RU2699751C1 - Method of sorting objects by their colour characteristics - Google Patents

Abstract

FIELD: transportation.

SUBSTANCE: invention relates to methods of sorting different objects into a limited number of classes in accordance with their colour characteristics. Main application of invention is found in processes of sorting precious stones by their colour, first of all, diamond material. Method includes stages, at which transportation of a sequence of objects is carried out by a special device through a surveillance zone, lighting of each object passing the observation zone individually, recording reflected and / or transmitted electromagnetic radiation in the form of electric signals, conversion and processing of obtained signals into digital form, analysis of said information in computing device for compliance with predetermined system parameters determining object belonging to a certain class, making a decision and issuing a command to the actuator about movement of the object to the corresponding hopper – accumulator of objects of one of the classes according to the selected classification. Distinctive features of the method are preliminary obtaining statistically reliable classification model or rule for decision on belonging of object to this or that class, analysis of features of classes by their significance and effect on classifying rule and use of methodology of the theory of multidimensional approximation and interpolation.

EFFECT: provides optimum combination of high rate of execution of program implementation of algorithms for solving classification problem with simultaneous reduction of percentage of unclassified objects in their total number due to lower probability of identification errors compared to existing methods.

1 cl, 1 dwg

Description

Technical field

The invention relates to methods for sorting various objects into a limited number of classes in accordance with their color. The invention finds its main application for the processes of sorting precious stones, primarily diamonds, by their color and for devices carrying out these processes.

State of the art

Known methods for sorting objects by color and other optical characteristics are based on the following general principles: arranging transportation of a sequence of objects with a special device through the observation zone, lighting individually each object passing through the observation zone, registering electromagnetic radiation reflected or transmitted through the object in the form of electrical signals, conversion and processing the received signals in digital form, the analysis of this information in a computing device for the subject of correspondence to a predetermined system of parameters that determine whether an object belongs to a certain class, making a decision and issuing a command to the actuator to move the object to the appropriate bunker - a drive of objects of one of the classes according to the selected classification.

So, there is a known method of sorting objects according to their optical properties (US patent 4,186,838, IPC В07С 5/342), in which sorting can be performed according to the characteristics of the illuminator of the selected spectral range reflected or transmitted through the object, that is, by color characteristics. Registration of radiation is carried out in the process of successive passage by each object in the phase of free fall through the sphere integrating radiation of the object. The photodetector devices, at least two photoelectronic multipliers (PMTs) with the appropriate optical filters that provide the necessary color-dependent signals to distinguish the sorted objects by color, were selected as devices for registration. The determination of sets (criteria for assigning to the color class) of color-dependent signal levels during the implementation of the method was carried out on exemplary collections of objects of a previously known color classification. The disadvantages of the method are manifested when it is necessary to sort into a large number of classes, rather than classes of three primary colors, for example. In this case, the sorting criteria based on the threshold (or interval) characteristics of the signals give significant errors due to the instability in time of the parameters of the electronic recording paths and lighting conditions of objects in the integrating sphere. The method also has limited possibilities for taking into account the size (physical size) of sorted objects, on which the signal levels of photodetectors depend.

A device is known (US patent 4,951,825, IPC B07C 5/342) for automatic sorting by predefined classes of granular material, including rough diamonds. In this device, the means of transporting objects through the lighting - observation zone is a rotary conveyor with evacuated seats for sorting objects, and radiation from the object is recorded by photodetector devices in different spectral ranges, similar to the patent source described above. The classification decision on assigning an object to a particular class is also made by the computing device on the basis of a system of predefined intervals (levels) of photodetector signals in the corresponding spectral range, together with an analysis of the error in determining the boundaries of these intervals or by the criterion of the minimum amount of deviations when the object is assigned to a predefined class. A system or a set of combinations of signal levels in the spectral channels of a device can be defined analytically or obtained empirically based on changes in representative collections of color objects classified in advance. To take into account the effect of the size of the sorting objects, additional LED sources of infrared radiation are used, while the errors in the instability of the parameters of the measuring paths in the spectral ranges are compensated by periodic calibration according to white and black standards. This patent source describes a method for sorting crystalline objects according to a number of optical characteristics and, in particular, by color. This method, although to a lesser extent, has the disadvantages already mentioned above. This is a practically limited number of sorting classes (4 and one class for unidentified objects), a fundamentally unavoidable error due to bright glare or dark defects on the surface of objects, a limited field of view of a surface reflecting (transmitting) radiation, a complex design and high requirements for equipment for implementation way.

There is a method of color classification of objects, which is one of the prototypes of the present method (RF patent No. 2468345, IPC ⁵ G01N 21/85, G01J 3/51). The method provides the possibility of simultaneous classification of several simultaneously both stationary and moving objects at their arbitrary location in the analysis zone, adaptation of measurements to changing conditions of identification of objects. The color image of an object obtained using multi-element matrix radiation detectors based on devices with a charge transfer effect (a camera with a CCD matrix) is converted into three two-dimensional arrays of integers, each of which contains information about the spatial distribution in the image of one of three colors - red, green and blue. These two-dimensional arrays are converted to the HLS color space, the value of the color coordinate is compared with the a priori known values of the so-called characteristic zones. For elements that do not coincide in the color coordinate, the color coordinates are aligned according to the ratio given in the source. This action is, in fact, a means of reducing the number of decomposition elements of the image of the analyzed object and smoothing the distribution characteristics of the color coordinates of the object in order to reduce the amount of processed information and simplify the classification criterion. The method and device for its implementation are focused primarily on the tasks of sorting objects of strong color contrast with a dominant predominance of one color. To sort objects of weakly expressed color shades, this method will have limited use.

It is known (RF Patent 2424859, IPC V07C 5/342) another device for sorting crystals of diamonds, which has a unit for piece feeding objects and a horizontal conveying disk. In this installation, there are lighting and observation means, a control computing device and a sorting unit for placing crystals in bins in accordance with classification criteria. The same patent source describes a method for color sorting (classification) of objects, which is also a prototype of the present invention. As a source of information, a digital image (pixel representation) of the sorting object in the format of three primary colors of the filter system of a multi-element radiation detector of a digital camera is used. The algorithm of a computer program that processes digital image arrays and develops a classification solution includes pixel-by-pixel analysis of the resulting image with identification of the fact that this decomposition element belongs to the background or to the object being analyzed, an object defect, or a diffraction glare on the crystal faces. The latter are excluded from the final processing - obtaining the integral color characteristics of the object for making a classification decision. The implementation of the method requires the receipt of a large amount of a priori information necessary for processing pixel arrays, while the analysis itself associated with the evaluation of each individual decomposition element in the pixel array is quite time-consuming and limits the production of the productivity required for processing rough diamonds.

Disclosure of invention

The purpose of the proposed technical solution is a method that provides a combination of high performance and efficiency of the processes of sorting objects by their color characteristics.

The technical result of the invention is expressed in providing a high performance method for sorting objects by color classes while reducing the percentage of classification errors expressed in the number of objects to be re-sorted or manually sorted. Another technical result of the invention is the provision of flexibility in the preparation of equipment for sorting, to work, increasing the efficiency of using equipment for sorting and simplifying the configuration process.

The technical result of the invention is achieved by the fact that in the proposed method for sorting objects in accordance with their color characteristics, which includes the piecewise supply of objects to the lighting-observation zone in the free fall phase, receiving images of objects in radiation from illuminators with two electronic image recording devices, which are, for example, two-dimensional matrix photodetectors, converting this image into electronic form - data file of one of the standard pixel image formats digital camera, transferring this file to a computing device, processing this file in order to assign the object in the image to one of the predefined classes, forming a command for the executive sorting device and placing the object in its corresponding storage hopper form, additionally, on the basis of a representative sample of classified (sorted) objects for each color class; image of each object of this sample is obtained in the conditions identical to the subsequent sorting processing the image data file with the conversion of each decomposition element into a vector representation in color space, obtaining histograms of the distribution of the primary colors of the elements of the decomposition of the sample by sorting class, clarifying the most informative signs of the vector representation in order to reduce the dimension of the attribute space, checking the significance of each sign with one of the known criteria and construction of a numerical model of a classifying solution based on the methodology of multidimensional approximation and interpolation, which allows you to unambiguously attribute the set of vector representations in the attribute space of a given sortable object to a predefined color class. In other words, the method includes two main stages: obtaining a statistically valid model (classifying rules) at the machine learning stage using a representative collection of objects assigned in advance to a particular class according to color characteristics, and actually subsequent sorting of an array of objects using the one obtained at the first stage models. The computational procedures of the method are implemented in the algorithms of programs and software libraries operating in a computing device that processes digital arrays of images of objects. The mathematical apparatus of these algorithms is built on the basis of the theory of random functions using the machine learning method based on the algorithm of multidimensional interpolation and approximation of random functions (T.A. Shirabakina, Yu.N. Bakhvalov, A.N. Zuev, Izv. VUZov. "Instrument Engineering" , Volume 48, 2005, No. 2, pp. 5-8; Yu.N. Bakhvalov, L. L. Malygin, P. S. Cherkass, Machine Learning Method Based on the Algorithm of Multidimensional Interpolation and Approximation of Random Functions.Vestnik of Cherepovets State University , Volume 2, 2012, No. 2, pp. 7-10). For the classification problem, which is the basis of the proposed method, each element of the training sample or each object of the representative collection at the training stage is perceived by the computing device as a set or vector of features obtained from its image by processing a pixel data array from a digital camera. In the general case, these can be the coordinates of the image decomposition element in one of the color representations (spaces), their selective distributions and statistical characteristics, the presence or absence of certain primary colors or their combinations in the class, various threshold features, and the like. Thus, at the stage of obtaining the classifying model, the entire set of features or the input feature vector of any element of the representative collection (training set) is compared with the output vector having the only feature - a predefined class of objects, in accordance with the present invention - this is a class according to color characteristics. The classifying model obtained at the training stage is optimized by eliminating the insignificant elements of the training set to reduce the execution time of computational procedures during sorting by reducing the number of terms of the function of the decision rule.

As a result of the first stage of the proposed method, the classification model or classifier is stored in the data and settings file and the functioning of the computer program of the optoelectronic classification system and is used to sort objects by color classes from the total unsorted mass. At this second stage, the set of features obtained by processing the pixel arrays of the image of the object or, in other words, the input feature vector of the sorted element, already determines the value of the function - the decisive rule of the classifier. In turn, the value of the function is compared with the predefined values for the sorting classes obtained at the first stage of the proposed method, that is, at the training stage with a representative collection of elements. The result of the comparison is a classification decision on assigning this element to a predefined class according to color characteristics.

A brief description of the notation of figure 1.

The figure 1 shows a block diagram of a device explaining the implementation of the proposed method for sorting objects according to their color characteristics.

1 - device piece feed objects;

2 - sensor entrance of the object into the lighting zone - observation;

3 - devices for recording images of objects (digital high-speed cameras);

4 - diffuser lighting objects;

5 - LED illuminators of objects;

6 - sensor input of the object into the sorting zone;

7 - receiving unit - buffer drum - intermediate drive with cells;

8 - bunker - drive for classes of objects;

9 - pneumatic cut-offs of objects for moving into a hopper - drive corresponding to their color.

Preferred Embodiment

The method of sorting objects according to their color characteristics consists in the unit-by-unit supply of objects to the lighting zone — the observation that the object passes in the free fall phase, simultaneously turning on the illuminators and both cameras, imaging objects in the form of digital data arrays, transferring data arrays to the computing device and processing them for assignment to predefined classes by color, passage of the sorting zone with intermediate drives with cells and distribution of the object to storage bins elyam according to the decision of the classifier software.

A device that implements the proposed method consists of a feed unit 1, which includes a vacuum rotary feed conveyor of objects and two vibratory feeders, forming a sequence of objects, which ensures a uniform, adjustable and reproducible flow of classified objects into the illumination-observation zone along the free fall path. At the entrance of the object to the lighting-observation zone, the first optoelectronic sensor for entering the object into the lighting zone 2, the observation, starts the synchronized process of shooting the object with each digital high-speed camera 3. The object is illuminated in the diffuser by 4 multi-element LED illuminators 5 objects of white radiation. The synchronization device (not shown in Fig. 1) simultaneously launches LED illuminators and cameras, forming pixel image files of the sorting object in the lighting - observation zone. Image files from two cameras are transmitted to a computing device — a personal computer (not shown in FIG. 1), where digital image processing and computational procedures of the proposed method are implemented.

Further, the object is identified by a specific color class or a decision is made about the impossibility of classifying the color of the object. Having passed the lighting - observation zone, the object in front of the cut-off zone is registered by the second optoelectronic sensor 6 of the object entering the sorting zone 7. The cut-off zone consists of a rotating buffer drum-intermediate drive 7 with cells, a number of pneumatic cutoffs 9 and storage bins 8 for sorted by shape objects. After passing the second optoelectronic sensor 6 of the entrance of the object into the sorting zone, the object enters one of the cells of the buffer drum 7, the angular position of which at that moment is fixed by the computing device and is used for subsequent movement of the object using the cutter 9 into the storage hopper 8 of the corresponding color class.

The set of distinguishing features of the described method for sorting by color, including the preliminary obtaining of a statistically reliable classifying model, the analysis of class attributes by their significance and influence on the classification rule, and, in general, using the multidimensional approximation and interpolation methodology, provides the optimal combination of high speed software implementation of problem solving algorithms classification with a simultaneous decrease in the share of unclassified objects in their total number due to enshey probability of identification errors as compared to the prior art. Classification models, in the presence of representative collections of objects, are easily created, saved as specialized components of these computer programs and provide quick reconfiguration of the sorting equipment during processing of large volumes of rough diamonds according to various criteria of color characteristics.

Industrial applicability and practical examples of the method

An example of the practical implementation of the proposed method is the installation of sorting rough diamonds by color ASA-1C. These plants are used for sorting rough diamonds in industrial conditions. Tests of the installation when sorting into 13 classes according to the color of the raw materials showed a productivity of up to 10 crystals per second. The share of unidentified crystals for re-sorting was less than 3% by weight relative to the initial, unsorted array.

Claims (2)

1. A method for sorting crystalline objects according to their color characteristics, including the piecewise supply of objects to the lighting zone — observation, obtaining an image of the object in reflected lighting by electronic image recording devices, converting this image into electronic form — a data file of one of the standard pixel formats of a digital camera, transferring this file to a computing device, processing this file in order to assign an object in an image to one of the predefined classes the colors and placement of this object in a storage bin corresponding to the class, characterized in that first, on the basis of a representative sample of classified (sorted) objects for each color class, an image is obtained of each object of this sample in conditions identical to the subsequent sorting, and the image data file is processed with transforming each decomposition element into a vector representation in the color space, obtaining histograms of the distribution of the primary colors of the decomposition elements selection of the sorting classes, refinement of the most informative features of the vector representation in order to reduce the dimension of the feature space, check the significance of each feature by one of the known criteria, and build a numerical model of the classification solution based on the multidimensional approximation and interpolation methodology, which allows you to unambiguously relate the set of vector representations in the feature the image space of this sortable object to a predefined color class. 2. The method according to claim 1, in which a representative sample of classified (sorted) crystalline objects of each class, selected by the expert assessment method, can be part of the whole set of objects assigned for sorting.

Images