WO2021075345A1

WO2021075345A1 - Signal processing method, learning model generation method, signal processing device, radiation detection device, and computer program

Info

Publication number: WO2021075345A1
Application number: PCT/JP2020/038107
Authority: WO
Inventors: 駿介村田
Original assignee: 株式会社堀場製作所
Priority date: 2019-10-18
Filing date: 2020-10-08
Publication date: 2021-04-22
Also published as: JPWO2021075345A1

Abstract

Provided are a signal processing method, a learning model generation method, a signal processing device, a radiation detection device, and a computer program which are capable of improving the accuracy of elemental analysis.　In a signal processing method for counting, by wave height, a staircase wave corresponding to radiation detection or a pulse wave obtained by converting the staircase wave, when at least one among a string of signal values constituting a signal including the staircase wave and a string of signal values constituting a signal including the pulse wave are input, at least one among the string of signal values constituting a signal including the staircase wave and the string of signal values constituting a signal including the pulse wave is input to a learning model that outputs information regarding whether a plurality of the pulse waves overlap or not, and the staircase wave or the pulse wave when the plurality of pulse waves do not overlap is counted in accordance with the information output by the learning model.

Description

Signal processing method, learning model generation method, signal processing device, radiation detection device and computer program

The present invention relates to a signal processing method for processing a signal generated by detection of radiation, a learning model generation method, a signal processing device, a radiation detection device, and a computer program.

The radiation detection device that detects radiation such as X-rays includes a radiation detector and a signal processing device that processes a signal output by the radiation detector. The radiation detector is configured by using a semiconductor radiation detection element or the like, and outputs a staircase wave each time radiation is detected. The signal processing device converts the staircase wave into a pulse wave and measures the height of the pulse wave. The height of the pulse wave corresponds to the energy of the radiation.

If the interval at which radiation is detected is short, multiple pulse waves may overlap, causing so-called pile-up in which the height of the pulse waves changes. At this time, the energy of erroneous radiation is measured, and a peak having erroneous energy, a so-called sum peak, is generated in the spectrum of radiation. Therefore, the signal processing device detects the overlap of the pulse waves and suppresses the occurrence of the sum peak by not measuring the wave height of the overlapped pulse waves. Conventionally, there are several methods for detecting the overlap of pulse waves. For example, a pulse wave is detected using a threshold value of wave height, and when the detection interval of a plurality of pulse waves is short, it is determined that overlap of pulse waves has occurred. Further, there is a method of determining whether or not the overlap of pulse waves has occurred based on the feature amount of the signal waveform such as the length of the base of the pulse wave or the rising speed of the pulse wave. Patent Document 1 discloses a technique for determining that overlap of pulse waves has occurred when the differential waveform of a pulse wave changes from a negative value to a positive value.

Japanese Unexamined Patent Publication No. 2009-229127

With any of the conventional methods, it may not be possible to detect the overlap of pulse waves. In the method based on the pulse wave detection interval, the signal processor may not be able to detect the overlap of a plurality of pulse waves when the detection interval is very short. The method based on the feature amount of the signal waveform such as the length of the base has a problem that it is vulnerable to noise because it focuses only on the feature amount. Further, when the threshold value for detecting the pulse wave is lowered, the overlap of a plurality of pulse waves having a high wave height cannot be detected, and when the threshold value is raised, the pulse wave having a low wave height cannot be detected. When performing elemental analysis based on the radiation spectrum, if the overlap of a plurality of pulse waves cannot be detected, a sum peak may occur and an element that does not exist may be erroneously detected. If a pulse wave with a low wave height cannot be detected, it becomes difficult to detect an element having a low radiation energy such as a light element.

The present invention has been made in view of such circumstances, and an object of the present invention is a signal processing method, a learning model generation method, and a signal capable of improving the accuracy of elemental analysis based on a radiation spectrum. To provide processing equipment, radiation detection equipment and computer programs.

The signal processing method according to the present invention is a signal processing method for counting a staircase wave corresponding to radiation detection or a pulse wave obtained by converting the staircase wave by wave height, and is a sequence of signal values constituting a signal including the staircase wave. And, when at least one of the sequence of signal values constituting the signal including the pulse wave is input, the signal including the step wave is configured in a learning model that outputs information regarding the presence or absence of overlap of the plurality of pulse waves. When at least one of the sequence of signal values and the sequence of signal values constituting the signal including the pulse wave is input, and the plurality of pulse waves do not overlap according to the information output by the learning model. It is characterized by counting the staircase wave or the pulse wave.

In one embodiment of the present invention, the overlap of pulse waves is detected by using a learning model. The training model outputs information regarding the presence or absence of overlap of a plurality of pulse waves when a sequence of signal values constituting a signal including a pulse wave and / or a sequence of signal values constituting a signal including a staircase wave is input. To do. By using the learning model, it is possible to effectively detect the overlap of a plurality of pulse waves.

The signal processing method according to the present invention is characterized in that the staircase wave or the pulse wave when a plurality of the pulse waves overlap is not counted according to the information output by the learning model.

In one embodiment of the present invention, when a plurality of pulse waves overlap, counting is not performed according to the detection of radiation. It is possible to prevent erroneous measurement of radiation energy due to the overlap of pulse waves.

In the signal processing method according to the present invention, the learning model is a learning model that outputs the information when a string of signal values constituting a signal including a wave obtained by shaping the pulse wave is further input, and the signal is used. It is characterized in that a sequence of constituent signal values is further input to the learning model.

In one embodiment of the present invention, in addition to the signal including the pulse wave and / or the signal including the step wave, a sequence of signal values constituting the signal including the wave formed by shaping the pulse wave is input to the learning model. The shape of the pulse wave shaped by a filter such as a differential filter differs depending on the presence or absence of a plurality of pulse waves. Therefore, by using a wave obtained by shaping a pulse wave, it is possible to more effectively detect the overlap of a plurality of pulse waves.

In the signal processing method according to the present invention, the learning model is a learning model that outputs the information when at least one feature amount of the staircase wave and the pulse wave is further input, and the feature amount is the learning model. It is characterized by inputting to.

In one embodiment of the present invention, the feature amount of the signal waveform such as the wave height or the time width is further input to the learning model. Since the feature amount of the signal waveform differs depending on the presence or absence of the plurality of pulse waves, the overlap of the plurality of pulse waves can be detected more effectively by using the feature amount of the signal waveform.

The learning model generation method according to the present invention is composed of a single pulse wave, and generates a plurality of pulse wave signals in which at least one of the height, rise, and roundness of the single pulse wave is random. Each of the plurality of pulse waves is composed of a superposed wave in which a plurality of pulse waves are overlapped, and a plurality of superposed wave signals in which at least one of the interval, the wave height, the rising edge and the roundness of the plurality of pulse waves is random are generated, and the plurality of pulse waves Using the signal and the plurality of superimposed wave signals as training data, a learning model is generated that outputs information regarding the presence or absence of overlap of a plurality of pulse waves when a sequence of signal values constituting a signal including an arbitrary pulse wave is input. It is characterized by that.

In one embodiment of the present invention, a plurality of pulse wave signals each consisting of a single pulse wave and a plurality of superimposed wave signals each consisting of a superposed wave in which a plurality of pulse waves are overlapped are subjected to wave height, rising edge, and randomness. Generate by simulation so that at least one of the above is random. The learning model is trained using the generated plurality of pulse wave signals and the plurality of superimposed wave signals as teacher data. A learning model that outputs information regarding the presence or absence of overlapping pulse waves can be generated. Further, by randomizing the bluntness of the pulse wave in the simulation, it is possible to generate a signal close to the actual signal.

The learning model generation method according to the present invention is composed of a single staircase wave, and generates a plurality of staircase wave signals in which at least one of the wave height, rise and roundness of the single staircase wave is random. Each consists of a plurality of staircase waves, and a plurality of pulse waves obtained by converting the plurality of staircase waves overlap each other, and at least one of the intervals, wave heights, rising edges, and rounds of the plurality of staircase waves is random. When the plurality of staircase wave signals and the proximity staircase wave signal are used as teacher data and a sequence of signal values constituting a signal including an arbitrary staircase wave is input, the plurality of staircase wave signals are generated. It is characterized by generating a learning model that outputs information on the presence or absence of overlap of a plurality of pulse waves converted from.

In one embodiment of the present invention, a plurality of staircase wave signals each consisting of a single staircase wave and a plurality of proximity staircase wave signals are simulated so that at least one of wave height, rise and roundness is random. Generated by. The learning model is trained using the generated plurality of staircase wave signals and a plurality of proximity staircase wave signals as teacher data. A learning model that outputs information regarding the presence or absence of overlap of pulse waves converted from staircase waves can be generated.

The signal processing method according to the present invention converts a step wave corresponding to the detection of radiation into a pulse wave, and constitutes a signal including the pulse wave in a signal processing method for counting the pulse wave according to the height of the pulse wave. A product-sum calculation is performed between the sequence of signal values to be performed and the predetermined coefficient sequence, and the predetermined coefficient sequence is between the signal value sequence and the predetermined coefficient sequence constituting the signal including the plurality of overlapping pulse waves. The result of the product-sum calculation and the result of the product-sum calculation between the string of signal values constituting the signal including the single pulse wave and the predetermined coefficient sequence are larger than the predetermined value. It is predetermined to be different depending on whether it is small or small, and it is determined whether or not a plurality of the pulse waves are overlapped according to the result of the product-sum calculation, and the non-overlapping pulse waves are counted. It is characterized by doing.

In one embodiment of the present invention, a product-sum calculation is performed on a sequence of signal values constituting a signal including a pulse wave and a predetermined coefficient sequence, and the overlap of pulse waves is detected based on the result of the product-sum calculation. The coefficient sequence is the result of the product-sum operation between the string of signal values and the coefficient sequence that compose the signal containing a single pulse wave, and the sequence and coefficient of the signal values that compose the signal containing multiple overlapping pulse waves. It is defined so that the result of the product-sum operation between columns is different. Also in this method, the overlap of a plurality of pulse waves can be effectively detected.

The signal processing device according to the present invention is a signal processing device that processes a signal including a staircase wave according to the detection of radiation, and is a sequence of signal values constituting the signal including the staircase wave and a pulse obtained by converting the staircase wave. It is characterized by including a learning model that outputs information regarding the presence or absence of overlap of a plurality of the pulse waves when at least one of a sequence of signal values constituting a signal including a wave is input.

The radiation detection device according to the present invention comprises a radiation detector that outputs a staircase wave according to the energy of radiation at the time of radiation detection, a conversion unit that converts the staircase wave into a pulse wave, and a signal including the staircase wave. A learning model that outputs information regarding the presence or absence of overlap of a plurality of the pulse waves when at least one of the signal value sequence and the signal value sequence constituting the signal including the pulse wave obtained by converting the staircase wave is input. A counting unit that counts the staircase wave or the pulse wave when there is no overlap of the plurality of pulse waves, a wave height of the staircase wave or the pulse wave, and a count number according to the information output by the learning model. It is characterized by including a spectrum generation unit that generates a radiation spectrum according to the above.

The computer program according to the present invention has at least one of a sequence of signal values constituting a signal including a staircase wave corresponding to the detection of radiation and a sequence of signal values constituting a signal including a pulse wave obtained by converting the staircase wave. A sequence of signal values constituting a signal including the staircase wave and a sequence of signal values constituting the signal including the pulse wave are sent to a learning model that outputs information regarding the presence or absence of overlap of a plurality of the pulse waves when input. It is characterized in that a computer is made to execute a process of inputting at least one of them and outputting the information.

In one embodiment of the present invention, the signal processing device uses a learning model to detect the overlap of pulse waves. The training model outputs information regarding the presence or absence of overlap of a plurality of pulse waves when a sequence of signal values constituting a signal including a pulse wave and / or a sequence of signal values constituting a signal including a staircase wave is input. To do. The radiation detector counts staircase waves or pulse waves when there is no overlap of a plurality of pulse waves, and generates a radiation spectrum. The signal processing device can effectively detect the overlap of a plurality of pulse waves, and the radiation detection device can improve the accuracy of elemental analysis based on the spectrum.

The present invention has excellent effects such as being able to effectively detect the overlap of pulse waves and improve the accuracy of elemental analysis.

It is a block diagram which shows the functional structure of the radiation detection apparatus which concerns on Embodiment 1. FIG. It is a schematic characteristic diagram which shows the example of a staircase wave and a pulse wave. It is a schematic characteristic diagram which shows the example of a staircase wave and a pulse wave. It is a schematic characteristic diagram which shows the example of the staircase wave and the pulse wave when the interval at which radiation is detected is short. It is a schematic characteristic diagram which shows the example of the staircase wave and the pulse wave when the interval at which radiation is detected is short. It is a graph which shows typically the example of the signal input to a processing part. It is a conceptual diagram which shows the functional structure example of the learning model which concerns on Embodiment 1. FIG. It is a block diagram which shows the structural example of the learning apparatus which trains a learning model. It is a flowchart which shows the procedure of the process which generates a learning model. It is a flowchart which shows the procedure of the process which the signal processing apparatus which concerns on Embodiment 1 performs. It is a spectrum diagram which shows the spectrum of the radiation obtained by a plurality of methods. It is a block diagram which shows the functional structure of the radiation detection apparatus which concerns on Embodiment 2. It is a block diagram which shows the functional structure of the radiation detection apparatus which concerns on Embodiment 3. It is a conceptual diagram which shows the functional structure example of the learning model which concerns on Embodiment 3. It is a flowchart which shows the procedure of the process which the signal processing apparatus which concerns on Embodiment 3 executes. It is a block diagram which shows the functional structure of the radiation detection apparatus which concerns on Embodiment 4. It is a conceptual diagram which shows the functional structure example of the learning model which concerns on Embodiment 4. It is a flowchart which shows the procedure of the process which the signal processing apparatus which concerns on Embodiment 4 executes. It is a block diagram which shows the functional structure of the processing part which concerns on Embodiment 5. It is a block diagram which shows the functional structure of the radiation detection apparatus which concerns on Embodiment 6. It is a graph which shows the example of the coefficient sequence. It is a graph which shows the example of the coefficient sequence. It is a flowchart which shows the procedure of the process which the signal processing apparatus which concerns on Embodiment 6 executes.

Hereinafter, the present invention will be specifically described with reference to the drawings showing the embodiments thereof.
<Embodiment 1>
FIG. 1 is a block diagram showing a functional configuration of the radiation detection device 10 according to the first embodiment. The radiation detection device 10 includes a radiation detector 1, a signal processing device 2, and an analysis unit 3. The radiation detector 1 includes a radiation detection element 11 and a preamplifier 12. The radiation detection element 11 is a semiconductor radiation detection element such as an SDD (Silicon Drift Detector), which generates an electric charge according to the energy of the incident radiation and outputs a current signal according to the generated electric charge. The preamplifier 12 converts the current signal output by the radiation detection element 11 into a voltage signal, and generates a step wave in which the signal value rises in a step-like manner at the time of radiation detection. The radiation detector 1 outputs a signal including a staircase wave generated by the preamplifier 12.

The signal output by the radiation detector 1 is input to the signal processing device 2. The signal processing device 2 executes the signal processing method. The signal processing device 2 includes an A / D (analog / digital) conversion unit 21. The A / D conversion unit 21 receives a signal including a staircase wave from the radiation detector 1 and A / D-converts the signal including the staircase wave. A waveform shaping unit 22 is connected to the A / D conversion unit 21. The waveform shaping unit 22 receives a signal including a staircase wave from the A / D conversion unit 21. The waveform shaping unit 22 converts the signal including the staircase wave into the signal including the pulse wave by passing the signal including the staircase wave through a predetermined filter and shaping the waveform of the signal. The filter used by the waveform shaping unit 22 is, for example, a differential filter or a trapezoidal shaping filter. By the processing in the waveform shaping unit 22, the staircase wave is converted into a pulse wave, the noise contained in the signal is reduced, and a predetermined amplification is performed. The waveform shaping unit 22 outputs a signal including a pulse wave. The waveform shaping unit 22 corresponds to the conversion unit.

2A and 2B are schematic characteristic diagrams showing examples of staircase waves and pulse waves. The horizontal axis in the figure shows time, and the vertical axis shows signal values. FIG. 2A shows a signal including a staircase wave output by the radiation detector 1. The radiation detector 1 outputs a step wave whose signal value rises in a step-like manner each time radiation is detected. In response to a single radiation detection, a step wave is generated in which the signal value rises in a step-like manner. When the radiation detector 1 detects radiation a plurality of times, a signal including a plurality of staircase waves is output. Each time radiation is detected, the signal value rises. The height of the step at which the signal value rises is defined as the wave height of the staircase wave. The height of the staircase wave corresponds to the energy of the radiation. In reality, the staircase wave is not completely stepped, and the signal waveform contains rising and rounding. The rising edge is the distortion of the signal waveform when the signal value rises from the reference value, and the rounding is the distortion of the signal waveform when the staircase wave ends.

FIG. 2B shows a signal obtained by converting the signal shown in FIG. 2A by the waveform shaping unit 22. The staircase wave is converted into a pulse wave. A pulse wave is a signal in which a signal value rises from a predetermined signal reference at which the signal value becomes zero to a peak value and then falls to a signal reference. The signal reference is, for example, zero. The height from the signal reference to the peak value is defined as the wave height of the pulse wave. The height of the pulse wave corresponds to the energy of the radiation. The shape of the pulse wave includes rising and rounding. The rising edge is the distortion of the signal waveform when the signal value rises from the reference value, and the rounding is the distortion of the signal waveform when the pulse wave ends. FIG. 2B shows an example in which the radiation detector 1 detects radiation a plurality of times at long intervals and a plurality of pulse waves are not superimposed.

3A and 3B are schematic characteristic diagrams showing examples of staircase waves and pulse waves when the intervals at which radiation is detected are short. The horizontal axis in the figure shows time, and the vertical axis shows signal values. FIG. 3A shows a signal including a staircase wave output by the radiation detector 1. Compared to the example shown in FIG. 2A, the interval at which the radiation detector 1 detects radiation a plurality of times is short, and the interval between the plurality of staircase waves is short. FIG. 3B shows a signal obtained by converting the signal shown in FIG. 3A by the waveform shaping unit 22. The interval between the plurality of pulse waves is short, and a superposed wave in which the plurality of pulse waves overlap is formed. The wave height of the superimposed wave is different from the wave height of a single pulse wave, and the erroneous radiation energy is measured according to the wave height of the superimposed wave.

The processing unit 23 and the pulse detection unit 24 are connected to the waveform shaping unit 22. The waveform shaping unit 22 inputs a signal including a pulse wave to the processing unit 23 and the pulse detecting unit 24. The pulse detection unit 24 receives a signal from the waveform shaping unit 22 and detects a pulse wave contained in the signal. For example, the pulse detection unit 24 determines that a pulse wave has been detected when the signal value exceeds a predetermined threshold value. The pulse detection unit 24 is connected to the processing unit 23. When the pulse detection unit 24 detects the pulse wave, the pulse detection unit 24 inputs information indicating that the pulse wave is detected to the processing unit 23.

The processing unit 23 receives a signal including a pulse wave from the waveform shaping unit 22, and an information indicating that the pulse wave is detected is input from the pulse detection unit 24. The processing unit 23 is configured by using an element that performs an operation. The processing unit 23 includes a buffer memory 231 and a learning model 232 for determining whether or not a plurality of pulse waves overlap. For example, the learning model 232 is configured by using FPGA (field-programmable gate array). When the pulse wave is detected, the processing unit 23 uses the learning model 232 to determine whether or not the signal includes the overlap of a plurality of pulse waves. A method for determining whether or not a plurality of pulse waves overlap will be described later.

A wave height measuring unit 25 is connected to the waveform shaping unit 22 and the processing unit 23. The waveform shaping unit 22 inputs a signal including a pulse wave to the wave height measuring unit 25. The processing unit 23 inputs information indicating whether or not a plurality of pulse waves overlap to the wave height measuring unit 25. The wave height measuring unit 25 measures the wave height of the pulse wave included in the signal input from the waveform shaping unit 22 when the plurality of pulse waves do not overlap. When there is an overlap of a plurality of pulse waves, the wave height measuring unit 25 does not measure the wave height of the pulse wave included in the signal input from the waveform shaping unit 22.

A counting unit 26 is connected to the wave height measuring unit 25. The wave height measuring unit 25 inputs the measured pulse wave height to the counting unit 26. The counting unit 26 counts pulse waves according to wave height. For example, the counting unit 26 is a multi-channel analyzer. The counting unit 26 may be in a form of counting pulse waves for all wave heights, or may be in a form of counting pulse waves only for a specific wave height. The signal processing device 2 outputs data showing the relationship between the wave height of the pulse wave and the number of counts counted by the counting unit 26. The count number corresponds to the number of times that the radiation detector 1 detects radiation having energy corresponding to the wave height of the pulse wave.

When there is an overlap of a plurality of pulse waves, the wave height measuring unit 25 does not measure the wave height of the pulse waves, so that the signal processing device 2 does not count the overlapped pulse waves. The wave height measuring unit 25 may measure the wave height of the overlapping pulse waves, but may not input the measured wave height to the counting unit 26. The wave height measuring unit 25 may also input the wave heights of the overlapping pulse waves to the counting unit 26, and the counting unit 26 may be in a form in which the overlapping pulse waves are not counted. The wave height measuring unit 25 may also input the wave heights of the overlapping pulse waves to the counting unit 26, and the counting unit 26 may be in a form of distinguishing between the non-overlapping pulse waves and the overlapping pulse waves.

The analysis unit 3 is composed of a computer such as a personal computer. The analysis unit 3 is input with the data output by the signal processing device 2. The analysis unit 3 performs a process of generating a spectrum of the radiation detected by the radiation detector 1 from the relationship between the wave height of the pulse wave and the count number. The analysis unit 3 corresponds to the spectrum generation unit. The analysis unit 3 may further perform further processing such as elemental analysis of the radiation source based on the generated spectrum of radiation. For example, the sample is irradiated with radiation, the characteristic X-rays generated from the sample are detected by the radiation detector 1, and qualitative analysis or quantitative analysis of the elements contained in the sample is performed based on the spectrum of the characteristic X-rays. The signal processing device 2 may also have a function of generating a radiation spectrum.

If the overlapping pulse waves are erroneously counted as a single pulse wave and the wrong radiation energy is measured, the radiation spectrum will include the sum peak, which is the peak with the wrong energy. When elemental analysis is performed based on a spectrum including a sum peak, there is a risk of erroneously detecting an element that does not exist according to the sum peak. Alternatively, the sum peak may overlap the peak of an existing element, and the amount of the element may be excessively detected. For example, when a plurality of pulse waves generated by detecting the K line of Fe (iron) overlap, the thumb peak overlaps the peak of the L line of Pb (lead), and the amount of Pb is excessively detected. When a plurality of pulse waves generated by detecting the K line of Cu (copper) overlap, the thumb peak overlaps the peak of the K line of Cd (cadmium), and the amount of Cd is excessively detected.

A method of determining whether or not a plurality of pulse waves overlap in the processing unit 23 will be described. FIG. 4 is a graph schematically showing an example of a signal input to the processing unit 23. The horizontal axis in the figure indicates time, and the vertical axis indicates signal value. The signal is composed of a time series of discrete signal values obtained at predetermined time intervals. That is, the signal is represented by one-dimensional data consisting of a sequence of signal values.

As shown in FIGS. 2B and 3B, a signal including a single pulse wave and a signal including a superimposed wave in which a plurality of pulse waves are overlapped have different signal shapes and different time changes in signal values. The learning model 232 is trained in advance so as to output information indicating whether or not a plurality of pulse waves are overlapped when a sequence of signal values constituting a signal including a pulse wave is input.

FIG. 5 is a conceptual diagram showing a functional configuration example of the learning model 232 according to the first embodiment. The learning model 232 uses a fully coupled neural network with an input layer, a plurality of intermediate layers, and an output layer, each having a plurality of nodes. The input layer has a plurality of nodes 41 into which a sequence of signal values constituting a signal including a pulse wave is input. One signal value in the signal value sequence is input to one node 41, and each signal value is input to one of the nodes 41. For example, the input layer includes m nodes 41, and m signal values constituting a signal including a pulse wave are input to the input layer.

The learning model 232 has an intermediate layer of n (n is a natural number) layer. The first intermediate layer has a plurality of nodes 421. Each node 41 of the input layer outputs a signal value to a plurality of nodes 421. The plurality of nodes 421 receive signal values from the nodes 41 of the input layer, calculate the signal values using parameters, and output the calculation result data to the plurality of nodes 422 included in the second intermediate layer. The nodes included in each intermediate layer receive data from a plurality of nodes in the previous intermediate layer, calculate the received data using parameters, and output the data to the nodes in the subsequent intermediate layer. For example, the node has f (Σ (w * x), where x is the value of the data received from each node in the previous layer, w is the weight corresponding to each node, b is the bias value, and f () is the activation function. ) + B) is performed, and the data of the calculation result is output to a plurality of nodes in the subsequent layer. The activation function is, for example, a relu function or a sigmoid function. The activation function may be another function generally used in machine learning.

The output layer of the learning model 232 has a single node 43. The plurality of nodes 42n included in the nth intermediate layer output data to the node 43 included in the output layer. The node 43 of the output layer receives data from a plurality of nodes 42n, calculates the received data using parameters, and outputs information indicating the presence or absence of overlap of the plurality of pulse waves. For example, the activation function at the node 43 is a function that outputs data indicating whether or not the calculation result of (Σ (w * x) + b) is a positive value. For example, the node 43 may output a value of 1 as data indicating a positive value and output a value of zero as data indicating a value of zero or less. For example, the data showing a positive value is information indicating that there is no overlap of a plurality of pulse waves, and the data showing a value of zero or less is information indicating that there is an overlap of a plurality of pulse waves. Note that the node 43 may output the probability that a plurality of pulse waves overlap each other as information indicating the presence or absence of the overlap of the pulse waves. As the learning model 232, a convolutional neural network (CNN: Convolutional Neural Network) or a recurrent neural network (RNN: Recurrent Neural Network) may be used as the neural network.

The learning of the learning model 232 is performed using a computer. FIG. 6 is a block diagram showing a configuration example of a learning device 5 that learns the learning model 232. The learning device 5 executes the learning model generation method. The learning device 5 is a computer such as a server device. The learning device 5 includes a calculation unit 51, a memory 52, a storage unit 53, a display unit 54, and an operation unit 55. The calculation unit 51 is configured by using, for example, a CPU (Central Processing Unit), a GPU (Graphics Processing Unit), or a multi-core CPU. The calculation unit 51 may be configured by using a quantum computer. The memory 52 stores temporary data generated by the calculation. The memory 52 is, for example, a RAM (Random Access Memory). The storage unit 53 is non-volatile, for example, a hard disk. The display unit 54 is, for example, a liquid crystal display or an EL display (Electroluminescent Display). The operation unit 55 accepts the input of information such as text by accepting the operation from the user. The operation unit 55 is, for example, a keyboard or a touch panel. The storage unit 53 stores the computer program 531. The calculation unit 51 executes the process according to the computer program 531.

FIG. 7 is a flowchart showing the procedure of the process of generating the learning model 232. Hereinafter, the step is abbreviated as S. The calculation unit 51 executes the following processing according to the computer program 531. The arithmetic unit 51 generates a plurality of pulse wave signals each consisting of a single pulse wave (S11). In S11, the calculation unit 51 generates a plurality of pulse wave signals in which at least one of the height, rise, and roundness of the pulse wave is random by simulation. The height, rise and roundness of the pulse wave may all be random. A pulse wave signal is composed of a sequence of signal values. For example, the number of signal values included in the pulse wave signal is the same as the number of nodes 41 in the input layer of the learning model 232.

The calculation unit 51 generates the wave height of the pulse wave, the rising time constant, and the rising start time from random numbers having a uniform distribution. As the random number, a random number having an exponential distribution or a Poisson distribution may be used. Further, the calculation unit 51 randomly generates a time constant of the bluntness of the pulse wave. The calculation unit 51 superimposes white noise on the pulse wave according to the generated parameter to generate a pulse wave signal. The noise superimposed on the pulse wave may be 1 / f noise, and the noise may not be superimposed. The noise superimposed on the pulse wave is preferably noise according to the characteristics of the radiation detector 1. The pulse wave signal is a signal as shown in FIG. 2B. The calculation unit 51 stores the pulse wave data 532 including the generated plurality of pulse wave signals in the storage unit 53. When the size of the radiation detector 1 is large, the roundness of the actually measured signal fluctuates greatly. By randomizing the bluntness of the pulse wave in the simulation, it is possible to generate a signal close to the actual signal.

Next, the calculation unit 51 generates a plurality of superimposed wave signals composed of superimposed waves in which a plurality of pulse waves are overlapped with each other (S12). In S12, the calculation unit 51 generates a plurality of superimposed wave signals in which two pulse waves having at least one of a wave height, a rising edge, and a randomness are random are overlapped with each other by simulation. The superimposed wave signal is also composed of a sequence of signal values. The calculation unit 51 generates the wave height, rise time constant, and rise start time of the first pulse wave, and the wave height, rise time constant, and rise start time of the second pulse wave from a uniformly distributed random number. Further, the calculation unit 51 randomly generates a time constant of the bluntness of the pulse wave. The time constant of the rounding may be the same for the two pulse waves. The calculation unit 51 superimposes noise on the superposed wave according to the generated parameter to generate a superposed wave signal. The superimposed wave signal is a signal as shown in FIG. 3B. The calculation unit 51 stores the superimposed wave data 533 including the generated plurality of superimposed wave signals in the storage unit 53.

Next, the calculation unit 51 performs processing for generating the learning model 232 by using the plurality of pulse wave signals included in the pulse wave data 532 and the plurality of superimposed wave signals included in the superimposed wave data 533 as teacher data ( S13). In S13, the calculation unit 51 inputs a sequence of signal values constituting the plurality of pulse wave signals and the plurality of superimposed wave signals to the input layer of the learning model 232, respectively. One signal value is input to each of the nodes 41 of the input layer. The calculation unit 51 associates the pulse wave signal with information indicating that the plurality of pulse waves do not overlap, and associates the superimposed wave signal with information indicating that the plurality of pulse waves overlap. The learning model 232 outputs information indicating whether or not a plurality of pulse waves overlap from the node 43 of the output layer. The calculation unit 51 calculates the error of the information by the error function using the information associated with the input pulse wave signal or the superimposed wave signal and the information output from the node 43 as variables, and the error is calculated by the error back propagation method. The calculation parameters of each node of the training model 232 are adjusted so as to be the minimum. That is, when a pulse wave signal is input, information close to the information indicating that there is no overlap of a plurality of pulse waves is output, and when a superimposed wave signal is input, information indicating that a plurality of pulse waves are overlapped is output. The parameters are adjusted so that close information is output.

The calculation unit 51 performs machine learning of the learning model 232 by repeating the process using the plurality of pulse wave signals and the plurality of superimposed wave signals and adjusting the parameters of each node of the learning model 232. The calculation unit 51 stores the learned data 534 in which the adjusted final parameters are recorded in the storage unit 53. In this way, the trained learning model 232 is generated. After the end of S13, the calculation unit 51 ends the process. The learning model 232 included in the processing unit 23 is manufactured based on the trained data 534. For example, the learning model 232 is manufactured by writing the parameters recorded in the trained data 534 to the FPGA included in the processing unit 23. The calculation unit 51 may perform machine learning of the learning model 232 using the actually measured pulse wave signal and the superimposed wave signal instead of using the pulse wave signal and the superimposed wave signal created by the simulation. ..

Next, the processing executed by the signal processing device 2 will be described. FIG. 8 is a flowchart showing a procedure of processing executed by the signal processing device 2 according to the first embodiment. When radiation is incident on the radiation detection element 11, the radiation detector 1 generates a staircase wave according to the energy of the radiation and outputs a signal including the staircase wave. The signal processing device 2 receives a signal including a staircase wave from the radiation detector 1 (S21). The A / D conversion unit 21 A / D converts the input signal (S22). The A / D conversion unit 21 inputs the A / D converted signal to the waveform shaping unit 22. The waveform shaping unit 22 shapes the waveform of the input signal (S23). By the waveform shaping, the waveform shaping unit 22 reduces the noise included in the signal and converts the staircase wave included in the signal into a pulse wave. The waveform shaping unit 22 inputs a signal including a pulse wave to the processing unit 23, the pulse detecting unit 24, and the wave height measuring unit 25.

The signal input to the processing unit 23 is composed of a time series of signal values. The processing unit 23 sequentially stores the signal values in the buffer memory 231 (S24). The processes S21 to S24 are individually and repeatedly executed, and the signal values are sequentially stored in the buffer memory 231. The buffer memory 231 is a first-in first-out memory, and stores a plurality of sequentially input signal values. When a new signal value is input while the amount of the plurality of signal values stored in the buffer memory 231 has reached the upper limit, the buffer memory 231 first stores the stored signal values. The signal value is erased and a new signal value is stored.

The pulse detection unit 24 waits for the detection of the pulse wave included in the input signal (S25). In S25, for example, the pulse detection unit 24 determines that the pulse wave included in the signal has been detected when the signal value exceeds a predetermined threshold value. The threshold value is stored in the processing unit 23 in advance. If no pulse wave is detected (S25: NO), the pulse detection unit 24 repeats the process of S25.

When the pulse wave included in the signal is detected (S25: YES), the pulse detection unit 24 inputs information indicating that the pulse wave is detected to the processing unit 23, and the processing unit 23 inputs the pulse wave. A sequence of signal values constituting the including signal is input to the learning model 232 (S26). In S26, for example, the processing unit 23 inputs a plurality of signal values stored in the buffer memory 231 to the learning model 232 at the time when the information indicating that the pulse wave is detected is input. The processing unit 23 may input the signal values constituting the signal to the learning model 232 after thinning out the signal values. As described above, the learning model 232 to which the signal including the pulse wave is input performs the calculation of the neural network and outputs the information indicating whether or not the plurality of pulse waves overlap.

The processing unit 23 inputs information indicating whether or not the plurality of pulse waves output by the learning model 232 overlap to the wave height measuring unit 25. The wave height measuring unit 25 identifies whether or not a plurality of pulse waves overlap each other based on the input information (S27). When information indicating that there is no overlap of the plurality of pulse waves is input and there is no overlap of the plurality of pulse waves (S27: NO), the wave height measuring unit 25 is the pulse wave included in the input signal. The wave height is measured (S28). The wave height measuring unit 25 inputs the measured pulse wave height to the counting unit 26. The counting unit 26 counts pulse waves according to the wave height input from the wave height measuring unit 25 (S29), and ends the process. When information indicating that there is an overlap of a plurality of pulse waves is input and there is an overlap of a plurality of pulse waves (S27: YES), the wave height measuring unit 25 measures the wave height for the plurality of overlapping pulse waves. Is not performed, and the signal processing device 2 ends the processing. As a result, the counting unit 26 does not count the plurality of overlapping pulse waves. The signal processing device 2 repeatedly executes the processes of S21 to S29 individually.

The signal processing device 2 outputs data showing the relationship between the height of the pulse wave and the number of counts counted by the counting unit 26. The analysis unit 3 inputs the data output by the signal processing device 2, and generates a spectrum of the radiation detected by the radiation detector 1 based on the data.

FIG. 9 is a spectrum diagram showing the spectra of radiation obtained by a plurality of methods. The horizontal axis in the figure shows the energy of radiation in units of keV, and the vertical axis shows the number of counts of radiation. The spectrum shown in FIG. 9 includes a peak near 11.8 keV and a peak near 12.4 keV. However, all of these peaks are sum peaks obtained by measuring the heights of a plurality of overlapping pulse waves. In FIG. 9, the spectrum when the overlap of the pulse waves is not detected is shown by a broken line, and the spectrum when the overlap of the pulse waves is detected by the conventional technique is shown by a thin solid line. In addition, the spectrum when the overlap of pulse waves is detected according to this embodiment is shown by a thick solid line.

As shown in FIG. 9, when the overlap of pulse waves is detected, the intensity of the thumb peak decreases. In the present embodiment, the intensity of the thumb peak is further reduced as compared with the prior art. Therefore, in the present embodiment, it is possible to effectively detect the overlap of a plurality of pulse waves, which has been difficult to detect in the past. By effectively detecting the overlap of a plurality of pulse waves, it is possible to suppress erroneous measurement of radiation energy due to the overlap of pulse waves. As a result, it is possible to suppress the occurrence of thumb peaks in the radiation spectrum and improve the accuracy of elemental analysis based on the spectrum. For example, it is possible to prevent erroneous detection of non-existent elements. Further, even if the threshold value for detecting the pulse wave is lowered, it is possible to detect the overlap of a plurality of pulse waves. Therefore, it is possible to lower the threshold value for detecting the pulse wave, and it becomes easy to detect an element having a low radiation energy such as a light element.

As described in detail above, the signal processing device 2 detects the overlap of a plurality of pulse waves by using the learning model 232. The learning model 232 outputs information indicating the presence or absence of overlap of a plurality of pulse waves when a sequence of signal values constituting a signal including a pulse wave is input. By using the learning model 232, it is possible to detect the overlap of a plurality of pulse waves, which has been difficult to detect in the past. Since the learning model 232 determines whether or not a plurality of pulse waves overlap from the entire signal waveform, it is less susceptible to noise than the conventional determination method based on the feature amount of the signal waveform, and is more reliable. The presence or absence of overlap of multiple pulse waves can be obtained. As a result, the radiation detection device 10 can improve the accuracy of elemental analysis based on the radiation spectrum. Further, the signal processing device 2 can execute the processing at a higher speed than the case of processing the image of the signal by processing the sequence of the signal values, and can detect the overlap of the pulse waves in almost real time. it can.

<Embodiment 2>
FIG. 10 is a block diagram showing a functional configuration of the radiation detection device 10 according to the second embodiment. The configuration and function of the radiation detector 1 and the analysis unit 3 are the same as those in the first embodiment. The processing unit 23 is connected to the waveform shaping unit 22 and the A / D conversion unit 21. The processing unit 23 inputs a signal including a pulse wave from the waveform shaping unit 22, and inputs a signal including a step wave from the A / D conversion unit 21. The processing unit 23 stores the signal value constituting the signal including the pulse wave and the signal value constituting the signal including the staircase wave in the buffer memory 231.

The learning model 232 is trained in advance so as to output information indicating whether or not a plurality of pulse waves overlap when a signal including a pulse wave and a signal including a staircase wave are input. The learning model 232 uses a fully connected neural network as in the first embodiment. The input layer of the learning model 232 includes a plurality of nodes 41 in which a sequence of signal values constituting a signal including a pulse wave is input, and a plurality of nodes 41 in which a sequence of signal values constituting a signal including a staircase wave is input. Includes node 41. The node 43 of the output layer of the learning model 232 outputs information indicating the presence or absence of overlapping of the plurality of pulse waves, as in the first embodiment. The configuration of the parts other than the A / D conversion unit 21 and the processing unit 23 of the signal processing device 2 is the same as that of the first embodiment.

The learning of the learning model 232 is performed by the learning device 5 as in the first embodiment. In addition to generating a plurality of pulse wave signals in S11, the calculation unit 51 of the learning device 5 generates a stair wave signal corresponding to the signal including the stair wave before the pulse wave is converted by the waveform shaping unit 22. .. On the contrary, the calculation unit 51 generates a plurality of staircase wave signals including staircase waves by simulation, in which at least one of the wave height, rise and roundness of the staircase wave is random, and white noise is superimposed, and the staircase wave signal is generated. A pulse wave signal may be generated by differentiating. The noise superimposed on the signal may be 1 / f noise, and the noise may not be superimposed. Further, in addition to the generation of the plurality of superimposed wave signals in S12, the arithmetic unit 51 generates a proximity staircase wave signal corresponding to the signal including the plurality of staircase waves before the superimposed wave is converted by the waveform shaping unit 22. To do. On the contrary, the calculation unit 51 generates a plurality of proximity staircase wave signals including two proximity staircase waves in which at least one of the wave height, the rise and the roundness is random by simulation, and differentiates the proximity staircase wave signals. May generate a superimposed wave signal. The calculation unit 51 may use the actually measured staircase wave signal and the proximity staircase wave signal instead of using the staircase wave signal and the proximity staircase wave signal created by the simulation.

In S13, the calculation unit 51 performs processing for generating the learning model 232 by using the plurality of pulse wave signals and staircase wave signals and the plurality of superimposed wave signals and proximity staircase wave signals as teacher data. In S13, the calculation unit 51 inputs a sequence of signal values constituting the pulse wave signal and the step wave signal to the input layer of the learning model 232. Further, the calculation unit 51 inputs a sequence of signal values constituting the superimposed wave signal and the proximity staircase wave signal to the input layer of the learning model 232. The calculation unit 51 adjusts the calculation parameters of each node of the learning model 232 by the error back propagation method. The calculation unit 51 repeats the process using the plurality of pulse wave signals and the plurality of superimposed wave signals to perform machine learning of the learning model 232.

The signal processing device 2 executes the processes of S21 to S25 as in the first embodiment. In S26, the processing unit 23 comprises a string of signal values constituting a signal including a pulse wave input from the waveform shaping unit 22, and a signal value constituting a signal including a staircase wave input from the A / D conversion unit 21. The column of is input to the training model 232. The learning model 232 performs a neural network operation and outputs information indicating whether or not a plurality of pulse waves overlap. The signal processing device 2 executes the processes of S27 to S29 as in the first embodiment. The signal processing device 2 outputs data showing the relationship between the wave height of the pulse wave and the number of counts, and the analysis unit 3 inputs the data output by the signal processing device 2 and the radiation spectrum detected by the radiation detector 1. To generate. The analysis unit 3 may perform elemental analysis of the radiation source based on the spectrum of radiation.

As described in detail above, in the second embodiment, the signal processing device 2 inputs a signal including a step wave before being converted into a pulse wave into the learning model 232 in addition to the signal including the pulse wave, and inputs the pulse wave. Detects overlap. When a plurality of pulse waves overlap, the interval between the staircase waves is short, and the waveform of the signal including the staircase wave is also different from that when the pulse waves do not overlap. Therefore, by using the signal including the staircase wave in addition to the signal including the pulse wave, it becomes easier to determine whether or not there is an overlap of the pulse waves. The signal processing device 2 can detect the overlap of a plurality of pulse waves more effectively as compared with the first embodiment, and the radiation detection device 10 can further improve the accuracy of elemental analysis based on the spectrum. it can.

Note that the signal processing device 2 may be in a form of detecting the overlap of a plurality of pulse waves by using a signal including a staircase wave without using a signal including a pulse wave. In this embodiment, the learning model 232 is trained in advance so as to output information indicating whether or not a plurality of pulse waves overlap when a signal including a staircase wave is input. The input layer of the learning model 232 includes a plurality of nodes 41 into which a sequence of signal values constituting a signal including a staircase wave is input. In the learning of the learning model 232, the learning device 5 generates a staircase wave signal and a proximity staircase wave signal, and performs machine learning of the learning model 232 using the plurality of staircase wave signals and the proximity staircase wave signal as teacher data. In S26, the signal processing device 2 does not input the sequence of signal values constituting the signal including the pulse wave to the learning model 232, but inputs the sequence of the signal values constituting the signal including the staircase wave to the learning model 232. Similarly, the learning model 232 outputs information indicating whether or not a plurality of pulse waves overlap. Similar to the first embodiment, the signal processing device 2 can effectively detect the overlap of a plurality of pulse waves.

Further, in addition to the signal including the pulse wave and / or the signal including the staircase wave, the signal processing device 2 further inputs the signal including the shaped wave obtained by shaping the pulse wave by the waveform shaping unit 22 into the learning model 232. , It may be in the form of detecting the overlap of a plurality of pulse waves. The shaped wave corresponds to, for example, the second derivative of the staircase wave. In the learning model 232, a sequence of signal values constituting a signal including a pulse wave and / or a sequence of signal values constituting a signal including a staircase wave and a sequence of signal values constituting a signal including a shaping wave are input. In this case, it is learned in advance to output information indicating whether or not a plurality of pulse waves overlap. The input layer of the learning model 232 includes a plurality of nodes 41 into which a sequence of signal values constituting a signal including a shaping wave is input. In the learning of the learning model 232, the learning device 5 generates a signal including a shaped wave and a signal obtained by shaping the superimposed wave signal, and generates a pulse wave signal and / or a staircase wave signal, a signal containing the shaped wave, and a superimposed wave. Machine learning is performed using the signal and / or the proximity step wave signal and the signal obtained by shaping the superimposed wave signal as training data. In S26, the signal processing device 2 includes a sequence of signal values constituting a signal including a pulse wave and / or a sequence of signal values constituting a signal including a staircase wave, and a sequence of signal values constituting a signal including a shaping wave. Is input to the learning model 232. The learning model 232 outputs information indicating whether or not a plurality of pulse waves overlap. Also in this form, the signal processing device 2 can more effectively detect the overlap of a plurality of pulse waves.

Further, the signal processing device 2 is in the form of detecting the overlap of the pulse waves by further inputting the feature amount of the signal waveform into the learning model 232 in addition to the signal including the pulse wave and / or the signal including the step wave. You may. The feature amount of the signal waveform is, for example, a wave height, a rising time constant, a time width of a signal wave, or the like. The learning model 232 receives a plurality of pulses when a sequence of signal values constituting a signal including a pulse wave and / or a sequence of signal values constituting a signal including a staircase wave and a feature amount of a signal waveform are input. It has been learned in advance to output information indicating whether or not there is overlap of waves. In S26, the signal processing device 2 inputs a sequence of signal values constituting a signal including a pulse wave and / or a sequence of signal values constituting a signal including a staircase wave and a feature amount of a signal waveform into the learning model 232. To do. The learning model 232 outputs information indicating whether or not a plurality of pulse waves overlap. Further, a signal including a shaping wave may also be used. Since the feature amount of the signal waveform is different between the single pulse wave and the superimposed wave, the overlap of a plurality of pulse waves can be detected more effectively by using the feature amount of the signal waveform. Therefore, the radiation detection device 10 can further improve the accuracy of elemental analysis based on the spectrum.

<Embodiment 3>
In the third embodiment, a mode in which the learning model 232 determines the number of overlapping pulse waves is shown. FIG. 11 is a block diagram showing a functional configuration of the radiation detection device 10 according to the third embodiment. The configuration and function of the radiation detector 1 and the analysis unit 3 are the same as those in the first embodiment. The signal processing device 2 includes an A / D conversion unit 21, a waveform shaping unit 22, a processing unit 23, a wave height measuring unit 25, and a counting unit 26. The processing unit 23 does not have a buffer memory. The processing unit 23 is input with at least one of a signal including a staircase wave from the A / D conversion unit 21 and a signal including a pulse wave from the waveform shaping unit 22. The processing unit 23 inputs information indicating the number of overlapping pulse waves to the wave height measuring unit 25.

FIG. 12 is a conceptual diagram showing a functional configuration example of the learning model 232 according to the third embodiment. The learning model 232 outputs the probability of the number of overlapping pulse waves when a sequence of signal values constituting a signal including a pulse wave and / or a sequence of signal values constituting a signal including a staircase wave is input. It has been learned in advance so as to do so. The output layer of the learning model 232 has a plurality of nodes 43. Each node 43 receives data from a plurality of nodes 42n, calculates the received data using parameters, and outputs the probability that the number of overlapping pulse waves included in the signal is a specific number. For example, one node 43 outputs the probability that the signal does not contain a pulse wave, and the other node 43 outputs the probability that a plurality of pulse waves do not overlap and the number of pulse waves is one. However, the other node 43 outputs the probability that the number of overlapping pulse waves is two. The node 43 may output the probability that the number of overlapping pulse waves is a specific number as a real number of 0 to 1, or may output as a binary value of 0 or 1.

In the training of the learning model 232, the learning device 5 uses the signal not including the pulse wave and the staircase wave, the pulse wave signal and / or the staircase wave signal, and the superimposed wave signal and / or the proximity staircase wave signal as training data. Perform machine learning. The learning device 5 associates information that the probability that there is no pulse wave is 1 and the other probability is 0 with the signal that does not include the pulse wave and the step wave, and the pulse wave signal and / or the step wave signal is pulsed. The information that the probability of one wave is 1 and the other probability is 0 is associated, and the superimposed wave signal and / or the proximity staircase wave signal has a probability of having multiple pulse waves of 1 and other probabilities. Associate information that is 0. The learning device 5 calculates an error of information by an error function having each probability indicated by the information associated with the signal input to the input layer and each probability output from the output layer as variables, and back-propagates the information. The calculation parameters of each node of the training model 232 are adjusted so that the error is minimized by the method.

FIG. 13 is a flowchart showing a processing procedure executed by the signal processing device 2 according to the third embodiment. The signal processing device 2 receives a signal including a step wave from the radiation detector 1 (S31), and the A / D conversion unit 21 converts the signal into A / D (S32). The A / D conversion unit 21 inputs the A / D converted signal to the waveform shaping unit 22. The waveform shaping unit 22 shapes the waveform of the input signal (S33). Waveform shaping converts the staircase wave into a pulse wave. The waveform shaping unit 22 inputs a signal including a pulse wave to the wave height measuring unit 25.

The processing unit 23 receives a signal including a step wave from the A / D conversion unit 21 and / or a signal including a pulse wave from the waveform shaping unit 22, and learns a sequence of signal values constituting the input signal. Input to model 232 (S34). As described above, the learning model 232 performs the operation of the neural network and outputs the probability of the number of overlapping pulse waves. The processing unit 23 inputs information indicating the number of overlapping pulse waves to the wave height measuring unit 25. At this time, the processing unit 23 may input the probability output by the learning model 232 to the wave height measuring unit 25 as information indicating the number of overlapping pulse waves. Further, the processing unit 23 determines the number of overlapping pulse waves based on the probability, and even if the information indicating the determined number is input to the wave height measuring unit 25 as the information indicating the number of overlapping pulse waves. Good. For example, the processing unit 23 determines that the number having the maximum probability or the number having the probability equal to or higher than a predetermined value is the number of overlapping pulse waves.

The wave height measuring unit 25 identifies whether or not the number of pulse waves is one based on the input information (S35). When the number of pulse waves is one (S35: YES), the wave height measuring unit 25 measures the wave height of the pulse wave included in the input signal (S36). The wave height measuring unit 25 inputs the measured pulse wave height to the counting unit 26. The counting unit 26 counts the pulse wave according to the wave height (S37), and ends the process. When the number of pulse waves is other than one (S35: NO), the wave height measuring unit 25 does not measure the wave height, and the signal processing device 2 ends the processing. As a result, the counting unit 26 does not count the plurality of overlapping pulse waves. The signal processing device 2 repeatedly executes the processes of S31 to S37.

The signal processing device 2 outputs data showing the relationship between the height of the pulse wave and the number of counts counted by the counting unit 26. The analysis unit 3 inputs the data output by the signal processing device 2, and generates a spectrum of the radiation detected by the radiation detector 1 based on the data. The analysis unit 3 may perform elemental analysis of the radiation source based on the spectrum of radiation. Also in this embodiment, the signal processing device 2 can effectively detect the overlap of a plurality of pulse waves, and the radiation detection device 10 can further improve the accuracy of elemental analysis based on the spectrum.

Note that the signal processing device 2 may be in a form of detecting the overlap of a plurality of pulse waves by using a signal including a shaped wave obtained by further shaping the pulse wave by the waveform shaping unit 22. Further, the signal processing device 2 may be in a form of detecting the overlap of a plurality of pulse waves by using the feature amount of the signal waveform.

<Embodiment 4>
In the fourth embodiment, the learning model 232 shows a mode in which the number of staircase waves and the wave height are determined. FIG. 14 is a block diagram showing a functional configuration of the radiation detection device 10 according to the fourth embodiment. The configuration and function of the radiation detector 1 and the analysis unit 3 are the same as those in the first embodiment. The signal processing device 2 includes an A / D conversion unit 21, a processing unit 23, and a counting unit 26. The A / D conversion unit 21 inputs a signal including a staircase wave to the processing unit 23.

FIG. 15 is a conceptual diagram showing a functional configuration example of the learning model 232 according to the fourth embodiment. The learning model 232 is pre-learned to output the probability of the number of staircase waves included in the signal and the wave height of the staircase wave when a sequence of signal values constituting the signal including the staircase wave is input. .. The output layer of the learning model 232 has a plurality of nodes 43. Each node 43 receives data from a plurality of nodes 42n, and calculates the received data using parameters. One node 43 outputs the height of the staircase wave included in the signal. The other node 43 outputs the probability of the number of staircase waves included in the signal. For example, a node 43 that outputs the probability that the signal does not include staircase waves, a node 43 that outputs the probability that the number of staircase waves is one, and a node that outputs the probability that the number of staircase waves is two. There is 43. The node 43 may output the probability that the number of staircase waves is a specific number as a real number from 0 to 1, or may output as a binary value of 0 or 1. The probability of the number of staircase waves included in the signal is information regarding the presence or absence of overlap of a plurality of pulse waves obtained by converting a plurality of staircase waves.

In the learning of the learning model 232, the learning device 5 performs machine learning using the signal not including the staircase wave, the staircase wave signal, and the proximity staircase wave signal as teacher data. The learning device 5 associates information indicating that the wave height is zero, the probability that there is no staircase wave is 1, and the other probabilities are 0 with the signal that does not include the staircase wave. The learning device 5 associates the wave height of the stair wave with the information indicating that the probability that there is one stair wave is 1 and the other probability is 0 in the stair wave signal. The learning device 5 associates the proximity staircase wave signal with the sum of the wave heights of the plurality of staircase waves and information indicating that the probability of having a plurality of staircase waves is 1 and the other probabilities are 0. The learning device 5 calculates an error of information by an error function having the wave height and each probability indicated by the information associated with the signal input to the input layer and each probability output from the output layer as variables, and the error. The calculation parameters of each node of the training model 232 are adjusted so that the error is minimized by the back propagation method.

FIG. 16 is a flowchart showing a processing procedure executed by the signal processing device 2 according to the fourth embodiment. The signal processing device 2 receives a signal including a step wave from the radiation detector 1 (S41), and the A / D conversion unit 21 converts the signal into A / D (S42). The A / D conversion unit 21 inputs the A / D converted signal to the processing unit 23.

The processing unit 23 inputs a sequence of signal values constituting the input signal to the learning model 232 (S43). As described above, the learning model 232 performs the calculation of the neural network and outputs the wave height of the staircase wave included in the signal and the probability of the number of staircase waves. The processing unit 23 inputs information indicating the wave height and the number of staircase waves to the counting unit 26. At this time, the processing unit 23 may input the probability output by the learning model 232 to the counting unit 26 as information indicating the number of staircase waves. Further, the processing unit 23 may determine the number of staircase waves based on the probability, and input the information indicating the determined number to the counting unit 26 as the information indicating the number of staircase waves. For example, the processing unit 23 determines that the number having the maximum probability or the number having the probability equal to or higher than a predetermined value is the number of staircase waves.

The counting unit 26 specifies whether or not the number of staircase waves is one based on the input information (S44). When the number of staircase waves is one (S44: YES), the counting unit 26 counts the staircase waves according to the wave height (S45), and ends the process. As a result, the counting unit 26 counts the staircase waves when there is no overlap of the plurality of pulse waves converted from the plurality of staircase waves. When the number of staircase waves is other than one (S44: NO), the counting unit 26 does not count, and the signal processing device 2 ends the processing. As a result, the counting unit 26 does not count the staircase wave when the plurality of pulse waves converted from the plurality of staircase waves overlap. The signal processing device 2 repeatedly executes the processes of S41 to S45.

The signal processing device 2 outputs data showing the relationship between the wave height of the staircase wave and the number of counts counted by the counting unit 26. The analysis unit 3 inputs the data output by the signal processing device 2, and generates a spectrum of the radiation detected by the radiation detector 1 based on the data. The analysis unit 3 may perform elemental analysis of the radiation source based on the spectrum of radiation. Also in this embodiment, the signal processing device 2 can efficiently detect the overlap of a plurality of pulse waves converted from a plurality of staircase waves, and the radiation detection device 10 can improve the accuracy of elemental analysis based on the spectrum. It can be improved further. The signal processing device 2 may be in a form of detecting the overlap of a plurality of pulse waves by also using the feature amount of the signal waveform.

<Embodiment 5>
FIG. 17 is a block diagram showing a functional configuration of the processing unit 23 according to the fifth embodiment. The processing unit 23 has a calculation unit 233 and a memory 234. The calculation unit 233 is configured by using, for example, a CPU, a GPU, or a multi-core CPU. The calculation unit 233 may be configured by using a quantum computer. The memory 234 is a non-volatile memory. The memory 234 stores the computer program 235. The calculation unit 233 executes the processing required for the processing unit 23 according to the computer program 235. The learning model 232 is realized by the arithmetic unit 233 executing information processing according to the computer program 235.

The calculation unit 233 executes information processing according to the computer program 235 to execute the processing required for the processing units 23 in the first to fourth embodiments. In this way, the processing unit 23 in the first to fourth embodiments is realized. The configurations and functions of the radiation detector 1 and the analysis unit 3 are the same as those in the first to fourth embodiments. The configuration and function of the parts other than the processing unit 23 of the signal processing device 2 are the same as those of the first to fourth embodiments. The signal processing device 2 and the radiation detection device 10 perform the same processing as in the first to fourth embodiments. Also in this embodiment, the signal processing device 2 can efficiently detect the overlap of a plurality of pulse waves converted from a plurality of staircase waves, and the radiation detection device 10 can improve the accuracy of elemental analysis based on the spectrum. It can be improved further. A part or all of the part other than the processing unit 23 of the signal processing device 2 may also be realized by using a computer program.

<Embodiment 6>
In the sixth embodiment, the signal processing device 2 effectively detects the overlap of pulse waves without using the learning model. FIG. 18 is a block diagram showing a functional configuration of the radiation detection device 10 according to the sixth embodiment. The configuration and function of the radiation detector 1 and the analysis unit 3 are the same as those in the first embodiment. The signal processing device 2 includes an A / D conversion unit 21, a waveform shaping unit 22, a processing unit 23, a pulse detection unit 24, a wave height measuring unit 25, a counting unit 26, and a storage unit 27. The processing unit 23 does not include the learning model. A storage unit 27 is connected to the processing unit 23. The storage unit 27 is non-volatile. For example, the storage unit 27 is composed of a non-volatile semiconductor memory. The storage unit 27 stores a predetermined coefficient sequence. The configuration of the parts other than the processing unit 23 and the storage unit 27 of the signal processing device 2 is the same as that of the first embodiment.

19 and 20 are graphs showing an example of a coefficient sequence. The horizontal axis in the figure indicates the time in units of clock, and the vertical axis indicates the coefficient value or signal value. The coefficient sequence is a one-dimensional array in which a plurality of coefficients are arranged. In FIG. 19, a signal including a single pulse wave is shown by a broken line, and a coefficient sequence is shown by a solid line. A sequence of signal values constituting a signal including a pulse wave is shown side by side at clock intervals. In addition, a plurality of coefficients included in the coefficient sequence are shown in order at clock intervals. The plurality of coefficients included in the coefficient sequence are for performing a product-sum operation with a sequence of signal values constituting a signal including a pulse wave. Let m be the number of coefficients included in the coefficient sequence. Let the coefficient sequence be a ₁ , a ₂ , ..., _Am , and the sequence of signal values constituting the signal including the pulse wave be x ₁ , x ₂ , .... Product-sum operation is represented by _{_{(a 1 * x 1 + a}} 2 * x 2 + ... + a m * x m + C). C is a constant value. The constant value C is stored in the storage unit 27.

As shown in FIG. 19, the coefficient sequence rises almost at the same time as the pulse wave, increases and decreases in a shorter period than the pulse wave, becomes almost zero near the peak of the pulse wave, and thereafter becomes a negative value. It slowly approaches the value of zero. The coefficient sequence is set so that the result of the product-sum operation with the sequence of signal values constituting the signal including a single pulse wave becomes almost zero. As shown in FIG. 19, the value of the signal including the pulse wave becomes almost zero at the time of the clock 150, and the product-sum operation after that becomes almost zero. Before the time of clock 150, the result of the product-sum calculation of the signal value before the peak of the pulse wave and the positive coefficient is a positive value, and the product-sum calculation of the signal value after the peak of the pulse wave and the negative coefficient is performed. The change in the coefficient value is defined so that the result is a negative value and the total is almost zero. Therefore, the result of the product-sum calculation of the sequence of signal values and the coefficient sequence constituting the signal including a single pulse wave becomes almost zero. In this way, the coefficient sequence is defined so that the result of the product-sum operation with the signal containing a single pulse wave becomes almost zero.

In FIG. 20, the signal including the superimposed wave is shown by a broken line, and the coefficient sequence is shown by a solid line. The coefficient sequences shown in FIGS. 19 and 20 are the same. As shown in FIG. 20, since the time width of the superimposed wave is larger than that of the single pulse wave, the signal value of the signal including the superimposed wave keeps a positive value even after the time of the clock 150. Further, the coefficient sequence has a negative value after the time of the clock 150. Therefore, the result of the product-sum calculation of the string of signal values and the coefficient string constituting the signal including the superimposed wave becomes a negative value after the time of the clock 150. Before the time of clock 150, the result of the product-sum operation is almost zero, as in the case of a single pulse wave. Therefore, the result of the product-sum calculation of the string of signal values and the sequence of coefficients constituting the signal including the superimposed wave is a negative value. In this way, the coefficient sequence is defined so that the result of the product-sum operation with the signal including the superimposed wave becomes a negative value. That is, with a predetermined value less than zero as the reference value, the result of the product-sum calculation with the signal containing a single pulse wave becomes larger than the reference value in the coefficient sequence, and the product-sum calculation with the signal containing the superimposed wave is performed. The result is set to be smaller than the reference value.

FIG. 21 is a flowchart showing a processing procedure executed by the signal processing device 2 according to the sixth embodiment. The signal processing device 2 executes the processing of S51 to S55 similar to the processing of S21 to S25 in the first embodiment. When the pulse wave included in the signal is detected (S25: YES), the pulse detection unit 24 inputs information indicating that the pulse wave is detected to the processing unit 23, and the processing unit 23 inputs the information. The product-sum calculation of the sequence of signal values constituting the signal and the sequence of coefficients stored in the storage unit 27 is performed (S56).

The processing unit 23 determines whether or not a plurality of pulse waves overlap in the input signal according to the result of the product-sum calculation (S57). In S57, the processing unit 23 determines that there is overlap of a plurality of pulse waves when the result of the product-sum calculation is less than the reference value of less than zero, and when the result of the product-sum calculation is greater than or equal to the reference value. , It is determined that there is no overlap of multiple pulse waves. The reference value is stored in advance in the processing unit 23 or the storage unit 27. The processing unit 23 inputs information indicating the presence or absence of overlapping of the plurality of pulse waves to the wave height measuring unit 25. When there is no overlap of a plurality of pulse waves (S57: NO), the wave height measuring unit 25 measures the wave height of the pulse wave included in the input signal (S58), and inputs the measured wave height to the counting unit 26. .. The counting unit 26 counts pulse waves according to the wave height input from the wave height measuring unit 25 (S59), and ends the process.

When there is an overlap of a plurality of pulse waves (S57: YES), the wave height measuring unit 25 does not measure the wave height for the overlapping plurality of pulse waves, and the signal processing device 2 ends the processing. As a result, the counting unit 26 does not count the plurality of overlapping pulse waves. The signal processing device 2 repeatedly executes the processes of S51 to S59 individually. The signal processing device 2 outputs data showing the relationship between the height of the pulse wave and the number of counts. The analysis unit 3 inputs the data output by the signal processing device 2, and generates a spectrum of the radiation detected by the radiation detector 1 based on the data. The analysis unit 3 may perform elemental analysis of the radiation source based on the spectrum of radiation.

In the coefficient sequence, the result of the product-sum calculation with the sequence of signal values constituting the signal including a single pulse wave is smaller than the reference value, and the product with the sequence of signal values constituting the signal including the superimposed wave. It may be set so that the result of the sum operation is larger than the reference value. For example, in the coefficient sequence, the result of the product-sum calculation of the signal value sequence and the coefficient sequence constituting the signal containing a single pulse wave becomes almost zero, and the result of the product-sum calculation with the signal including the superimposed wave is almost zero. It may be set to be a positive value. In this embodiment, in S57, the processing unit 23 determines that there is an overlap of a plurality of pulse waves when the result of the product-sum calculation exceeds the reference value, using a predetermined value exceeding zero as a reference value. When the result of the product-sum operation is equal to or less than the reference value, it is determined that the plurality of pulse waves do not overlap.

As described in detail above, in the sixth embodiment, the signal processing device 2 performs a product-sum calculation of a string of signal values constituting a signal including a pulse wave and a predetermined coefficient string, and is based on the result of the product-sum calculation. Then, the overlap of pulse waves is detected. Also in the sixth embodiment, the signal processing device 2 can effectively detect the overlap of pulse waves, and the radiation detection device 10 can further improve the accuracy of elemental analysis based on the spectrum. Since the determination is performed by the product-sum calculation without using the learning model, the signal processing device 2 can execute the processing at a higher speed.

The signal processing device 2 may be in a form in which both the method using the learning model 232 shown in the first or second embodiment and the method using the result of the product-sum calculation shown in the sixth embodiment are executed. For example, the signal processing device 2 determines whether or not there is overlap of pulse waves in both the method using the learning model 232 and the method using the result of the product-sum calculation, and the overlap of pulse waves is performed by either method. When it is determined that there is, it is determined that the overlap of pulse waves has occurred. In the first to sixth embodiments, the signal processing device 2 includes the counting unit 26, but the counting unit 26 may be provided outside the signal processing device 2. Further, in the first to sixth embodiments, an example in which the radiation detection element 11 is a semiconductor radiation detection element is shown, but the radiation detector 1 may have a form in which the radiation detection element 11 other than the semiconductor radiation detection element is used. Good. For example, the radiation detector 1 may be a scintillation detector.

The present invention is not limited to the contents of the above-described embodiment, and various modifications can be made within the scope of the claims. That is, an embodiment obtained by combining technical means appropriately modified within the scope of the claims is also included in the technical scope of the present invention.

1 Radiation detector 10 Radiation detection device 11 Radiation detection element 2 Signal processing device 21 A / D conversion unit 22 Waveform shaping unit 23 Processing unit 232 Learning model 235 Computer program 25 Wave height measurement unit 26 Counting unit 27 Storage unit 3 Analysis unit 5 Learning apparatus

Claims

In a signal processing method that counts a staircase wave corresponding to radiation detection or a pulse wave obtained by converting the staircase wave by wave height.
When at least one of the sequence of signal values constituting the signal including the staircase wave and the sequence of signal values constituting the signal including the pulse wave is input, information regarding the presence or absence of overlapping of the plurality of pulse waves is output. At least one of the sequence of signal values constituting the signal including the staircase wave and the sequence of signal values constituting the signal including the pulse wave are input to the learning model to be performed.
A signal processing method comprising counting the staircase wave or the pulse wave when there is no overlap of a plurality of the pulse waves according to the information output by the learning model.
The signal processing method according to claim 1, wherein the step wave or the pulse wave is not counted when a plurality of the pulse waves overlap according to the information output by the learning model.
The learning model is a learning model that outputs the information when a sequence of signal values constituting a signal including a wave obtained by shaping the pulse wave is further input.
The signal processing method according to claim 1 or 2, wherein a sequence of signal values constituting the signal is further input to the learning model.
The learning model is a learning model that outputs the information when at least one feature amount of the staircase wave and the pulse wave is further input.
The signal processing method according to any one of claims 1 to 3, wherein the feature amount is input to the learning model.
Each of them consists of a single pulse wave, and generates a plurality of pulse wave signals in which at least one of the height, rise and roundness of the single pulse wave is random.
Each of the plurality of pulse waves is composed of a superposed wave in which a plurality of pulse waves are overlapped, and a plurality of superposed wave signals in which at least one of the interval, wave height, rising edge, and rounding of the plurality of pulse waves is random are generated.
When the plurality of pulse wave signals and the plurality of superimposed wave signals are used as teacher data and a sequence of signal values constituting a signal including an arbitrary pulse wave is input, information regarding the presence or absence of overlap of the plurality of pulse waves is output. A training model generation method characterized by generating a training model.
Each staircase wave consists of a single staircase wave, generating multiple staircase wave signals in which at least one of the height, rise and roundness of the single staircase wave is random.
Each consists of a plurality of staircase waves, and a plurality of pulse waves obtained by converting the plurality of staircase waves overlap each other, and at least one of the intervals, wave heights, rises, and rounds of the plurality of staircase waves is random. Generates a proximity staircase wave signal,
Overlapping of a plurality of pulse waves obtained by converting the plurality of staircase waves when a sequence of signal values constituting a signal including an arbitrary staircase wave is input using the plurality of staircase wave signals and the proximity staircase wave signal as training data. A learning model generation method characterized by generating a learning model that outputs information regarding the presence or absence of.
In a signal processing method in which a staircase wave corresponding to radiation detection is converted into a pulse wave and the pulse wave is counted according to the height of the pulse wave.
A product-sum calculation is performed on a sequence of signal values constituting the signal including the pulse wave and a predetermined sequence of coefficients.
The predetermined coefficient sequence includes a sequence of signal values constituting a signal including a plurality of overlapping pulse waves, a result of a product-sum calculation between the predetermined coefficient sequences, and a signal including a single pulse wave. It is predetermined so that the result of the product-sum calculation between the constituent signal value sequence and the predetermined coefficient sequence differs depending on whether it is large or small with respect to the predetermined value.
Depending on the result of the product-sum calculation, it is determined whether or not the plurality of pulse waves overlap.
A signal processing method comprising counting the non-overlapping pulse waves.
In a signal processing device that processes a signal including staircase waves in response to radiation detection
When at least one of the sequence of signal values constituting the signal including the staircase wave and the sequence of signal values constituting the signal including the pulse wave obtained by converting the staircase wave is input, a plurality of the pulse waves are overlapped. A signal processing device characterized by having a learning model that outputs information regarding the presence or absence of a signal.
A radiation detector that outputs staircase waves according to the energy of radiation when detecting radiation,
A conversion unit that converts the staircase wave into a pulse wave,
When at least one of the signal value sequence constituting the signal including the staircase wave and the signal value sequence constituting the signal including the pulse wave obtained by converting the staircase wave is input, the presence or absence of overlap of the plurality of pulse waves. A learning model that outputs information about
A counting unit that counts the staircase wave or the pulse wave when there is no overlap of the plurality of pulse waves according to the information output by the learning model.
A radiation detection device including a spectrum generation unit that generates a spectrum of radiation according to the wave height of the staircase wave or the pulse wave and the number of counts.
When at least one of a sequence of signal values constituting a signal including a staircase wave according to radiation detection and a sequence of signal values constituting a signal including a pulse wave obtained by converting the staircase wave is input, a plurality of the pulses are performed. At least one of the sequence of signal values constituting the signal including the staircase wave and the sequence of signal values constituting the signal including the pulse wave is input to the learning model that outputs information regarding the presence or absence of overlap of the waves. A computer program characterized by having a computer execute a process that outputs information.