like the reading device 5.

The payment system 6 can present various information to the user 90 by display or audio, and can accept operations (including audio input) by the user 90, by using the user interface 4 of the reading device 5 described below. However, the presentation of information by the payment system 6 needs to be realized by at least one of display and audio, and may be realized by either one of display and audio, or may be realized by a combination of display and audio.

As an example, the payment system 6 can present messages such as "Welcome" and "Thank you" to the user 90 when a visitor is detected and when the payment process is completed. This can give the user 90 a sense of familiarity similar to that of a store clerk serving customers. The user 90 can also, for example, check the list of items 9 to be purchased displayed on the user interface 4 and the payment amount (settlement amount), and after checking, operate the user interface 4 to express his/her intention to accept, thereby performing the payment process in the payment system 6. The payment process may be realized, for example, by short-range wireless communication between a mobile information terminal (smartphone or wearable terminal, etc.) or an IC (Integrated Circuit) card, etc., and the payment system 6. In this case, the user 90 causes the payment system 6 to perform the payment process by bringing the mobile information terminal or IC card, etc., close to a predetermined location (for example, near the user interface 4). Here, the payment system 6 may have a printer, and in that case, when the payment process is completed, a receipt, etc. can be issued (printed) by the printer. However, accepting operations by the user 90 during payment processing and issuing receipts are not essential components of the shopping support system 200.

In addition, the list of items 9 to be purchased and information such as the payment amount (hereinafter referred to as "purchase information") may be transmitted from the payment system 6 to a mobile information terminal (such as a smartphone or tablet terminal) owned by the user 90, or to a store terminal, etc. This makes it possible for the user 90 to view the purchase information or issue a receipt on a mobile information terminal owned by the user 90 or an information terminal other than the payment system 6. Furthermore, by transmitting the purchase information from the store terminal, etc. to a server, the user 90 can view the purchase information on the mobile information terminal owned by the user 90 and have the mobile information terminal issue a receipt, even after leaving the store.

In this embodiment, the payment system 6 is a computer system whose main components are one or more processors and memory as hardware. In this computer system, the functions of the payment system 6 are realized by executing a program recorded in the memory with one or more processors. The program may be pre-recorded in the memory of the computer system, may be provided via a telecommunications line, or may be recorded and provided on a non-transitory recording medium such as an optical disk or hard disk drive that can be read by the computer system.

(2.2) Reading Device Next, the configuration of the reading device 5 will be described with reference to Fig. 1. The reading device 5 includes a first reading unit 2A, a main computer 10, a detection unit 3, and a user interface 4.

The first reading unit 2A is configured to acquire the item information by reading the item information from the first electronic tag 91 attached to the item 9. That is, the first reading unit 2A acquires the item information stored in the first electronic tag 91 by performing wireless communication using radio waves as a medium between the first reading unit 2A and the first electronic tag 91 attached to the item 9.

The first electronic tag 91 is, for example, a passive RF tag, and has at least a memory for storing item information. Here, a plurality of first electronic tags 91 are associated one-to-one with a plurality of items 9. The first electronic tag 91 stores item information about the corresponding item 9, and the first electronic tag 91 is attached to the corresponding item 9.

The first electronic tag 91 may be attached to the item 9 in a state where it can be handled together with the item 9, and there are various specific aspects in which the first electronic tag 91 is attached to the item 9. In the present embodiment, as an example, the first electronic tag 91 is in the form of a sticker and is attached to the item 9. In addition, the first electronic tag 91 may be connected to the item 9 by, for example, a string or the like, integrated into the packaging material of the item 9, embedded in the item 9, or incorporated into the item 9. Furthermore, for example, by using a technology such as coated semiconductor, the first electronic tag 91 may be directly formed by printing on the surface of the item 9 itself or the packaging material of the item 9.

As shown in FIG. 2, the first reading unit 2A according to this embodiment is formed in a rectangular plate shape and is installed below the placement area A11 of the register counter 7. The first reading unit 2A has an antenna 21 and a communication unit 22 inside (see FIG. 1), and is a reader that constitutes an RFID (Radio Frequency Identification) system. The first reading unit 2A basically performs wireless communication with the first electronic tag 91 of the item 9 placed in the placement area A11 above the first reading unit 2A. In other words, the first reading unit 2A is installed facing the first area A1, and the first area A1 includes the placement area A11 on which the item 9 to be read is placed. In this embodiment, the first reading unit 2A is a so-called open-type reading device that does not have a shield or the like to reduce leakage of radio waves from the placement area A11.

The antenna 21 transmits and receives radio waves as a communication medium between the first electronic tag 91 located in the placement area A11 on the first reading unit 2A and performs wireless communication. That is, in this embodiment, the antenna 21 is positioned so that it can transmit and receive radio waves from below the processing area. Taking into account the polarization plane, it is preferable that the antenna 21 is configured as a circularly polarized antenna. The antenna 21 is electrically connected to the communication unit 22. The antenna 21 is configured as, for example, a patch antenna, a monopole antenna, an inverted F-shaped antenna, or a slot antenna.

However, in this embodiment, the first reading unit 2A is an open type, and therefore no particular structure is provided to restrict the range of the radio waves transmitted from the antenna 21. Therefore, the radio waves for communication emitted from the antenna 21 may reach an item 9 located outside the placement area A11, such as a product placed near the cash register counter 7. Therefore, the first reading unit 2A may read item information from an item 9 located outside the placement area A11, such as a product placed near the cash register counter 7.

The communication unit 22 transmits radio waves from the antenna 21 to the first electronic tag 91, and receives the article information from the first electronic tag 91 activated by the radio waves at the antenna 21. The communication unit 22 measures the received signal strength of the wireless signal (radio waves) at least when receiving the article information. When the communication unit 22 receives the article information from the first electronic tag 91, it transmits the received signal strength to the main computer 10 in addition to the article information.

The detection unit 3 detects the position of the item 9. As will be described in detail later, when the detection unit 3 detects that the item 9 has entered an area (peripheral area A21) at a certain distance from the register counter 7 and that the item 9 has been placed in the placement area A11, wireless communication with the first electronic tag 91 in the reading device 5 is initiated. In other words, upon receiving a trigger from the detection unit 3, the main computer 10 causes the control unit 15 (described later) to initiate wireless communication with the first electronic tag 91 in the reading device 5. In this embodiment, the detection unit 3 has a first sensor 31 and a second sensor 32.

The first sensor 31 is realized, for example, by a distance measurement sensor that measures the distance from the register counter 7 to the user 90. Specifically, the first sensor 31 is realized, for example, by a non-contact distance measurement sensor such as an optical, ultrasonic, or radio wave type. The first sensor 31 detects that the item 9 has entered an area (surrounding area A21) at a certain distance from the register counter 7 by measuring the distance from the register counter 7 to the user 90 carrying the item 9 (or a shopping basket 8 containing the item 9).

The second sensor 32 is a weight sensor that measures the weight of an object placed in the placement area A11. The second sensor 32 does not measure the weight of an object if an item 9 (or a shopping basket 8 containing an item 9) is not placed in the placement area A11, but measures the weight of an object if an item 9 (or a shopping basket 8 containing an item 9) is placed in the placement area A11. Therefore, the second sensor 32 detects whether an item 9 is present in the placement area A11 depending on whether the weight of the object has been measured.

That is, the detection unit 3 detects the presence or absence of at least one of an item 9 in the detection range (here, the placement area A11 and the surrounding area A21) and a moving object 8 moving with the item 9 (here, a shopping basket). In other words, in this embodiment, the learning method further includes a detection step of detecting the presence or absence of at least one of an item 9 in the detection range and a moving object 8 moving with the item 9. And, the detection unit 3 is the entity that executes the detection step.

In this embodiment, the detection range includes the placement area A11 and the surrounding area A21. As already mentioned, the placement area A11 is the area where the item 9 to be read is placed. The surrounding area A21 is an area at a certain distance from the register counter 7, that is, an area up to a position a predetermined distance away from the placement area A11.

The user interface 4 has the function of presenting various information to the user 90 by display or sound on the reading device 5, and accepting operations (including voice input) by the user 90. In this embodiment, the user interface 4 has a display unit 41 and an operation unit 42. Furthermore, in this embodiment, the display unit 41 and the operation unit 42 are integrated as a touch panel display. As an example, the user interface 4 is installed on the register counter 7 so as to face the user 90 (forward). The user interface 4 is not limited to display and sound, and may present information by, for example, the lighting state of a lamp (on, off, light color, blinking pattern or brightness, etc.) or vibration using a vibrator.

The main computer 10 is a computer system whose main components are one or more processors and memory as hardware. In this computer system, the functions of the main computer 10 are realized by executing a program recorded in the memory with one or more processors. The program may be pre-recorded in the memory of the computer system, may be provided via a telecommunications line, or may be recorded and provided on a non-transitory recording medium such as an optical disk or hard disk drive that can be read by the computer system.

(2.3) Learning System Next, the configuration of the learning system 100 will be described in detail. In this embodiment, the main computer 10 constitutes the learning system 100. The main computer 10 has an acquisition unit 11, a label assignment unit 12, a learning unit 13, a classifier 14, a control unit 15, a storage unit 16, and an input reception unit 17. However, the learning system 100 does not need to have all of the components of the main computer 10, and it is sufficient that the learning system 100 has at least an acquisition unit (first acquisition unit) 11A, a label assignment unit 12, and a learning unit 13.

The acquisition unit 11 acquires multiple pieces of article information about multiple items 9 from the first reading unit 2A. As already described, when wireless communication with the first electronic tag 91 attached to the item 9 is established, the first reading unit 2A receives the article information about the item 9 to which the first electronic tag 91 is attached from the first electronic tag 91. The acquisition unit 11 acquires from the first reading unit 2A the article information that the first reading unit 2A reads (receives) from the first electronic tag 91 in this manner. That is, in this embodiment, the acquisition unit 11 is the entity that executes the first acquisition step.

Here, when the first reading unit 2A simultaneously reads the item information for multiple items 9, the acquisition unit 11 acquires the item information for each item 9. The acquisition unit 11 acquires the item information from the first reading unit 2A at regular time intervals (e.g., several milliseconds) and stores it in the memory unit 16.

The classifier 14 classifies whether the item 9 is a target to be read or not based on a comparison between the read information of the first electronic tag 91 and a judgment criterion. Specifically, when one or more parameters from the read information of the first electronic tag 91 are input as explanatory variables, the classifier 14 compares the input one or more parameters with the judgment criterion and outputs binary information indicating whether the item 9 to which the first electronic tag 91 is attached is a target to be read or not as a target variable.

The classifier 14 may include a linear classifier such as a Support Vector Machine (SVM), a classifier using a neural network, or a classifier generated by deep learning using a multi-layer neural network. In this embodiment, the classifier 14 is a classifier using a trained neural network. The trained neural network may include, for example, a Convolutional Neural Network (CNN) or a Bayesian Neural Network (BNN). The classifier 14 is realized by implementing a trained neural network in an integrated circuit such as an Application Specific Integrated Circuit (ASIC) or a Field-Programmable Gate Array (FPGA).

The judgment criteria of the classifier 14 are obtained by machine learning using a large amount of training data in the learning unit 13. In this disclosure, "machine learning" may include not only learning performed on an untrained classifier 14, but also re-learning performed on a trained classifier 14. The "training data" here is a data set that combines the input information input to the classifier 14 and the labels assigned to the input information.

The "input information" is an explanatory variable, which is one or more parameters included in the read information of the first electronic tag 91. The explanatory variable may include, as an example, the change over time in the number of times the first electronic tag 91 is read by the first reading unit 2A, or the change over time in the strength of the received radio waves (RSSI value). The explanatory variable may also include, as an example, the variance in the number of times the first electronic tag 91 is read by the first reading unit 2A, the strength (RSSI value) or strength variance of the received radio waves, or the phase or phase variance of the received radio waves.

A "label" is data that represents the objective variable that the classifier 14 should output for the explanatory variables input to the classifier 14, that is, the correct answer that the classifier 14 should output. In this embodiment, the label is data that represents whether or not the item 9 is to be read. As an example, in training data for an item 9 to be read (i.e., an item 9 placed in the placement area A11), the label is data that represents that the item 9 is to be read. Also, in training data for an item 9 that is not to be read (i.e., an item 9 outside the placement area A11), the label is data that represents that the item 9 is not to be read.

In this embodiment, the input information does not include a product identification code such as an EPC. In other words, in this embodiment, the explanatory variables input to the classifier 14 do not include information that identifies the item 9. Here, if the input information includes a product identification code, the classifier 14 may be trained to determine whether or not the item 9 is a target for reading, depending on the information that identifies the item 9. In other words, by excluding the information that identifies the item 9 from the input information, it is possible to prevent the classification result of the classifier 14 from being influenced by the information that identifies the item 9.

The label assignment unit 12 assigns a label indicating whether or not the item 9 is to be read to the read information of the first electronic tag 91 obtained by acquiring item information by the acquisition unit 11. That is, in this embodiment, the label assignment unit 12 is the entity that executes the label assignment step. A data set including the label assigned by the label assignment unit 12 and the read information of the first electronic tag 91 to which the label is to be assigned is stored in the storage unit 16 as training data.

In this embodiment, the label assignment unit 12 assigns a label to the read information of the first electronic tag 91 in a different manner between the learning phase and the inference phase. The learning phase is a phase in which the judgment criteria of the classifier 14 are calculated by machine learning during a period in which the classifier 14 is not used. As an example, the learning phase is executed at the stage of introducing the shopping support system 200 to a new store or at the stage of changing the layout of the store. The inference phase is a phase in which the classifier 14 is used to classify whether or not the item 9 is a target for reading. In this embodiment, the judgment criteria of the classifier 14 can be updated by machine learning (relearning) even in the inference phase.

In the learning phase, the labeling unit 12 compares the item information acquired by the acquisition unit 11 with the list stored in the memory unit 16 to determine the label to be assigned to the read information of the first electronic tag 91. One example of the item information is a product identification code such as EPC. The list includes item information of the item 9 to be read (i.e., the item 9 placed in the placement area A11). When the item information is included in the list, the labeling unit 12 assigns a correct label indicating that the item 9 is the item to be read to the read information of the first electronic tag 91 corresponding to this item information. On the other hand, when the item information is not included in the list, the labeling unit 12 assigns an incorrect label indicating that the item 9 is not the item to be read to the read information of the first electronic tag 91 corresponding to this item information.

In the inference phase, the labeling unit 12 determines a label to be assigned to the read information of the first electronic tag 91 based on the evaluation of the classification result of the classifier 14 received by the input receiving unit 17 (described later). In other words, in this embodiment, the labeling step determines a label based on the input received by the input receiving unit 17 (i.e., the input receiving step). For example, it is assumed that the classification result of the classifier 14 for a certain item 9 is "read target" and the input receiving unit 17 receives an input that this evaluation is correct. In this case, the labeling unit 12 assigns a label indicating that the item 9 is a read target to the read information of the first electronic tag 91 attached to this item 9. It is also assumed that the classification result of the classifier 14 for a certain item 9 is "read target" and the input receiving unit 17 receives an input that this evaluation is incorrect. In this case, the labeling unit 12 assigns an incorrect label to the read information of the first electronic tag 91 attached to this item 9, indicating that the item 9 is not the item to be read.

In this embodiment, the label attachment unit 12 determines the label based on the period from when wireless communication between the first reading unit 2A and the first electronic tag 91 is started by the control unit 15 until when it is ended. As already described, this period is determined based on the detection results of the first sensor 31 and the second sensor 32 of the detection unit 3. In other words, the execution period of the label attachment step is determined based on the detection result of the detection step.

The learning unit 13 uses the training data to calculate the judgment criteria of the classifier 14 through machine learning. The training data is data including one or more parameters included in the read information of the first electronic tag 91 acquired by the acquisition unit 11 (first acquisition step) and the label assigned by the label assignment unit 12 (label assignment step). In other words, in this embodiment, the learning unit 13 is the entity that executes the learning step. The learning unit 13 is composed of one or more processors. The "processor" here may include, for example, a general-purpose processor such as a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit), as well as a dedicated processor specialized for calculations in a neural network.

In this embodiment, the learning unit 13 uses a large amount of training data to perform machine learning of the neural network used in the classifier 14, thereby performing machine learning of the judgment criteria of the classifier 14. For each of the multiple training data, the learning unit 13 inputs input information to the input layer of the neural network and performs a calculation. Then, the learning unit 13 performs backpropagation processing using output values of multiple neurons in the output layer of the neural network and labels corresponding to the input information. Here, the multiple neurons in the output layer each correspond to multiple possible states of the item 9. The "multiple states" here include a state in which the item 9 is a reading target and a state in which the item 9 is not a reading target.

In the backpropagation process, the learning unit 13 updates the weighting coefficients of the neural network so as to maximize the output value of the neurons in the output layer that correspond to the labels included in the training data. The learning unit 13 optimizes the weighting coefficients of the neural network by performing the backpropagation process on all the training data. This completes the learning of the neural network.

The control unit 15 issues instructions to start and end wireless communication with at least the first reading unit 2A and the first electronic tag 91. Specifically, when starting wireless communication between the first reading unit 2A and the first electronic tag 91, the control unit 15 controls the first reading unit 2A to start outputting radio waves for wireless communication from the first reading unit 2A. In addition, when ending wireless communication between the first reading unit 2A and the first electronic tag 91, the control unit 15 controls the first reading unit 2A to stop outputting radio waves for wireless communication from the first reading unit 2A.

In this embodiment, when the first sensor 31 of the detection unit 3 detects that an item 9 has entered the surrounding area A21, the control unit 15 controls the first reading unit 2A to start outputting radio waves for wireless communication from the first reading unit 2A. Also, in this embodiment, the control unit 15 controls the first reading unit 2A to stop outputting radio waves for wireless communication from the first reading unit 2A when a predetermined time has elapsed since the second sensor 32 of the detection unit 3 detected the item 9 (or a shopping basket 8 containing the item 9).

The storage unit 16 is realized by a non-transitory recording medium such as a rewritable non-volatile semiconductor memory. The storage unit 16 stores the item information of the first electronic tag 91 acquired by the acquisition unit 11. The storage unit 16 also stores read information of the first electronic tag 91 to which a label has been assigned by the label assignment unit 12, that is, training data. The storage unit 16 also stores the judgment criteria of the classifier 14 learned by the learning unit 13.

The input receiving unit 17 receives an input for evaluating the classification result of the classifier 14. In other words, the learning method of this embodiment has an input receiving step for receiving an input for evaluating the classification result of the classifier 14. The input receiving unit 17 is the entity that executes the input receiving step. In this embodiment, the input is performed by the operation unit 42 of the user interface 4. Specifically, the display unit 41 of the user interface 4 displays, for each item 9 classified by the classifier 14, the classification result of the classifier 14, a message inquiring whether the classification result is correct, and a first icon that is operated when the classification result is incorrect. In addition, the display unit 41 displays a second icon that is operated when the evaluation of the classification result is completed.

If the classification result of the classifier 14 is incorrect for any of the items 9, the user 90 operates the operation unit 42 to select a first icon on the display unit 41 that corresponds to the relevant item 9. Then, when the user 90 has completed evaluation of the classifier 14 for all of the items 9, the user 90 operates the operation unit 42 to select a second icon on the display unit 41. This causes the input receiving unit 17 to receive evaluations of the classification results of the classifier 14 for all of the items 9 classified by the classifier 14. The user 90 who operates the operation unit 42 may be a customer or a store clerk, etc.

(3) Operation An example of the operation of the shopping support system 200 of this embodiment and an example of the operation of the learning system 100 of this embodiment, in other words, an example of a learning method, will be described below.

(3.1) Learning Phase First, the operation of the learning system 100 in the learning phase will be described. Machine learning in the learning phase is performed, for example, in a store that introduces the shopping support system 200. In the store, machine learning of the neural network used in the classifier 14 is performed using the learning unit 13. When machine learning is performed at the stage of introducing the shopping support system 200 to a new store, it is preferable that the weighting coefficients of the neural network are initialized. On the other hand, when the shopping support system 200 has already been introduced in a store and machine learning is performed at the stage of changing the layout of the store, the weighting coefficients of the neural network may be initialized or may remain as they are.

As shown in FIG. 3, a list including the item information of the item 9 to be read (i.e., the item 9 placed in the placement area A11) is stored in the memory unit 16 (S1). Next, the items 9 to be read that are listed in the list are placed in the placement area A11 one by one. The items 9 may be placed in the placement area A11 one by one, or multiple items may be placed in the placement area A11. Then, when the detection unit 3 (here, the second sensor 32) detects that the item 9 has been placed in the placement area A11 (S2: Yes), the first reading unit 2A reads the item information from the first electronic tag 91 attached to the item 9. As a result, the acquisition unit 11 acquires the item information of the item 9 placed in the placement area A11 (S3). At this time, the acquisition unit 11 also acquires the read information of the first electronic tag 91 (S4).

Next, if the item information of the acquired item 9 is included in the list (S5: Yes), the labeling unit 12 assigns a correct label to the read information of the first electronic tag 91 attached to the item 9, indicating that the item 9 is the object to be read (S6A). On the other hand, if the item information of the acquired item 9 is not included in the list (S5: No), the labeling unit 12 assigns an incorrect label to the read information of the first electronic tag 91 attached to the item 9, indicating that the item 9 is not the object to be read (S6B). The latter case may occur, for example, when the item information of the first electronic tag 91 attached to the item 9 in the vicinity of the placement area A11 is read by the first reading unit 2A. Then, the labeling unit 12 stores a data set including the assigned label and the read information of the first electronic tag 91 to which the label is to be assigned in the storage unit 16 as training data (S7).

If the number of accumulated training data has not reached the predetermined number (S8: No), the learning system 100 accumulates training data by repeating the above-mentioned processes of S2 to S7. If the number of training data reaches the predetermined number (S8: Yes), the learning unit 13 uses the accumulated training data stored in the memory unit 16 to machine-learn the judgment criteria of the classifier 14 (S9). If the accuracy rate of the classification result of whether or not the article 9 is a reading target by the classifier 14 reaches a predetermined value or more, the machine learning of the judgment criteria of the classifier 14 is completed.

(3.2) Inference Phase Next, the operation of the shopping support system 200 and the learning system 100 in the inference phase will be described. As shown in FIG. 4, first, the classifier 14 reads the judgment criteria from the memory unit 16 (S11). Next, the user 90 sequentially places the items 9 to be purchased (i.e., the items to be read) in the placement area A11. Then, when the detection unit 3 (here, the second sensor 32) detects that the item 9 has been placed in the placement area A11 (S11: Yes), the acquisition unit 11 acquires the item information of the item 9 placed in the placement area A11 (S12) in the same manner as in the learning phase. At this time, the acquisition unit 11 also acquires the read information of the first electronic tag 91 in the same manner as in the learning phase (S13).

Next, one or more parameters included in the read information of the first electronic tag 91 acquired by the acquisition unit 11 are input to the classifier 14, and the classifier 14 classifies whether the item 9 is a target for reading or not (S14). The classification result of the classifier 14 is then displayed on the display unit 41 of the user interface 4 together with a message inquiring whether the classification result is correct or not, the first icon, and the second icon (S15). This allows the user 90 to evaluate the classification result of the classifier 14 by selecting the first icon or the second icon while viewing the image displayed on the display unit 41.

When the input reception unit 17 receives an evaluation of the classification result of the classifier 14 (S16), the label assignment unit 12 assigns a correct label or an incorrect label to the read information of the first electronic tag 91 based on the evaluation received by the input reception unit 17 (S17). Then, the label assignment unit 12 stores a data set including the assigned label and the read information of the first electronic tag 91 to which the label is to be assigned in the storage unit 16 as training data (S18).

Then, the learning unit 13 re-learns the judgment criteria of the classifier 14 using the accumulated training data stored in the memory unit 16 (S19). When the re-learning is completed, the learning unit 13 updates the judgment criteria by storing the re-learned judgment criteria of the classifier 14 in the memory unit 16 (S20). Note that if the re-learning by the learning unit 13 takes time, the re-learning may be performed during a time period when the shopping support system 200 is not in use, such as at night.

(4) Advantages The advantages of the learning method (learning system 100) of this embodiment will be described below, along with a comparison with the shopping support system 300 of the comparative example. As shown in Figs. 5 and 6, the shopping support system 300 of the comparative example is built into the checkout counter 7 as the main computer of the reading device 5. However, the shopping support system 300 of the comparative example differs from the shopping support system 200 of this embodiment in that it does not include the above-mentioned learning system 100. Also, in the shopping support system 300 of the comparative example, it is assumed that the classifier is configured to classify whether or not the item 9 is a target for reading, using a predetermined algorithm that compares one or more parameters included in the read information of the first electronic tag 91 with a threshold value.

In the comparative shopping support system 300, it is assumed that the classifier threshold is optimized in advance based on the store layout shown in FIG. 5. Therefore, in the comparative shopping support system 300, it is assumed that the classifier is able to classify whether or not the item 9 is a target for reading in the store layout shown in FIG. 5. The store layout shown in FIG. 5 is the same as the store layout shown in FIG. 2.

Now, assume that the layout of the store has been changed by moving display shelf B1 closer to the reading device 5, as shown in FIG. 6. In this case, in the comparative shopping support system 300, the change in the store layout may cause the classifier to be unable to classify whether or not an item 9 is a target for reading. For example, even if the classifier had classified an item 9 displayed on display shelf B1 as not a target for reading before the store layout was changed, the classifier may erroneously classify the item as a target for reading after the store layout was changed.

To enable the classifier to correctly classify whether an item 9 is a target for reading even after the layout of the store is changed, the threshold of the classifier needs to be readjusted. However, in the shopping support system 300 of the comparative example, the threshold needs to be readjusted by observing the distribution of one or more parameters contained in the read information of the first electronic tag 91 while the reading device 5 reads the item information of the first electronic tag 91 attached to a large number of items 9. Such threshold readjustment is difficult without the assistance of a system specialist engineer, and it is not realistic to summon an engineer to the store every time the layout of the store is changed.

In contrast, in the learning system 100 of this embodiment, it is possible to learn the judgment criteria of the classifier 14 using training data based on the acquired read information of the first electronic tag 91. Therefore, in this embodiment, even if the environment in which the item 9 is placed changes due to a change in the layout of the store, it is possible to classify whether the item 9 is a target for reading even in the changed environment by readjusting the judgment criteria of the classifier 14 using the training data. This readjustment of the judgment criteria of the classifier 14 is executed by inputting the training data into the classifier 14, and therefore, unlike the shopping support system 300 of the comparative example, there is no need to enlist the help of a system specialist engineer.

As described above, the present embodiment has the advantage that it is easy to improve the accuracy of classifying whether or not the item 9 is a target for reading, regardless of the environment in which the item 9 is placed. In addition, the present embodiment also has the advantage that it is possible to easily readjust the judgment criteria of the classifier 14, since it is not necessary to enlist the assistance of a system specialist engineer.

(5) Modifications The above-described embodiment is merely one of various embodiments of the present disclosure. The above-described embodiment can be modified in various ways depending on the design, etc., as long as the object of the present disclosure can be achieved. In addition, a function similar to the learning method (learning system 100) may be embodied in a (computer) program, a non-transitory recording medium on which a program is recorded, or the like. A program according to one aspect of the present disclosure causes one or more processors to execute the above-described learning method.

Below, we list some variations of the above-mentioned embodiment. The variations described below can be applied in appropriate combinations.

In the learning system 100 of the present disclosure, for example, the main computer 10 includes a computer system. The computer system is mainly composed of a processor and a memory as hardware. The processor executes a program recorded in the memory of the computer system to realize the functions of the learning system 100 of the present disclosure. The program may be pre-recorded in the memory of the computer system, may be provided through an electric communication line, or may be recorded and provided on a non-transitory recording medium such as a memory card, an optical disk, or a hard disk drive that can be read by the computer system. The processor of the computer system is composed of one or more electronic circuits including a semiconductor integrated circuit (IC) or a large-scale integrated circuit (LSI). The integrated circuits such as IC or LSI referred to here are called different names depending on the degree of integration, and include integrated circuits called system LSI, VLSI (Very Large Scale Integration), or ULSI (Ultra Large Scale Integration). Furthermore, a field-programmable gate array (FPGA) that is programmed after the LSI is manufactured, or a logic device that allows reconfiguration of the connection relationship within the LSI or reconfiguration of the circuit partition within the LSI, can also be used as a processor. The electronic circuits may be integrated into one chip, or may be distributed among multiple chips. The chips may be integrated into one device, or may be distributed among multiple devices. The computer system referred to here includes a microcontroller having one or more processors and one or more memories. Therefore, the microcontroller is also composed of one or more electronic circuits including a semiconductor integrated circuit or a large-scale integrated circuit.

Furthermore, it is not essential for the learning system 100 that multiple functions are concentrated in one housing. The components of the learning system 100 may be distributed across multiple housings. Furthermore, at least some of the functions of the learning system 100 may be realized, for example, by a server device and the cloud (cloud computing), etc.

As an example, in the above embodiment, the learning system 100 is realized by a main computer 10 in the store, but this is not limited thereto, and the learning system 100 may be realized by a server device in a remote location away from the store.

In the above embodiment, the input receiving unit 17 may receive an input when an index regarding the reliability of the classification result by the classifier 14 is within a predetermined range. In other words, the input receiving step may be executed when an index regarding the reliability of the classification result by the classifier 14 is within a predetermined range. The "index regarding the reliability of the classification result" here is, for example, the classification probability of whether or not the item 9 is a reading target by the classifier 14. For example, when the classification probability is less than a threshold value (for example, less than 90%), the input receiving unit 17 receives an input. In other words, when the classification probability is less than a threshold value, there is a possibility that the item 9 has not been accurately classified as a reading target or not. Therefore, in such a case, it is possible to complement the classification accuracy of the classifier 14 by receiving an evaluation of the classification result by the classifier 14 by the user 90. In other words, when the classification probability is equal to or greater than the threshold value, the input receiving unit 17 does not need to receive an input of an evaluation of the classification result by the classifier 14.

In the above-described embodiment, the detection unit 3 may have a function of acquiring status information representing the status of the item 9, such as the volume or weight of the item 9. In other words, the learning method may further have a status acquisition step of acquiring the status information. The explanatory variables input to the classifier 14 may include the status information. In this aspect, the judgment criteria of the classifier 14 are learned to classify whether or not the item 9 is a reading target taking into account the status of the item 9, which has the advantage that the classification accuracy of the classifier 14 can be expected to be improved.

In the above-described embodiment, for example, as shown in FIG. 7, a second electronic tag 81 may be attached to a mobile object (here, a shopping basket) 8. The second electronic tag 81 is a passive RF tag like the first electronic tag 91, and has a memory that stores mobile object information including at least information for identifying the mobile object 8, which includes a unique identification code assigned to each mobile object 8. Like the first electronic tag 91, the second electronic tag 81 may be attached to the mobile object 8 in a state that allows it to be handled together with the mobile object 8.

The acquisition unit 11 may acquire not only the read information of the first electronic tag 91, but also the read information of the second electronic tag 81 obtained by wireless communication between the first reading unit 2A and the second electronic tag 81. In other words, the learning method may further include a third acquisition step of acquiring the read information of the second electronic tag 81 by the first reading unit 2A performing wireless communication using radio waves as a medium with the second electronic tag 81 attached to the moving object 8 that moves together with the article 9. The labeling step may determine a label based on a comparison result between the read information of the first electronic tag 91 and the read information of the second electronic tag 81.

For example, an item 9 to be purchased (i.e., the item to be read) is placed in a mobile object (shopping basket) 8 on the placement area A11 of the cash register counter 7. Therefore, the read information of the first electronic tag 91 attached to this item 9 and the read information of the second electronic tag 81 attached to the mobile object 8 in which this item 9 is stored follow almost the same changes over time.

Therefore, when the read information of the first electronic tag 91 and the read information of the second electronic tag 81 follow almost the same changes over time, the label attachment unit 12 attaches a correct label to the read information of the first electronic tag 91, indicating that the item 9 is the object to be read. On the other hand, when the read information of the first electronic tag 91 and the read information of the second electronic tag 81 differ from each other in changes over time, the label attachment unit 12 attaches an incorrect label to the read information of the first electronic tag 91, indicating that the item 9 is not the object to be read. This aspect has the advantage that it is easier to attach an appropriate label to the read information of the first electronic tag 91 compared to the case where the read information of the second electronic tag 81 is not referenced. In addition, this aspect has the advantage that it is possible to automatically attach a label to the read information of the first electronic tag 91 without the need for the user 90 to make a decision.

In the above embodiment, multiple second electronic tags 81 may be attached to the mobile object 8. The multiple second electronic tags 81 may be arranged to face in different directions. This embodiment has the advantage that it is easier to assign a more appropriate label to the read information of the first electronic tag 91, since the read information of the multiple second electronic tags 81 that change over time is referenced. This embodiment also has the advantage that the user 90 can place the mobile object (shopping basket) 8 on the placement area A11 of the checkout counter 7 without considering the orientation of the mobile object.

In the inference phase of the above-described embodiment, the learning unit 13 may not use all of the training data obtained each time the classifier 14 outputs a classification result to re-learn the judgment criteria of the classifier 14. For example, the learning unit 13 may exclude relatively unreliable data from the training data based on the reliability of the training data. In other words, the learning step may include an exclusion step of excluding data whose index relating to the reliability of the training data is below a threshold value from the training data.

The "reliability index" here is determined, for example, based on the input time from when the classification result of the classifier 14 is displayed on the display unit 41 until the evaluation is input. For example, the shorter the input time, the smaller the reliability index will be, since it is considered that the user 90 is inputting without thinking. In other words, the index is determined based on the input in the input reception step.

Also, as an example, the index may increase or decrease depending on whether the user 90 in the input reception step is a store clerk or a customer. For example, if the user 90 is a store clerk, the index may be higher than a threshold, and if the user 90 is a customer, the index may be lower than the threshold. In this case, the user 90 may be identified as a store clerk if, for example, the store clerk ID card is authenticated before the evaluation is input in the input reception unit 17, and as a customer otherwise.

Alternatively, the index may be determined based on the period for which the training data has been stored in the memory unit 16. As an example, the exclusion step may exclude, from the training data stored in the memory unit 16, training data that has been stored in the memory unit 16 for a period longer than a predetermined period.

In the above embodiment, the learning unit 13 may re-learn the judgment criteria of the classifier 14 when the layout of the store is changed. In other words, the learning step may be executed when the arrangement of the items 9 that are not to be read is changed. As an example, the re-learning of the judgment criteria of the classifier 14 is executed when the user 90 (here, a store clerk) performs a predetermined operation on the operation unit 42 of the user interface 4 after the layout of the store is changed. This aspect has the advantage that it is easy to optimize the judgment criteria of the classifier 14 according to the environment in which the items 9 are placed.

In the above embodiment, in the learning phase, the classifier 14 learns its judgment criteria using training data on the object 9 to be read, but this is not limited to the above. For example, in the learning phase, the classifier 14 may learn its judgment criteria by also using training data on the object 9 that is not the object to be read. This aspect has the advantage that it is easier to have the classifier 14 learn the boundary between whether or not the object 9 is the object to be read, compared to when the classifier 14 is trained using only training data on the object 9 to be read.

(Embodiment 2)
The learning system 100A of this embodiment differs from the learning system 100 of the first embodiment in that it has two reading units (a first reading unit 2A and a second reading unit 2B) as shown in Fig. 8. That is, the register counter 7 has a second reading unit 2B different from the first reading unit 2A in addition to the first reading unit 2A. The second reading unit 2B has an antenna 23 and a communication unit 24. The configurations of the antenna 23 and the communication unit 24 are the same as the antenna 21 and the communication unit 22 of the first reading unit 2A, respectively.

The antenna 21 (first reading unit 2A) is installed facing the first area A1, as in the first embodiment. The first area A1 includes a placement area A11 where the item 9 to be read is placed. On the other hand, the antenna 23 (second reading unit 2B) is installed facing a second area A2 different from the first area A1, as shown in Figures 9 and 10. The second area A2 includes the surrounding area A21. In other words, the second area A2 is an area that does not include the placement area A11. The second area A2 is, for example, an area around the register counter 7 where users 90 are lined up.

In this embodiment, the acquisition unit 11 acquires the article information from the first electronic tag 91 by performing wireless communication using radio waves with the first electronic tag 91 at the second reading unit 2B. In other words, in this embodiment, the learning method further includes a second acquisition step of acquiring the article information from the first electronic tag 91 by performing wireless communication using radio waves with the first electronic tag 91 at the second reading unit 2B. In this embodiment, the labeling step determines a label based on a comparison result between the read information of the first electronic tag 91 obtained at the first reading unit 2A (i.e., obtained by the first acquisition step) and the read information of the first electronic tag 91 obtained at the second reading unit 2B (i.e., obtained by the second acquisition step).

In particular, in this embodiment, the labeling step determines the label based on whether the magnitude relationship between the first radio wave intensity and the second radio wave intensity is inverted. The first radio wave intensity is the intensity of the radio wave received by the first reading unit 2A from the first electronic tag 91. The second radio wave intensity is the intensity of the radio wave received by the second reading unit 2B from the first electronic tag 91.

An example of the operation of the learning system 100A of this embodiment will be described below with reference to Figures 9 to 13. In Figures 11 to 13, "RSSI" on the vertical axis represents the RSSI value of the first radio wave intensity (or the second radio wave intensity). Also, "difference" on the vertical axis represents the value obtained by subtracting the second radio wave intensity from the first radio wave intensity. Therefore, if the first radio wave intensity is greater than the second radio wave intensity, the difference is a positive value, and if the first radio wave intensity is less than the second radio wave intensity, the difference is a negative value. In Figures 11 to 13, "t1" on the horizontal axis represents the time when the first sensor 31 detects an object (here, the item 9, the shopping basket 8, or the user 90) in the surrounding area A21. Also, "t2" on the horizontal axis represents the time when the second sensor 32 detects that the item 9 (or the shopping basket 8 containing the item 9) has been placed in the placement area A11. Furthermore, in Figures 11 to 13, circles represent the first radio wave intensity, and crosses represent the second radio wave intensity.

Now, assume that a user 90 carrying a shopping basket 8 containing an item 9 to be purchased (i.e., an item to be read) enters the surrounding area A21 from outside the surrounding area A21 shown in FIG. 9 and places the shopping basket 8 in the placement area A11 as shown in FIG. 10. In this case, as shown in FIG. 11, in the first reading unit 2A, the first radio wave intensity increases over time because the item 9 approaches the placement area A11, i.e., approaches the first area A1, from time t1 to time t2. Then, after time t2, the item 9 remains in the placement area A11, so the first radio wave intensity becomes almost constant. On the other hand, in the second reading unit 2B, the item 9 moves from the surrounding area A21 to the placement area A11 from time t1 to time t2, i.e., the item 9 moves away from the second area A2, so the second radio wave intensity decreases over time. Then, after time t2, the item 9 remains in the placement area A11, so the second radio wave intensity becomes approximately a constant value.

The difference between the first radio wave intensity and the second radio wave intensity is a negative value at time t1, but increases over time to become a positive value between time t1 and time t2. In other words, for the first electronic tag 91 attached to the article 9 to be read, the magnitude relationship between the first radio wave intensity and the second radio wave intensity is reversed.

On the other hand, for a first electronic tag 91 attached to an item 9 that is not a target for purchase (i.e., not a target for reading), since the item 9 does not move, the first radio wave intensity and the second radio wave intensity are both approximately constant values, as shown in FIG. 12. Therefore, the difference between the first radio wave intensity and the second radio wave intensity is also approximately constant. In other words, for a first electronic tag 91 attached to an item 9 that is not a target for reading, the magnitude relationship between the first radio wave intensity and the second radio wave intensity is not reversed.

If the first electronic tag 91 attached to the item 9 to be read overlaps with a first electronic tag 91 attached to another item 9, it is expected that the strength of the radio waves received from the first electronic tag 91 will be reduced in both the first reading unit 2A and the second reading unit 2B. Even in such a case, as shown in FIG. 13, for the first electronic tag 91 attached to the item 9 to be read, the difference between the first radio wave intensity and the second radio wave intensity will change over time, and the magnitude relationship will be reversed.

When the magnitude relationship between the first radio wave intensity and the second radio wave intensity is reversed, the label assignment unit 12 assigns a correct label to the read information of the first electronic tag 91, indicating that the item 9 is the object to be read. On the other hand, when the magnitude relationship between the first radio wave intensity and the second radio wave intensity is not reversed, the label assignment unit 12 assigns an incorrect label to the read information of the first electronic tag 91, indicating that the item 9 is not the object to be read.

Advantages of the learning system 100A of this embodiment will be described below with a comparison with the learning system 100 of embodiment 1. In the learning system 100 of embodiment 1, since the reading unit is the first reading unit 2A, the acquisition unit 11 acquires only the first radio wave intensity. Here, the temporal change in the first radio wave intensity in the learning system 100 of embodiment 1 is shown in FIG. 14 and FIG. 15. In both FIG. 14 and FIG. 15, the black circle represents the first radio wave intensity of the first electronic tag 91 attached to the item 9 to be read (hereinafter referred to as the "radio wave intensity of the object to be read"). Also, the white circle represents the first radio wave intensity of the first electronic tag 91 attached to the item 9 that is not the object to be read (hereinafter referred to as the "radio wave intensity of the non-read object"). Also, FIG. 14 shows a case where the first electronic tag 91 attached to the item 9 to be read and the first electronic tag 91 attached to the other item 9 do not overlap. On the other hand, FIG. 15 shows a case where a first electronic tag 91 attached to the item 9 to be read overlaps with a first electronic tag 91 attached to another item 9.

In the example shown in FIG. 14, there is a significant difference between the radio wave strength of the object to be read and the radio wave strength of the object not to be read after time t2, so by setting an appropriate threshold, it is possible for the labeling unit 12 to assign an appropriate label. However, in the example shown in FIG. 15, there is almost no difference between the radio wave strength of the object to be read and the radio wave strength of the object not to be read even after time t2, making it difficult for the labeling unit 12 to assign an appropriate label.

In contrast, in this embodiment, the labeling unit 12 can assign an appropriate label regardless of whether multiple first electronic tags 91 overlap. In other words, this embodiment has the advantage that it is easier to assign an appropriate label to the read information of the first electronic tag 91 compared to when there is only one reading unit. In addition, this embodiment has the advantage that it is possible to automatically assign a label to the read information of the first electronic tag 91 without the need for judgment by the user 90.

The configuration described in embodiment 2 can be adopted in appropriate combination with the various configurations (including modified examples) described in embodiment 1.

(summary)
As described above, the learning method according to the first aspect includes a first acquisition step, a labeling step, and a learning step. The first acquisition step is a step of acquiring article information related to the article (9) from the first electronic tag (91) by performing wireless communication using radio waves as a medium with the first reading unit (2A) and the first electronic tag (91) attached to the article (9). The labeling step is a step of assigning a label indicating whether the article (9) is a read target or not to the read information of the first electronic tag (91) obtained by acquiring the article information. The learning step is a step of calculating a judgment criterion for the classifier (14) by machine learning using training data. The classifier (14) classifies whether the article (9) is a read target or not based on a comparison between the read information of the first electronic tag (91) and the judgment criterion. The training data is data including one or more parameters included in the read information of the first electronic tag (91) acquired in the first acquisition step and the label assigned in the labeling step.

This aspect has the advantage that it is easier to improve the accuracy of classifying whether or not an item (9) is a target for reading, regardless of the environment in which the item (9) is placed.

The learning method according to the second aspect further includes an input receiving step of receiving an input for evaluating the classification result of the classifier (14) in the first aspect. The labeling step determines a label based on the input received in the input receiving step.

This embodiment has the advantage that by referring to the evaluation of the classification results of the classifier (14), it is easier to assign an appropriate label to the read information of the first electronic tag (91).

In the learning method according to the third aspect, in the second aspect, the input receiving step is executed when an index relating to the certainty of the classification result by the classifier (14) is within a predetermined range.

This aspect has the advantage that it is possible to complement the classification accuracy of the classifier (14) when the classifier (14) is insufficient in classifying whether the item (9) is a target for reading or not.

The learning method according to the fourth aspect is the second or third aspect, and further includes a second acquisition step. The second acquisition step is a step of acquiring item information from the first electronic tag (91) by performing wireless communication using radio waves as a medium with the first electronic tag (91) at a second reading unit (2B) different from the first reading unit (2A). The labeling step determines a label based on the result of comparing the read information of the first electronic tag (91) obtained by the first acquisition step with the read information of the first electronic tag (91) obtained by the second acquisition step.

This embodiment has the advantage that it is easier to assign an appropriate label to the read information of the first electronic tag (91) compared to when there is only one reading unit.

In the learning method according to the fifth aspect, in the fourth aspect, the first reading unit (2A) is installed facing the first area (A1). The second reading unit (2B) is installed facing the second area (A2) different from the first area (A1).

This aspect has the advantage that it is easier to differentiate between the results of reading information obtained by the first reading unit (2A) and the second reading unit (2B).

In the learning method according to the sixth aspect, in the fifth aspect, the first area (A1) includes a placement area (A11) on which the item (9) to be read is placed.

This aspect has the advantage that it is easier to differentiate between the results of reading information obtained by the first reading unit (2A) and the second reading unit (2B).

In the learning method according to the seventh aspect, in the fifth or sixth aspect, the labeling step determines the label based on whether or not there is a reversal of the magnitude relationship between the intensity of the radio wave received by the first reading unit (2A) from the first electronic tag (91) and the intensity of the radio wave received by the second reading unit (2B) from the first electronic tag (91).

This aspect has the advantage that it is possible to assign an appropriate label to the read information of the first electronic tag (91), regardless of whether or not multiple first electronic tags (91) overlap.

The learning method according to the eighth aspect is any one of the first to seventh aspects, and further includes a detection step. The detection step is a step of detecting the presence or absence of at least one of an item (9) in the detection range and a moving object (8) moving together with the item (9). The execution period of the labeling step is determined based on the detection result of the detection step.

This aspect has the advantage that it is easier to perform the labeling step while the item (9) to be read is present.

In the learning method according to the ninth aspect, in the eighth aspect, the detection range includes a placement area (A11) and a surrounding area (A21). The placement area (A11) is an area where the item (9) to be read is placed. The surrounding area (A21) is an area up to a position a predetermined distance away from the placement area (A11).

This aspect has the advantage that it is easier to perform the labeling step while the item (9) to be read is present.

The learning method according to the tenth aspect, in any one of the first to ninth aspects, further includes a state acquisition step of acquiring state information representing the state of the item (9). The explanatory variables input to the classifier (14) include the state information.

According to this embodiment, the judgment criteria of the classifier (14) are learned to classify whether or not the item (9) is a reading target, taking into account the condition of the item (9), which has the advantage that the classification accuracy of the classifier (14) can be expected to improve.

The learning method according to the eleventh aspect is any one of the first to tenth aspects, and further includes a third acquisition step. The third acquisition step is a step in which the first reading unit (2A) acquires read information of the second electronic tag (81) attached to the moving body (8) that moves together with the article (9) by performing wireless communication using radio waves as a medium with the second electronic tag (81). The labeling step determines a label based on the result of comparing the read information of the first electronic tag (91) with the read information of the second electronic tag (81).

This embodiment has the advantage that it is easier to assign an appropriate label to the read information of the first electronic tag (91) compared to a case where the read information of the second electronic tag (81) is not referenced.

In the learning method according to the twelfth aspect, in the eleventh aspect, a plurality of second electronic tags (81) are attached to the moving body (8). The plurality of second electronic tags (81) are arranged so that they face in different directions.

This embodiment has the advantage that it is easier to assign a more appropriate label to the read information of the first electronic tag (91) because it refers to the read information of multiple second electronic tags (81) that change over time differently.

In the learning method according to the thirteenth aspect, in any one of the first to twelfth aspects, the learning step includes an exclusion step. The exclusion step is a step of excluding data whose index relating to the reliability of the training data is below a threshold value from the training data.

This embodiment has the advantage that it is easier to improve the classification accuracy of the classifier (14) by excluding training data with relatively low reliability.

The learning method according to the 14th aspect is the 13th aspect, and further includes an input receiving step of receiving an input for evaluating the classification result of the classifier (14). The index is determined based on the input in the input receiving step.

This has the advantage that it is easier to evaluate the reliability of the training data.

In the learning method according to the fifteenth aspect, in any of the first to fourteenth aspects, the explanatory variables input to the classifier (14) do not include information that identifies the item (9).

This aspect has the advantage that by eliminating information that identifies the item (9), it is possible to prevent the classification results of the classifier (14) from being influenced by information that identifies the item (9).

In the learning method according to the sixteenth aspect, in any of the first to fifteenth aspects, the learning step is performed when the arrangement of the item (9) that is not to be read is changed.

This aspect has the advantage that it is easier to optimize the judgment criteria of the classifier (14) depending on the environment in which the item (9) is placed.

The program according to the seventeenth aspect causes one or more processors to execute the learning method according to any one of the first to sixteenth aspects.

This aspect has the advantage that it is easier to improve the accuracy of classifying whether or not an item (9) is a target for reading, regardless of the environment in which the item (9) is placed.

The learning system (100) according to the eighteenth aspect includes an acquisition unit (11), a labeling unit (12), and a learning unit (13). The acquisition unit (11) acquires item information on the item (9) from the first electronic tag (91) by performing wireless communication using radio waves as a medium with the first electronic tag (91) attached to the item (9) by the first reading unit (2A). The labeling unit (12) assigns a label indicating whether the item (9) is a target to be read to the read information of the first electronic tag (91) obtained by acquiring the item information. The learning unit (13) uses the training data to calculate the judgment criterion of the classifier (14) by machine learning. The classifier (14) classifies whether the item (9) is a target to be read based on a comparison between the read information of the first electronic tag (91) and the judgment criterion. The training data includes one or more parameters included in the read information of the first electronic tag (91) acquired by the acquisition unit (11) and a label assigned by the label assignment unit (12).

This aspect has the advantage that it is easier to improve the accuracy of classifying whether or not an item (9) is a target for reading, regardless of the environment in which the item (9) is placed.

The methods according to the second to sixteenth aspects are not essential to the learning method and may be omitted as appropriate.

By the way, by using the acquisition result of the third acquisition step and the acquisition result of the first acquisition step in the learning method according to the eleventh aspect, it is also possible to estimate whether the item 9 is a reading target without using the classifier 14. That is, the estimation method according to the nineteenth aspect has a first acquisition step, a third acquisition step, and an estimation step. The first acquisition step is a step of acquiring item information related to the item (9) from the first electronic tag (91) by performing wireless communication using radio waves with the first electronic tag (91) attached to the item (9) in the first reading unit (2A). The third acquisition step is a step of acquiring read information of the second electronic tag (81) by performing wireless communication using radio waves with the second electronic tag (81) attached to the moving body (8) moving together with the item (9) in the first reading unit (2A). The estimation step is a step of estimating whether or not the item (9) is a target for reading based on the results of a comparison between the read information of the first electronic tag (91) obtained in the first acquisition step and the read information of the second electronic tag (81) obtained in the third acquisition step.

In addition, by using the acquisition result of the second acquisition step and the acquisition result of the first acquisition step in the learning method according to the fourth aspect, it is also possible to estimate whether the item 9 is a target for reading without using the classifier 14. That is, the estimation method according to the twentieth aspect has a first acquisition step, a second acquisition step, and an estimation step. The first acquisition step is a step of acquiring item information related to the item (9) from the first electronic tag (91) by performing wireless communication using radio waves with the first electronic tag (91) attached to the item (9) in the first reading unit (2A). The second acquisition step is a step of acquiring item information from the first electronic tag (91) by performing wireless communication using radio waves with the first electronic tag (91) in the second reading unit (2B) different from the first reading unit (2A). The estimation step is a step of estimating whether or not the item (9) is a target for reading based on the results of a comparison between the read information of the first electronic tag (91) obtained in the first acquisition step and the read information of the first electronic tag (91) obtained in the second acquisition step.

100 Learning system 11 Acquisition unit 13 Learning unit 14 Classifier 2A First reading unit 2B Second reading unit 8 Shopping basket (mobile object)
81 Second electronic tag 9 Article 91 First electronic tag A1 First area A11 Placement area A2 Second area A21 Surrounding area

Claims (19)

a first acquisition step of acquiring article information related to the article from a first electronic tag of the article by performing wireless communication using radio waves as a medium between a first reader and a first electronic tag attached to the article;
a second acquisition step of acquiring the article information of the article from the first electronic tag of the article by performing wireless communication using radio waves with the first electronic tag of the article by a second reading unit different from the first reading unit;
a labeling step of adding a label indicating whether or not the article is a target for reading to the read information of the first electronic tag of the article based on a comparison result between the read information of the first electronic tag of the article obtained by acquiring the article information of the article in the first acquisition step and the read information of the first electronic tag of the article obtained by acquiring the article information of the article in the second acquisition step ;
a learning step of calculating, by machine learning, a judgment criterion of a classifier that classifies whether or not the article to be read is the article to be read, using training data including one or more parameters included in the read information of the first electronic tag of the article acquired in the first acquisition step, one or more parameters included in the read information of the first electronic tag of the article acquired in the second acquisition step, and the label assigned in the label assignment step;
A first step of acquiring article information related to the object to be read from the first electronic tag of the object to be read by performing wireless communication using radio waves as a medium between the first reading unit and the first electronic tag of the object to be read;
A second step of acquiring the article information of the article to be read from the first electronic tag of the article to be read by performing wireless communication using radio waves as a medium between the second reading unit and the first electronic tag of the article to be read;
an input receiving step of receiving an input for evaluating a classification result of the classifier after the classifier classifies whether the object to be read is the object to be read or not based on a comparison between the read information of the first electronic tag of the object to be read, which is obtained by acquiring the object information of the object to be read in the first step, and the read information of the first electronic tag of the object to be read, which is obtained by acquiring the object information of the object to be read in the second step, and the judgment criterion;
and a third step of adding a label indicating whether the article to be read is the article to be read or not to the read information of the first electronic tag of the article to be read based on the input received in the input receiving step,
Executed by one or more processors,
How to learn. the first reading unit is installed facing a first area,
The second reading unit is installed facing a second area different from the first area.
The learning method according to claim 1 . The first area includes a placement area on which the article to be read is placed.
The learning method according to claim 2 . In the labeling step, the label is determined based on whether or not there is a reversal of the magnitude relationship between the intensity of the radio wave received by the first reading unit from the first electronic tag of the article and the intensity of the radio wave received by the second reading unit from the first electronic tag of the article.
The learning method according to claim 2 or 3. a first acquisition step of acquiring article information related to the article from a first electronic tag of the article by performing wireless communication using radio waves as a medium between a first reader and a first electronic tag attached to the article;
a labeling step of attaching a label indicating whether the article is a target for reading to the read information of the first electronic tag of the article obtained by acquiring the article information of the article in the first acquisition step;
a learning step of calculating, by machine learning, a judgment criterion of a classifier that classifies whether or not the article to be read is the article to be read, using training data including one or more parameters included in the read information of the first electronic tag of the article acquired in the first acquisition step and the label assigned in the label assignment step;
A first step of acquiring article information related to the object to be read from the first electronic tag of the object to be read by performing wireless communication using radio waves as a medium between the first reading unit and the first electronic tag of the object to be read;
an input receiving step of receiving an input for evaluating a classification result of the classifier after the classifier classifies whether the object to be read is the object to be read or not based on a comparison between the read information of the first electronic tag of the object to be read, which is obtained by acquiring the object information of the object to be read in the first step, and the judgment criterion;
a second step of adding a label indicating whether the article to be read is the article to be read or not to the read information of the first electronic tag of the article to be read based on the input received in the input receiving step;
A status acquisition step of acquiring status information representing a status of the item,
The state of the item includes at least the volume or weight of the item,
The explanatory variables input to the classifier include the state information,
In the learning step, the judgment criterion of the classifier is learned so as to classify whether the article is the reading target or not based on the state information;
Executed by one or more processors,
How to learn. a first acquisition step of acquiring article information related to the article from a first electronic tag of the article by performing wireless communication using radio waves as a medium between a first reader and a first electronic tag attached to the article;
a second acquisition step of acquiring read information of the second electronic tag of the moving object by performing wireless communication using radio waves with the second electronic tag attached to the moving object that moves together with the article, using the first reading unit;
a labeling step of attaching a label indicating whether the article is a target for reading based on a comparison result between the read information of the first electronic tag of the article obtained by acquiring the article information of the article in the first acquisition step and the read information of the second electronic tag of the moving object acquired in the second acquisition step;
a learning step of calculating, by machine learning, a judgment criterion of a classifier that classifies whether or not the article to be read is the article to be read, using training data including one or more parameters included in the read information of the first electronic tag of the article acquired in the first acquisition step, one or more parameters included in the read information of the second electronic tag of the moving object acquired in the second acquisition step, and the label assigned in the label assignment step;
A first step of acquiring article information related to the object to be read from the first electronic tag of the object to be read by performing wireless communication using radio waves as a medium between the first reading unit and the first electronic tag of the object to be read;
a second step of acquiring read information of the second electronic tag of the second moving object by performing wireless communication using radio waves with the second electronic tag attached to the second moving object that moves together with the object to be read, in the first reading unit;
an input receiving step of receiving an input for evaluating a classification result of the classifier after the classifier classifies whether the object to be read is the object to be read based on a comparison between the read information of the first electronic tag of the object to be read and the read information of the second electronic tag of the second moving body and the judgment criterion;
and a third step of adding a label indicating whether the article to be read is the article to be read or not to the read information of the first electronic tag of the article to be read based on the input received in the input receiving step,
Executed by one or more processors,
How to learn. A plurality of the second electronic tags are attached to the moving object,
The second electronic tags are arranged so as to face in different directions.
The learning method according to claim 6. adding a data set including one or more parameters included in the read information of the first electronic tag of the article to be read, which is obtained by acquiring the article information of the article to be read, and the label assigned to the read information of the first electronic tag of the article to be read, to the training data and storing the data set in a storage unit;
re-learning the judgment criteria of the classifier using the training data stored in the storage unit;
and updating the criteria when the re-learning is completed.
A learning method according to any one of claims 1 to 7. The input receiving step is executed when an index relating to the certainty of the classification result by the classifier is within a predetermined range.
A learning method according to any one of claims 1 to 8. The method further includes a detection step of detecting the presence or absence of at least one of the object and a moving object moving together with the object in a detection range,
an execution period of the labeling step is determined based on a detection result of the detection step;
The learning method according to any one of claims 1 to 9. The detection range is:
a placement area on which the object to be read is placed;
A surrounding area up to a position a predetermined distance away from the placement area,
The learning method according to claim 10. The learning step includes an exclusion step of excluding data whose index regarding the reliability of the training data is below a threshold from the training data.
A learning method according to any one of claims 1 to 11. The indicator includes information indicative of a user or a customer;
The index is determined based on an input in the input receiving step,
In the excluding step, predetermined data is excluded from the training data based on the index.
The learning method according to claim 12. The explanatory variables input to the classifier do not include information identifying the article.
A learning method according to any one of claims 1 to 13. the learning step is executed when a user operates a user interface in a case where a position of the article that is not a reading target is changed.
A learning method according to any one of claims 1 to 14. One or more processors,
Executing the learning method according to any one of claims 1 to 15,
Program. a first acquisition unit that acquires article information related to the article from a first electronic tag of the article by performing wireless communication using radio waves as a medium between the first reading unit and a first electronic tag attached to the article;
a second acquisition unit that acquires the article information of the article from the first electronic tag of the article by performing wireless communication using radio waves as a medium with the first electronic tag of the article using a second reading unit different from the first reading unit;
a label attachment unit that attaches a label indicating whether or not the article is a target for reading to the read information of the first electronic tag of the article based on a comparison result between the read information of the first electronic tag of the article obtained by acquiring the article information of the article by the first acquisition unit and the read information of the first electronic tag of the article obtained by acquiring the article information of the article by the second acquisition unit;
a learning unit that calculates, by machine learning, a judgment criterion for a classifier that classifies whether or not an article to be read is the article to be read, using training data including one or more parameters included in the read information of the first electronic tag of the article acquired by the first acquisition unit, one or more parameters included in the read information of the first electronic tag of the article acquired by the second acquisition unit, and the label assigned by the label assignment unit; and
An input receiving unit,
The first acquisition unit acquires article information related to the read target article from the first electronic tag of the read target article by performing wireless communication using radio waves as a medium with the first electronic tag of the read target article by the first reading unit,
The second acquisition unit acquires the article information of the object to be read from the first electronic tag of the object to be read by performing wireless communication using radio waves as a medium with the first electronic tag of the object to be read by the second reading unit,
the input receiving unit receives an input for evaluating a classification result of the classifier after the classifier classifies whether the object to be read is the object to be read or not based on a comparison between the read information of the first electronic tag of the object to be read, which is obtained by acquiring the object information of the object to be read by the first acquiring unit, and the read information of the first electronic tag of the object to be read, which is obtained by acquiring the object information of the object to be read by the second acquiring unit, and the judgment criterion;
The labeling unit assigns a label indicating whether the article to be read is the article to be read, to the read information of the first electronic tag of the article to be read, based on the input accepted by the input accepting unit.
Learning system. a first acquisition unit that acquires article information related to the article from a first electronic tag of the article by performing wireless communication using radio waves as a medium between the first reading unit and a first electronic tag attached to the article;
a label attachment unit that attaches a label indicating whether or not the article is a target for reading to read information of the first electronic tag of the article obtained by acquiring the article information of the article by the first acquisition unit;
a learning unit that uses training data including one or more parameters included in the read information of the first electronic tag of the article acquired by the first acquisition unit and the label assigned by the label assignment unit to calculate, by machine learning, a judgment criterion for a classifier that classifies whether or not the article to be read is the article to be read;
An input receiving unit;
A status acquisition unit,
The first acquisition unit acquires article information related to the read target article from the first electronic tag of the read target article by performing wireless communication using radio waves as a medium with the first electronic tag of the read target article by the first reading unit,
the input receiving unit receives an input for evaluating a classification result of the classifier after the classifier classifies whether the object to be read is the object to be read or not based on a comparison between the read information of the first electronic tag of the object to be read, which is obtained by acquiring the object information of the object to be read by the first acquisition unit, and the judgment criterion;
the labeling unit assigns a label indicating whether the article to be read is the article to be read, to the read information of the first electronic tag of the article to be read, based on the input accepted by the input accepting unit;
The status acquisition unit acquires status information representing a status of the item,
The state of the item includes at least the volume or weight of the item,
The explanatory variables input to the classifier include the state information,
The learning unit learns the determination criterion of the classifier so as to classify whether the article is the reading target or not based on the status information.
Learning system. a first acquisition unit that acquires article information related to the article from a first electronic tag of the article by performing wireless communication using radio waves as a medium between the first reading unit and a first electronic tag attached to the article;
a second acquisition unit that acquires read information of a second electronic tag of a moving object that moves together with the article by performing wireless communication using radio waves with the second electronic tag attached to the moving object by the first reading unit;
a label attachment unit that attaches a label indicating whether the article is a target for reading based on a comparison result between read information of the first electronic tag of the article obtained by acquiring the article information of the article by the first acquisition unit and read information of the second electronic tag of the moving object acquired by the second acquisition unit;
a learning unit that calculates, by machine learning, a judgment criterion for a classifier that classifies whether or not an article to be read is the article to be read, using training data including one or more parameters included in the read information of the first electronic tag of the article acquired by the first acquisition unit, one or more parameters included in the read information of the second electronic tag of the moving object acquired by the second acquisition unit, and the label assigned by the label assignment unit; and
An input receiving unit,
The first acquisition unit acquires article information related to the read target article from the first electronic tag of the read target article by performing wireless communication using radio waves as a medium with the first electronic tag of the read target article by the first reading unit,
The second acquisition unit acquires read information of the second electronic tag of the second moving body by performing wireless communication using radio waves as a medium with a second electronic tag attached to a second moving body that moves together with the object to be read, using the first reading unit;
the input receiving unit receives the input for evaluating a classification result of the classifier after the classifier classifies whether the object to be read is the object to be read based on a comparison between the read information of the first electronic tag of the object to be read and the read information of the second electronic tag of the second moving body and the judgment criterion;
The labeling unit assigns a label indicating whether the article to be read is the article to be read, to the read information of the first electronic tag of the article to be read, based on the input accepted by the input accepting unit.
Learning system.

Images