JP7795730B2 - Data processing device, data processing program, method for determining whether a group of substrates is similar or not, system for evaluating a group of substrates, and method for evaluating a group of substrates - Google Patents

Description

The present invention relates to a data processing device, a data processing program, a method for determining whether a group of substrates is similar, a system for evaluating a group of substrates, and a method for evaluating a group of substrates.

The electronic circuit boards that make up electronic devices contain metals with high economic value, so in recent years, discarded electronic circuit boards (hereinafter referred to as "waste electronic circuit boards") have been recycled to extract and refine major metals from these waste electronic circuit boards.

When recycling waste electronic boards, it is extremely important to understand in advance the amount of metal resources contained in those boards. For this reason, it has been proposed to estimate the amount of metal resources contained in waste electronic boards to be recycled by utilizing past data obtained during previous recycling and determining whether image data of the waste electronic boards is similar to the past data (see, for example, Patent Document 1).

Japanese Patent Application Laid-Open No. 2020-190881

When recycling waste electronic circuit boards, it is common to process a collection of multiple waste electronic circuit boards (a group of boards) at once. Furthermore, a collection of waste electronic circuit boards can contain multiple types of boards, such as a group of expansion board boards or a group of CPU boards. Even in such cases, being able to efficiently and reliably determine whether the image data of a collection of waste electronic circuit boards is similar to past data is extremely important for recycling waste electronic circuit boards to become a viable business.

The present invention aims to provide technology that contributes to the establishment of a recycling business for waste electronic circuit boards by enabling highly reliable and efficient determination of the similarity between image data of a collection of waste electronic circuit boards and image data constituting past data.

A first aspect of the present invention is a method for manufacturing a semiconductor device comprising:
a data acquisition unit that acquires image data of the collection of waste electronic substrates;
a feature extraction unit that extracts image complexity from the image data and sets it as a feature of the image data;
a similarity determination unit that determines whether one image data and another image data acquired by the data acquisition unit are similar or not by using the feature amount as an index;
The data processing device includes:

A second aspect of the present invention is
In the data processing device according to the first aspect, the feature extraction unit is configured to extract the complexity of the image by analyzing the texture of the image.

A third aspect of the present invention is
In the data processing device according to the first or second aspect, the feature extraction unit uses a calculation result of a coefficient of variation of the complexity as the feature.

A fourth aspect of the present invention is
a region classification unit that classifies the image data into a substrate region and a background region based on the feature extracted by the feature extraction unit,
In the data processing device according to any one of the first to third aspects, the similarity determination unit is configured to determine whether the substrate region is similar or not.

A fifth aspect of the present invention is
a different feature extraction unit that extracts a different feature from the feature from the image data;
The data processing device according to any one of the first to fourth aspects, wherein the similarity determination unit is configured to perform similarity determination using, as an index, the feature extracted by the feature extraction unit and another feature extracted by the another feature extraction unit.

A sixth aspect of the present invention is
In the data processing device according to the fifth aspect, the similarity determination unit is configured to weight each feature amount used in the similarity determination when making the similarity determination.

A seventh aspect of the present invention is a method for manufacturing a semiconductor device comprising:
an imaging unit that captures an image of the collection of waste electronic substrates to obtain the image data;
an illumination unit that irradiates light onto the collection of waste electronic substrates that are to be imaged by the imaging unit,
In the data processing device according to any one of the first to sixth aspects, the illumination unit is configured to irradiate the collection of waste electronic circuit boards with indirect light while blocking external light.

An eighth aspect of the present invention is a method for manufacturing a semiconductor device comprising:
Computer,
a data acquisition unit that acquires image data of the collection of waste electronic substrates;
a feature extraction unit that extracts image complexity from the image data and sets it as a feature of the image data;
a similarity determination unit that determines whether one image data and another image data acquired by the data acquisition unit are similar or not by using the feature amount as an index;
It is a data processing program that functions as a

A ninth aspect of the present invention is a method for manufacturing a semiconductor device comprising:
a data acquisition step of acquiring image data of the collection of waste electronic substrates;
a feature extraction step of extracting image complexity from the image data and setting the image complexity as a feature of the image data;
a similarity determination step of determining whether one image data and another image data acquired in the data acquisition step are similar to each other by using the feature amount as an index;
The method for determining whether a group of substrates is similar or not comprises the steps of:

A tenth aspect of the present invention is a method for manufacturing a semiconductor device comprising:
an image capturing unit that captures an image of the collection of waste electronic substrates;
a resource amount estimation unit that analyzes the aggregate and estimates the amount of metal resources contained in the aggregate;
a database unit that stores and accumulates the image capturing results from the image capturing unit and the estimation results from the resource amount estimation unit in association with each other;
a data acquisition unit that acquires image data of the collection from the image capturing unit or the database unit;
a feature extraction unit that extracts image complexity from the image data and sets it as a feature of the image data;
a similarity determination unit that uses the feature amount as an index to determine whether one piece of image data acquired by the data acquisition unit regarding the analysis target in the resource amount estimation unit is similar to another piece of image data acquired by the data acquisition unit from the database unit for comparison with the one piece of image data;
an evaluation information output unit that extracts from the database information corresponding to image data that has been determined to have a high similarity to the one image data by the similarity determination unit, and outputs the extracted information as evaluation information about the collection that is the analysis target of the resource amount estimation unit;
The substrate group evaluation system includes:

An eleventh aspect of the present invention is a method for manufacturing a semiconductor device comprising:
an image capturing step of capturing an image of the collection of waste electronic substrates with an image capturing unit;
a resource amount estimation step of analyzing the aggregate to estimate the amount of metal resources contained in the aggregate;
a data storage step of storing and accumulating the image capturing result in the image capturing step and the estimation result in the resource amount estimation step in a database unit in association with each other;
a data acquisition step of acquiring image data of the collection from the image capturing unit or the database unit;
a feature extraction step of extracting image complexity from the image data and setting the image complexity as a feature of the image data;
a similarity determination step of determining whether one image data item acquired in the data acquisition step for the analysis target in the resource amount estimation step is similar to another image data item acquired from the database unit in the data acquisition step for comparison with the one image data item, using the feature amount as an index;
an evaluation information output step of retrieving, from the database unit, correspondence information of image data determined to have a high similarity to the one image data in the similarity determination step, and outputting the information as evaluation information about the collection to be analyzed in the resource amount estimation step;
The substrate group evaluation method includes the steps of:

This invention makes it possible to efficiently and reliably determine whether image data of a collection of discarded electronic circuit boards is similar to other image data, thereby contributing to the establishment of a recycling business for discarded electronic circuit boards.

FIG. 10 is a flow chart showing an example of a procedure for an analysis process in a recycling process of waste electronic substrates. 1 is a schematic diagram showing a system configuration of a substrate group evaluation system according to the present invention; 1 is a block diagram showing an example of a functional configuration of a substrate group evaluation system according to the present invention; 1 is a flowchart showing an example of the procedure of a method for determining whether a group of substrates is similar or not according to the present invention. 1 is an explanatory diagram showing a specific example of a board image to be handled by the board group similarity determination method according to the present invention; 1 is an explanatory diagram showing an example of a board region and a background region in a board image handled by the board group similarity determination method according to the present invention; FIG. 10 is an explanatory diagram (part 1) showing a specific example of the procedure for extracting a fourth characteristic amount in the substrate group similarity determination method according to the present invention. FIG. 10 is an explanatory diagram (part 2) showing a specific example of the procedure for extracting the fourth characteristic amount in the method for determining whether a group of substrates is similar or not according to the present invention. 10A and 10B are explanatory diagrams showing a specific example of a board image and a complexity heat map used in the board group similarity determination method according to the present invention;

The following describes the data processing device, data processing program, board group similarity determination method, board group evaluation system, and board group evaluation method according to the present invention, with reference to the drawings.

<Recycling process overview>
First, an outline of the recycling process for waste electronic substrates will be explained.

Recycling of waste electronic substrates can involve extracting and refining major metals from the waste electronic substrates at a non-ferrous metal smelter, and also includes analytical processing to determine the amount of metal contained in the waste electronic substrates.
The waste electronic circuit boards to be processed are printed circuit boards with electronic components mounted on them, and contain metal resources such as gold, silver, copper, platinum, palladium, etc. Specific examples include expansion board boards, motherboard boards, CPU boards, memory boards, network boards, server boards, finger boards, HDD boards, and mobile phone boards.
In the recycling process, waste electronic circuit boards are grouped together in a certain quantity of the same type (for example, boards with the same or similar configuration of electronic components mounted), and this group is treated as a single unit (hereinafter, a single unit is referred to as a "lot"). Therefore, different types of waste electronic circuit boards are treated as separate lots.

The analysis process in the recycling process is carried out, for example, according to the procedure described below.
FIG. 1 is a flow chart showing an example of the procedure of an analysis process in the recycling process of waste electronic substrates.
When performing the analysis process, first, an image of the collection of waste electronic boards in the target lot is taken to obtain an image of the collection (hereinafter also referred to as a "board image") (step 101, hereinafter, step is abbreviated as "S"). Next, the waste electronic boards in the target lot are crushed, and a portion of the crushed boards is used as a sample (S102), and the composition of the sample is analyzed (S103). This allows the metal content ratio of the waste electronic boards to be obtained. The metal content ratio refers to the proportion of major metals estimated to be contained in a predetermined amount (e.g., 1 ton) of waste electronic boards. In other words, the amount of metal resources contained in the collection of waste electronic boards is estimated by analyzing the composition of the sample.
When determining an appropriate analytical procedure for waste electronic substrates, it is desirable to be able to grasp the metal content ratio of the waste electronic substrates in advance. Therefore, as shown in the dashed box in the figure, when selecting an analytical procedure (S106), by using a substrate image of the target waste electronic substrate and referring to information on similar substrates that have been handled in the past (S104), it becomes possible to more accurately estimate the metal content ratio (S105).

The analysis process described above is carried out for each lot, which is a collection of waste electronic boards. In other words, a collection of multiple waste electronic boards (a group of boards) that make up one lot is processed all at once. This is therefore extremely effective in achieving efficient processing.
However, in the above-mentioned analysis process, if the similarity determination of the circuit board images etc. is made by the subjective judgment of the operator, it becomes a process that depends on the individual, which is not desirable from the viewpoint of ensuring a high reliability of the process. Furthermore, if the amount of data to be compared is enormous, it will be a heavy burden on the operator and will take a lot of time, which is not desirable from the viewpoint of realizing efficient processing.
For this reason, in this embodiment, an algorithm is used to automate the determination of similarity of a collection of waste electronic substrates in the analysis process of the recycling process, thereby ensuring high reliability of the estimation results for the amount of metal resources contained in the collection of waste electronic substrates and making it possible to efficiently carry out the processing required for this.

<System configuration example>
Next, an example of the configuration of a system used in this embodiment to enable automated determination of the similarity of a collection of waste electronic circuit boards will be described. The system given as an example here corresponds to a specific example of the "circuit board group evaluation system" according to the present invention.
Fig. 2 is a schematic diagram showing the system configuration (hardware configuration) of the substrate group evaluation system exemplified in this embodiment. Fig. 3 is a block diagram showing an example of the functional configuration (software configuration) of the substrate group evaluation system shown in Fig. 2.

(Overall structure)
2 , the system exemplified in this embodiment includes a belt conveyor 2 that receives and transports a collection of waste electronic circuit boards (hereinafter also referred to as a “board group”) 1 that constitutes a target lot, a camera 3 that serves as an imaging unit that photographs the board group 1 flowing on the belt conveyor 2 to obtain board images, a resource amount estimation unit 4 that performs a component analysis of the board group 1 received from the belt conveyor 2 to determine the metal content ratio and thereby estimates the amount of metal resources contained in the board group 1, a server device 5 and a data processing device 6 that are configured as computers, and an illumination unit 7 that irradiates light onto the board group 1 that is the subject of imaging by the camera 3. Note that the server device 5 and the data processing device 6 may be configured separately from each other, or may be configured as an integrated unit.

Of these, the belt conveyor 2, the camera 3 and the resource amount estimation unit 4 may be configured using known technology, and detailed description thereof will be omitted here.
Furthermore, in this embodiment, a system configuration in which the camera 3 photographs the group of boards 1 on the belt conveyor 2 is exemplified, but this is not limiting and other system configurations may be used. For example, the camera 3 may photograph the group of boards 1 before being transported by the belt conveyor 2, or may photograph the group of boards 1 being transported by a transport tool other than the belt conveyor 2, or may photograph the group of boards 1 being placed at a predetermined location. Furthermore, the camera 3 may be installed outside the system (for example, an external camera installed in a location remote from the system) as long as it is capable of providing image data to the data acquisition unit 61, database unit 51, etc., which will be described later.

(Server device)
3, the server device 5 has a database unit 51 that stores past data obtained within the system during past recycling processes. The data stored in the database unit 51 includes image data of the substrate images obtained by the camera 3 and data related to the results of estimation of the amount of metal resources by the resource amount estimation unit 4 (hereinafter simply referred to as "resource amount data").

Image data of substrate images is stored for each lot processed within the system. Each image data for each lot is associated with identification information to identify it. Examples of identification information include information on the type of substrate (product name, etc.) for each lot, the lot number, and the processing date and time, but there are no particular limitations as long as it enables identification. The format of the identification information also includes, for example, text format, two-dimensional or one-dimensional barcode format, but there are no particular limitations as long as it is recognizable.

The resource quantity data, which is the estimation result from the resource quantity estimation unit 4, is associated with image data and stored for each lot processed within the system. In other words, the database unit 51 associates image data with resource quantity data and stores it for each lot.

(Data Processing Device)
The data processing device 6 corresponds to one specific example of the "data processing device" according to the present invention, and is configured with hardware resources such as an arithmetic unit consisting of a combination of a CPU, ROM, RAM, etc., for controlling the operation of the entire system, and a storage unit such as a flash memory, HDD, etc. In other words, the data processing device 6 is configured with hardware resources as a computer device, and when the arithmetic unit executes a predetermined program stored in the storage unit, the program (software) and the hardware resources work together to control the operation of the entire system.

In addition, the data processing device 6 is configured to function as at least a data acquisition unit 61, a first feature extraction unit 62, a second feature extraction unit 63, a third feature extraction unit 64, a fourth feature extraction unit 65, a similarity determination unit 66, and an evaluation information output unit 67, as a result of the calculation unit executing a predetermined program.

The data acquisition unit 61 has the function of acquiring image data of board images from the camera 3 or the database unit 51. Specifically, the data acquisition unit 61 acquires image data of board images of the target lot to be analyzed by the resource quantity estimation unit 4 from the camera 3, and acquires image data of board images to be compared with the target lot from the database unit 51. Note that the data format of the acquired image data is not particularly limited.

The first feature extraction unit 62 has a function of extracting, as the first feature, a feature related to the color information of the image data acquired by the data acquisition unit 61. In other words, the first feature extraction unit 62 extracts a different type of feature from the fourth feature extraction unit 65 described below, and corresponds to one specific example of a "different type of feature extraction unit" according to the present invention. The first feature will be described in more detail below.

The second feature extraction unit 63 has a function of extracting, as a second feature, a feature related to the coefficient of variation of the substrate distribution of the image data acquired by the data acquisition unit 61. In other words, the second feature extraction unit 63 extracts a different type of feature from the fourth feature extraction unit 65 described below, and corresponds to another specific example of the "different feature extraction unit" according to the present invention. The second feature will be described in more detail below.

The third feature extraction unit 64 has a function of extracting, as a third feature, a feature relating to the amount of a predetermined shape portion of a waste electronic circuit board in the image data acquired by the data acquisition unit 61. In other words, the third feature extraction unit 64 extracts a different type of feature from the fourth feature extraction unit 65 described below, and corresponds to yet another specific example of a "different type of feature extraction unit" according to the present invention. The third feature will be described in more detail below.

The fourth feature extraction unit 65 has a function of extracting the image complexity specified by the image data acquired by the data acquisition unit 61 and setting this as the fourth feature of the image data. In other words, the fourth feature extraction unit 65 corresponds to a specific example of a "feature extraction unit" according to the present invention. The fourth feature will be described in more detail below.

The fourth feature extraction unit 65 also functions as an area separation unit 65a. The area separation unit 65a separates the image data acquired by the data acquisition unit 61 into a board area and a background area based on the fourth feature extracted by the fourth feature extraction unit 65. The board area and background area will be described in more detail below.

The similarity determination unit 66 has a function for determining whether one image data acquired from the camera 3 by the data acquisition unit 61 for a target lot to be analyzed by the resource amount estimation unit 4 is similar to other image data acquired from the database unit 51 by the data acquisition unit 61 for comparison with the one image data. The similarity determination unit 66 performs similarity determination using at least the fourth feature amount, and preferably the first, second, and third feature amounts in addition to the fourth feature amount, as indicators. The similarity determination unit 66 also narrows down the image data to be subjected to similarity determination based on the identification information of the image data.

The evaluation information output unit 67 receives the results of the similarity determination unit 66 and outputs various pieces of information identified from the results of the similarity determination as evaluation information for the group of boards in the target lot. The evaluation information output by the evaluation information output unit 67 is information provided for the process of approving the component analysis results in the resource quantity estimation unit 4. Specifically, the evaluation information includes, for example, other image data determined by the similarity determination unit 66 to have a high degree of similarity to one piece of image data for the target lot, and resource quantity data corresponding to the image data with a high degree of similarity. This information is extracted from the database unit 51 by the evaluation information output unit 67 to form the evaluation information.

(program)
In the data processing device 6 having the above-described functional configuration, the functions of the above-described units 61 to 67 are realized by the execution of a predetermined program by the calculation unit constituting the data processing device 6. In other words, the predetermined program that realizes the functions of the units 61 to 67 corresponds to one embodiment of the "data processing program" according to the present invention.

In this case, the specific programs that realize each function may be provided by being stored on a recording medium readable by the computer device (e.g., a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc.), as long as they can be installed in the data processing device 6 as a computer device, or they may be provided from outside via a network such as the Internet or a dedicated line.

(Lighting Department)
The lighting unit 7 irradiates light onto the group of boards 1 to be imaged by the camera 3, and for this purpose, it is equipped with a light-shielding film 7a arranged to cover the camera 3 and part of the belt conveyor 2, a reflector 7b arranged in the space covered by the light-shielding film 7a, and an illumination lamp 7c that emits light toward the reflector 7b. By providing such light-shielding film 7a, reflector 7b, and illumination lamp 7c, the lighting unit 7 is configured to irradiate indirect light onto the group of boards 1 on the belt conveyor 2 while blocking out external light.

<Procedure for determining whether a group of boards is similar or not>
Next, an example of processing operations in the system configured as described above will be described. Here, the procedure for determining whether a group of boards is similar or not by the data processing device 6 will be mainly specifically described. The procedure given as an example here corresponds to one specific example of the "method for determining whether a group of boards is similar or not" according to the present invention. The procedure of the "method for determining whether a group of boards is similar or not" constitutes part of the procedure of the "method for evaluating a group of boards" according to the present invention.

Figure 4 is a flow chart showing the steps of the method for determining whether a group of substrates is similar, as exemplified in this embodiment.

(Data acquisition)
4, in determining whether a group of boards is similar or not, first, board images of the group of boards 1 of a target lot (hereinafter referred to as the "input lot") supplied onto the belt conveyor 2 are taken by the camera 3, and the image data resulting from this image capture is acquired by the data acquisition unit 61 from the camera 3 (S201). Note that it is preferable that the image data of the board images is associated with unique identification information, for example, when the image is captured by the camera 3.

At this time, the group of boards 1 of the input lot is transported into a space covered by the light-shielding film 7a of the lighting unit 7 and is imaged by the camera 3 within that space. Within that space, light emitted from the illumination lamp 7c is reflected by the reflector 7b, and the reflected light (i.e., indirect light) reaches the group of boards 1 on the belt conveyor 2. Therefore, the group of boards 1 on the belt conveyor 2 is not affected by external light due to the shielding provided by the light-shielding film 7a. Furthermore, the group of boards 1 on the belt conveyor 2 is imaged by the camera 3 in a state where indirect light uniformly illuminates the entire imaging area without localized concentrations of direct light. More specifically, the group of boards 1 on the belt conveyor 2 is imaged by the camera 3 in a state where differences in brightness (luminance information) between the center and edges are suppressed and shadow areas are eliminated on the belt conveyor 2. Furthermore, while direct light can cause overexposure in areas where light is concentrated, which can result in the inability to capture color information in those areas, this can be prevented by using indirect light. In other words, the camera 3 can capture images of the boards in the group 1 on the belt conveyor 2 under stable conditions that are generally constant, while eliminating the effects of imaging timing, environmental conditions, etc.

5A and 5B show specific examples of board images. The example in Fig. 5A shows a case where the type of waste electronic board is an expansion board type board, and multiple expansion board type boards are grouped together so as to overlap each other to form board group 1. The example in Fig. 5B shows a case where the type of waste electronic board is a CPU board, and multiple CPU boards are grouped together so as to overlap each other to form board group 1.
In this way, the images of the boards in the board group 1 show a collection of one type of waste electronic boards, but the images will be different depending on the type of waste electronic board. Furthermore, even if the waste electronic boards in the board group 1 are the same type, the images of the boards in the board group 1 will not be the same for different lots, because the overlapping state and position on the belt conveyor 2 are indefinite and will differ for each lot.

When the image data of the input lot is acquired, as shown in FIG. 4, the data acquisition unit 61 accesses the database unit 51 and acquires the image data of the processed lot stored in the database unit 51 (hereinafter referred to as "database image data") (S202).

Since the image data in the database constitutes past data, the amount of data is enormous and it is likely to include data from discarded electronic circuit boards of different types from the input lot. Therefore, the similarity determination unit 66 may narrow down the image data in the database to be used for similarity determination. Specifically, for example, the similarity determination unit 66 focuses on the identification information of the image data of the input lot and the identification information of the image data in the database, and calculates the similarity between them. The similarity may be calculated using publicly known techniques, such as the Levenshtein distance if the identification information is text information specifying the circuit board product name. Once the similarity between them is calculated, the similarity is compared with a preset threshold, and only image data from lots with a similarity equal to or greater than the threshold is used for similarity determination. Narrowing down the image data in this way improves processing efficiency compared to not narrowing down the image data.

(Extraction of first feature amount)
Thereafter, the first feature extraction unit 62 extracts first features for each of the image data of the input lot and the image data in the database (for each image data of each lot in the case of multiple lots, the same applies below) (S203). More specifically, the first feature extraction unit 62 calculates a feature related to the color information of the image data as the first feature. The types of colors contained in a board image vary depending on the type of waste electronic board that constitutes the board image. Therefore, color information is considered to be very useful in characterizing the board image.

The first feature extraction unit 62 can calculate a feature related to the hue value of the HSV color system as a feature related to color information (i.e., a first feature). The HSV color system is expressed using three components: H (hue), S (saturation), and V (lightness), and compared to the RGB color system, it can capture color differences in a manner closer to human perception. In particular, feature values that focus on the hue value of the HSV color system provide robustness to lighting conditions, etc., when capturing images with the camera 3, making it possible to obtain judgment results that are close to those of an operator's visual judgment.

The calculation of features related to hue values can be performed using the specific method disclosed in Patent Document 1.

As a result, the first feature quantity extraction section 62 extracts the first feature quantity (f _ci ) of the image data of the input lot and the first feature quantity (f _cdi ) of the image data in the database.

(Extraction of second feature amount)
Furthermore, the second feature extraction unit 63 extracts second feature quantities from the image data of the input lot and the image data of the database (S204). More specifically, the second feature extraction unit 63 calculates, as the second feature quantity, a feature quantity related to the coefficient of variation of the substrate distribution. The substrate distribution can be identified based on the distribution of predetermined-shape portions of the waste electronic substrate in the image. Examples of predetermined-shape portions of the waste electronic substrate include linear-shaped portions corresponding to the terminal portions of the waste electronic substrate. In other words, the second feature extraction unit 63 calculates, as the feature quantity related to the coefficient of variation of the substrate distribution (i.e., the second feature quantity), a feature quantity that represents the distribution status of the terminal portions (linear-shaped portions) of the waste electronic substrate using a coefficient of variation.

Waste electronic circuit boards generally have terminal sections where electrode terminals are arranged in rows, and this arrangement varies depending on the type of waste electronic board. Therefore, quantifying the distribution of such terminal sections (linear sections) within an image is thought to be extremely useful in characterizing the board image. Furthermore, waste electronic boards are equipped with a variety of components in addition to the terminal sections, but focusing only on the terminal sections (linear sections) can improve processing efficiency.

The feature quantity representing the distribution of terminal portions (linear portions) of waste electronic circuit boards using a coefficient of variation can be calculated using the specific method disclosed in Patent Document 1.

As a result, the second feature quantity extraction section 63 extracts the second feature quantity (f _v ) of the image data of the input lot and the second feature quantity (f _vd ) of the image data in the database.

(Extraction of third feature amount)
Furthermore, the third feature extraction unit 64 extracts a third feature from each of the image data of the input lot and the image data in the database (S205). More specifically, the third feature extraction unit 64 calculates, as the third feature, a feature related to the quantity of predetermined-shape portions of the waste electronic board. As in the case of the second feature, the predetermined-shape portions are linear-shaped portions corresponding to the terminal portions of the waste electronic board. The quantity of the predetermined-shape portions can be exemplified by the number of linear-shaped portions. In other words, the third feature extraction unit 64 calculates, as the feature related to the quantity of predetermined-shape portions of the waste electronic board (i.e., the third feature), a feature related to the number of terminal portions (linear-shaped portions) of the waste electronic board (i.e., the number of linear portions in the image).

In PCB images, the overlapping state and position of each discarded electronic board varies from lot to lot. Therefore, if the number of straight lines in the image can be quantified and the number of terminals present can be estimated, it will be possible to grasp the degree and amount of overlapping of each board, which is thought to be extremely useful in characterizing PCB images.

The feature value related to the number of straight lines in an image can be calculated using the specific method disclosed in Patent Document 1.

As a result, the third feature quantity extraction section 64 extracts the third feature quantity (f _L ) of the image data of the input lot and the third feature quantity (f _dL ) of the image data of the database.

(Extraction of Fourth Feature Amount)
Furthermore, the fourth feature quantity extraction unit 65 extracts a fourth feature quantity from each of the image data of the input lot and the image data of the database (S206). More specifically, the fourth feature quantity extraction unit 65 obtains a feature quantity related to the complexity of the board image of the waste electronic board as the fourth feature quantity.

The complexity (Dissimilarity) of a circuit board image is an index that represents the uniformity of the image, and takes on a value between 0.00 and 1.00. Specifically, the more non-uniform the pixel values that make up the image are and the greater the difference between these values, the higher the complexity of the image. On the other hand, if the pixel values are uniform and the difference between shades is small, the complexity of the image will be low.

For example, in the group of expansion boards shown in Figure 5(a), the overlapping expansion boards are relatively large, so the complexity of the board images for that group tends to be low. In contrast, in the group of CPU boards shown in Figure 5(b), the overlapping CPU boards are relatively small, so the complexity of the board images for that group tends to be high. In this way, when dealing with a collection of discarded electronic boards, the complexity of the board images will vary depending on the type of discarded electronic boards, the degree of overlap, etc. Therefore, image complexity is thought to be very useful in characterizing board images.

The complexity of an image can be calculated by analyzing the texture of the image, and the fourth feature amount extraction unit 65 is configured to extract the complexity of the image by analyzing the texture of the image.
Texture analysis is a technique for quantitatively classifying images by quantifying the spatial patterns of the image. Specifically, for example, a gray-level co-occurrence matrix (GLCM) is generated and statistical information is extracted from the matrix to describe the texture characteristics of the image. Note that specific techniques for texture analysis and complexity extraction using texture analysis can be performed using known techniques, and detailed descriptions thereof will be omitted here.

For the complexity of such an image, the fourth feature extraction unit 65 calculates the coefficient of variation of the complexity and sets the result as the fourth feature. By using the coefficient of variation of the complexity as the fourth feature, it becomes possible to determine the size and distribution of the waste electronic boards in the board image from the fourth feature. Specifically, if the board image contains, for example, a large number of small waste electronic boards, the coefficient of variation tends to be large, and if the board image contains a small number of large waste electronic boards, the coefficient of variation tends to be small. Furthermore, for example, if the waste electronic boards are evenly scattered throughout the board image, the coefficient of variation tends to be large, and if the waste electronic boards are concentrated in a certain area, the coefficient of variation tends to be small. Specific methods for calculating the coefficient of variation of the complexity will be described in detail below.

Here, we will explain in more detail the specific steps for extracting the fourth feature by the fourth feature extraction unit 65.

When extracting the fourth feature amount, the fourth feature amount extracting section 65 first separates the substrate region and the background region using the region separating section 65a.
FIG. 6 is an explanatory diagram showing an example of the substrate region and the background region.
As shown in the figure, the circuit board image captured by camera 3 contains a mixture of a circuit board area, which is an area where overlapping waste electronic circuit boards are captured, and a background (belt conveyor) area, which is an area where no waste electronic circuit boards are present and where the surface of the belt conveyor 2 is captured in the background. In such a circuit board image, the background area corresponds to a portion that is not affected by the size or scattering of the waste electronic circuit boards, i.e., a portion that becomes a noise component when extracting the fourth feature. Therefore, when extracting the fourth feature, it is desirable to exclude the background area and use only the circuit board area as the extraction target area.
Therefore, the region separating unit 65a separates the image data of the input lot and the image data of the database into a substrate region and a background region, respectively. The separation of the substrate region and the background region may be performed based on the complexity of the substrate image extracted by the fourth feature amount extracting unit 65, as will be described below.

FIG. 7 is an explanatory diagram (part 1) showing a specific example of a procedure for extracting the complexity of a board image, which is the fourth feature amount.
For example, when image data of the board image shown in Fig. 7(a) is acquired, the fourth feature amount extraction unit 65 divides the board image into an arbitrary number of vertical and horizontal portions as shown in Fig. 7(b). The number of portions is not particularly limited, but a small number of portions reduces the resolution for distinguishing between the board region and the background region, while a large number of portions can increase the processing load. Therefore, it is considered that the number of portions is set to, for example, 48 x 48. As a result, the board image is divided into a plurality of cells, for example, 48 x 48.

After dividing the board image into multiple cells, the fourth feature extraction unit 65 then calculates the image complexity for each cell. As described above, the image complexity can be calculated by analyzing the texture of the image. As a result, a value between 0.00 and 1.00 is calculated as the image complexity for each cell.

Then, after calculating the image complexity for each cell, the fourth feature extraction unit 65 corresponds the calculation result to a hue value in the HSV color system, which is expressed using the three components H (hue), S (saturation), and V (lightness). This creates the complexity heat map shown in Figure 7(c). In the complexity heat map, each cell is expressed using a different hue value depending on the complexity of that cell.

The fourth feature extraction unit 65 then compares the complexity value of each cell in the created complexity heat map with a preset threshold value. The threshold value is set based on empirical rules such as simulations to a value that allows for good discrimination between the board region and the background region. The fourth feature extraction unit 65 then performs a binarization process by separating each cell into cells above the threshold and cells below the threshold, replacing cells above the threshold with white pixel values and cells below the threshold with black pixel values. This results in the creation of the board region mask image shown in FIG. 7(d). In the board region mask image, the board region is represented by white pixel values and the background region is represented by black pixel values. By focusing on the white pixel value portions in this board region mask image, the board image is separated into the board region and the background region, making it possible to exclude the background region and leave only the board region as the extraction target region.

After creating the substrate region mask image, the fourth feature amount extraction unit 65 subsequently calculates the coefficient of variation of the complexity.
FIG. 8 is an explanatory diagram (part 2) showing a specific example of the procedure for extracting the complexity of the board image, which is the fourth feature amount.
As shown in the figure, the fourth feature extraction unit 65 calculates the coefficient of variation of complexity using a virtual grid set in the substrate region mask image. Specifically, for example, as shown in FIG. 8A, virtual perpendicular lines are set at predetermined intervals (e.g., every five pixels) from one end (e.g., the left end) of the substrate region mask image, and the number of white pixels present on each perpendicular line is calculated. Also, for example, as shown in FIG. 8B, virtual horizontal lines are set at predetermined intervals (e.g., every five pixels) from one end (e.g., the top end) of the substrate region mask image, and the number of white pixels present on each horizontal line is calculated. After calculating the number of white pixels present on each perpendicular line and each horizontal line, the coefficient of variation is calculated using a predetermined equation, including an equation that divides the variance (standard deviation) of the distribution of white pixels by the average number of white pixels. Specifically, for example, the coefficient of variation _fs is calculated using the following equation (1):

In equation (1), n represents the number of vertical or horizontal lines, _xk represents the number of white pixels on the kth vertical line, the overlined x represents the average value of white pixels on the vertical lines, _yk represents the number of white pixels on the kth horizontal line, and the overlined y represents the average value of white pixels on the horizontal lines.

The coefficient of variation fs of the distribution of white pixels obtained in this way is used as a feature related to the coefficient of variation of the complexity of the board region in the board image, i.e., the fourth feature in this embodiment. This fourth feature makes it possible to determine the size and distribution of the waste electronic boards in the board image. Specifically, for example, if a board image contains many small waste electronic boards, the coefficient of variation tends to be large, and if a small number of large waste electronic boards are shown, the coefficient of variation tends to be small. Furthermore, for example, if waste electronic boards are evenly scattered throughout the board image, the coefficient of variation tends to be large, and if the waste electronic boards are concentrated in a certain area, the coefficient of variation tends to be small.

The fourth feature quantity extraction unit 65 extracts the above-described fourth feature quantity for each of the image data of the input lot and the image data of the database. In other words, the fourth feature quantity extraction unit 65 extracts the fourth feature quantity (f _s ) of the image data of the input lot and the fourth feature quantity (f _sd ) of the image data of the database.

9A and 9B show specific examples of circuit board images and corresponding complexity heat maps. Fig. 9A shows a circuit board image and a complexity heat map when the discarded electronic circuit board is an expansion board type circuit board, Fig. 9B shows a circuit board image and a complexity heat map when the discarded electronic circuit board is a motherboard type circuit board, Fig. 9C shows a circuit board image and a complexity heat map when the discarded electronic circuit board is a CPU circuit board, and Fig. 9D shows a circuit board image and a complexity heat map when the discarded electronic circuit board is a mobile phone. In each of Figs. 9A to 9D, the left side of the figure is the circuit board image, and the right side of the figure is the complexity heat map.
9(a) to (d) show that the complexity heat maps show clear differences depending on the type of waste electronic board. Specifically, for example, in a board image showing several large boards, as in FIG. 9(a) or (b), the areas with high complexity are distributed clustered together in the complexity heat map, whereas in a board image showing many small boards, as in FIG. 9(c) or (d), the areas with high complexity are distributed scatteredly in the complexity heat map. Thus, it is clear that the type of waste electronic board causes differences in the distribution of complexity in board images.
Therefore, calculating the coefficient of variation of image complexity and using it as the fourth feature is extremely useful for determining the size and scattering of waste electronic boards in a board image and classifying the type of the waste electronic boards. By classifying the type of waste electronic boards, it becomes possible to grasp in advance the usefulness of board group 1 of the input lot.

(similarity judgment)
After extracting the first, second, third, and fourth feature quantities, the similarity determination unit 66 determines whether the image data of the input lot and the image data in the database are similar to each other, as shown in Figure 4. At this time, the similarity determination unit 66 uses at least the fourth feature quantity, and preferably the first, second, and third feature quantities in addition to the fourth feature quantity, as an index to calculate the image similarity between the image data of the input lot and the image data in the database, thereby determining whether they are similar to each other (S207). The calculation of the image similarity is performed, for example, using the following procedure.

First, the similarity determination unit 66 calculates the feature similarity _Fc between the image data of the input lot and the image data in the database for the first feature amount. Specifically, when the first feature amount of the image data of the input lot is _fci , the first feature amount of the image data in the database is _fcdi , and the number of regions in the hue division used to calculate the first feature amount is 24, the feature similarity _Fc is calculated using the following formula (2). Note that when i=1 in formula (2), _fci and _fcdi are the proportions of image data classified into hue division 1 when the hue circle is divided into 24 stages.

Furthermore, the similarity determination unit 66 calculates the feature similarity _Fv between the image data of the input lot and the image data of the database for the second feature amount. Specifically, when the second feature amount of the image data of the input lot is _fv and the second feature amount of the image data of the database is _fvd , the feature similarity _Fv is calculated using the following formula (3):

Furthermore, the similarity determination unit 66 calculates the feature similarity F _L between the image data of the input lot and the image data of the database for the third feature amount. Specifically, when the third feature amount of the image data of the input lot is f _L and the third feature amount of the image data of the database is f _dL , the feature similarity F _L is calculated using the following formula (4):

Furthermore, the similarity determination unit 66 calculates the feature similarity _Fs between the image data of the input lot and the image data of the database for the fourth feature amount. Specifically, when the fourth feature amount of the image data of the input lot is _fs and the fourth feature amount of the image data of the database is _fsd , the feature similarity _Fs is calculated using the following equation (5):

Then, the similarity determination unit 66 calculates the image similarity _S between the image data of the input lot and the image data in the database using the following equation ( ₆ ) based on the calculation results of the four types of feature similarities ( _Fc , _Fv , FL, Fs).

In equation (6), coefficient α represents a weighting coefficient for the first feature amount, coefficient β represents a weighting coefficient for the second feature amount, coefficient γ represents a weighting coefficient for the third feature amount, and coefficient δ represents a weighting coefficient for the fourth feature amount. In other words, when calculating the image similarity S, the similarity determination unit 66 assigns a weight to each of the first to fourth feature amounts used as an index for similarity determination.

The balance of the magnitudes of the coefficients α, β, γ, and δ is not particularly limited, and may be set appropriately depending on experimental and simulation results, etc. For example, if the fourth feature value is used as an indicator for determining similarity and the first, second, and third feature values are not taken into consideration, the coefficients α, β, and γ would be set to "0."

By going through the above steps, the similarity determination unit 66 calculates the image similarity S between the image data of the input lot and the image data in the database. Note that if the database contains image data for multiple lots, the similarity determination unit 66 will calculate the image similarity S for each of the multiple lots relative to the image data of the input lot.

(Evaluation information output)
When the similarity determination unit 66 calculates the image similarity S, the evaluation information output unit 67 outputs various information including the calculation result as evaluation information for the input lot. The output may be performed by a known method such as image display or printout. By referring to this output result, the system user can understand the content of the evaluation information for the input lot, including the result of the similarity determination.

By outputting such evaluation information, the evaluation results can be attached to the estimation results of the amount of metal resources contained in the group of substrates in the input lot. This makes it possible to ensure a high level of reliability in the estimation results of the amount of metal resources for the input lot.

(Specific calculation example)
Here, a brief explanation will be given of a specific example of the calculation result of the image similarity S. Here, the example will be one in which the image similarity S between the image data of an input lot and the image data of a database of multiple lots is calculated, and based on the calculation result, the image data of the database of multiple lots with the highest image similarity S is extracted, the value of the content ratio of a predetermined metal (e.g., gold) is recognized based on the resource amount data for the image data of the extracted lot, and if the metal content ratio is within ±10% of the metal content ratio for the input lot, the judgment is successful, and the ratio of the number of successful judgments to the number of data used is calculated as the judgment success rate.

As a result, it was confirmed that if similarity determination is performed using at least the fourth feature amount as an index, a determination success rate of, for example, 74% or more can be achieved.
In particular, it was confirmed that a high determination success rate of 86.3% could be obtained by using all of the first, second, third, and fourth feature quantities as indices and weighting each feature quantity, for example, with coefficients α = 0.8, β = 1.0, γ = 0.4, and δ = 0.6. Incidentally, when the first, second, and third feature quantities were used as indices but the fourth feature quantity was not used, the determination success rate was 84.2%, whereas using the fourth feature quantity as an index increased the determination success rate by 2.1%.
This indicates that the feature related to image complexity (fourth feature) contributes to improving the accuracy of similarity determination. Furthermore, by appropriately setting the weighting coefficients for each feature, it becomes possible to vary the degree of contribution of each feature, which is expected to further improve the accuracy of similarity determination.

The above shows that the similarity determination method in this embodiment is extremely useful for calculating similarity in circuit board images and determining similarity based on the calculation results.

<Effects of this embodiment>
According to the data processing device, data processing program, substrate group similarity determination method, substrate group evaluation system, and substrate group evaluation method described in this embodiment, the following effects can be obtained.

(a) According to this embodiment, the similarity determination of the image data of a board image is performed by extracting the complexity of the board image as a fourth feature and using the fourth feature as an index. Therefore, when a collection of discarded electronic boards (a group of boards) is processed collectively, even if there are multiple types of boards to be processed, the complexity varies depending on the type of board and the degree of overlap between the boards, etc., so that the similarity determination can be performed with high reliability. Moreover, by utilizing the differences in complexity depending on the type of board, the degree of overlap, etc., these differences are immediately reflected in the results of the similarity determination, thereby improving the efficiency of the similarity determination.
In other words, according to this embodiment, it is possible to accurately and efficiently determine whether the image data of a collection of waste electronic substrates is similar to other image data, thereby contributing to the establishment of a recycling business for waste electronic substrates.

(b) According to this embodiment, when extracting the fourth feature, the complexity of the image is extracted by analyzing the image's texture. Therefore, the complexity of the image can be quantitatively extracted, which is extremely useful for efficiently and reliably determining whether the image data of board images are similar.

(c) According to this embodiment, the coefficient of variation of the complexity of the board image is calculated, and the result of this calculation is used as the fourth feature. Therefore, the fourth feature also reflects the distribution of complexity and reveals differences in the degree of dispersion of the boards, making it extremely useful for efficiently and reliably determining whether the image data of board images are similar.

(d) According to this embodiment, prior to determining whether the image data of the board image is similar, the region separating unit 65a separates the board image into a board region and a background region. Therefore, when extracting the fourth feature amount that serves as an index for determining whether the board image is similar, it is possible to exclude the background region that becomes a noise component and to use only the board region as the region to be extracted for the fourth feature amount, which is extremely useful for efficiently determining whether the image data of the board image is similar with high reliability.
Moreover, according to this embodiment, the separation between the board region and the background region is performed based on the complexity of the board image, so that the separation between the board region and the background region can be performed efficiently with high reliability.

(e) According to this embodiment, the similarity determination for the image data of the circuit board image is performed using, in addition to the fourth feature, the first feature, second feature, and third feature, which are different types of feature from the fourth feature, as indicators. The first feature, second feature, and third feature are each of a completely different type from the fourth feature, and therefore can play a complementary role to the fourth feature. Therefore, the addition of the first feature, second feature, and third feature further improves the reliability of the similarity determination.

(f) According to this embodiment, when determining whether the image data of a circuit board image is similar, each of the first to fourth feature amounts is used as an index, and a weight is assigned to each feature amount. Therefore, by adjusting which feature amount is emphasized depending on the degree of weighting, it is possible to vary the degree of contribution of each feature amount, which is expected to further improve the accuracy of similarity determination.

(g) According to this embodiment, when capturing images of the substrate group 1, the illumination unit 7 illuminates the substrate group 1, but the illumination unit 7 is configured to illuminate the substrate group 1 with indirect light while blocking external light. Therefore, the substrate group 1 is captured without being affected by external light, and with indirect light illuminating the entire imaging area uniformly. In other words, according to this embodiment, the substrate images of the substrate group 1 can be captured under stable conditions that are generally constant, while eliminating the influence of imaging timing, environmental conditions, etc. This is extremely useful for further improving the reliability of similarity determinations by acquiring image data under stable conditions.

<Modification>
Although the embodiments of the present invention have been described above, the above disclosure shows exemplary embodiments of the present invention, and the technical scope of the present invention is not limited to the above exemplary embodiments.

In the above-described embodiment, we have mainly used all of the first through fourth feature quantities as indices, but the present invention is not limited to this. That is, the feature quantity used as an index needs to be at least the fourth feature quantity. Furthermore, it is also possible to use a different type of feature quantity from the fourth feature quantity as an index, but even in this case, it is not necessary to use all of the first through third feature quantities; at least one of these may be used.

1...Collection of waste electronic circuit boards (group of circuit boards), 2...Belt conveyor, 3...Camera, 4...Resource amount estimation unit, 5...Server device, 6...Data processing device, 7...Illumination unit, 51...Database unit, 61...Data acquisition unit, 62...First feature amount extraction unit, 63...Second feature amount extraction unit, 64...Third feature amount extraction unit, 65...Fourth feature amount extraction unit, 66...Similarity determination unit, 67...Evaluation information output unit

Claims (10)

a data acquisition unit that acquires image data of the collection of waste electronic substrates;
a feature extraction unit that extracts image complexity from the image data and sets it as a feature of the image data;
a similarity determination unit that determines whether one image data and another image data acquired by the data acquisition unit are similar or not by using the feature amount as an index;
Equipped with
the complexity of the image extracted by the feature extraction unit is an index representing the uniformity of pixel values constituting the image;
The feature extraction unit calculates a coefficient of variation of the complexity using a predetermined arithmetic expression including an arithmetic expression that divides a variance value of a distribution of a predetermined number of pixels in a plot space of the complexity of each pixel value by an average value of the predetermined number of pixels, and sets the calculation result as the feature. The data processing device according to claim 1 , wherein the feature extraction unit is configured to extract the complexity of the image by analyzing the texture of the image. a region classification unit that classifies the image data into a substrate region and a background region based on the feature extracted by the feature extraction unit,
The data processing device according to claim 1 , wherein the similarity determination section is configured to determine whether the substrate area section is a target for similarity determination. a different feature extraction unit that extracts a different feature from the feature from the image data;
The data processing device according to claim 1 , wherein the similarity determination unit is configured to perform similarity determination using, as an index, the feature extracted by the feature extraction unit and another feature extracted by the another feature extraction unit. The data processing device according to claim 4 , wherein the similarity determination unit is configured to weight each feature amount used in the similarity determination when making the similarity determination. an imaging unit that captures an image of the collection of waste electronic substrates to obtain the image data;
an illumination unit that irradiates light onto the collection of waste electronic substrates that are to be imaged by the imaging unit,
The data processing device according to claim 1 , wherein the illumination unit is configured to irradiate the collection of waste electronic circuit boards with indirect light while blocking external light. Computer,
a data acquisition unit that acquires image data of the collection of waste electronic substrates;
a feature extraction unit that extracts image complexity from the image data and sets it as a feature of the image data;
a similarity determination unit that determines whether one image data and another image data acquired by the data acquisition unit are similar or not by using the feature amount as an index;
It also functions as a
the complexity of the image extracted by the feature extraction unit is an index representing the uniformity of pixel values constituting the image;
The feature extraction unit calculates a coefficient of variation of the complexity using a predetermined arithmetic expression including an arithmetic expression that divides the variance of the distribution of a predetermined number of pixels in a plot space of the complexity of each pixel value by the average value of the predetermined number of pixels, and sets the calculation result as the feature. a data acquisition step of acquiring image data of the collection of waste electronic substrates;
a feature extraction step of extracting image complexity from the image data and setting the image complexity as a feature of the image data;
a similarity determination step of determining whether one image data and another image data acquired in the data acquisition step are similar to each other by using the feature amount as an index;
Equipped with
The complexity of the image extracted in the feature extraction step is an index representing the uniformity of pixel values constituting the image,
In the feature extraction step, a coefficient of variation of the complexity is calculated using a predetermined arithmetic expression including an arithmetic expression for dividing the variance of a distribution of a predetermined number of pixels in a plot space of the complexity of each pixel value by the average value of the predetermined number of pixels, and the calculation result is used as the feature. an image capturing unit that captures an image of the collection of waste electronic substrates;
a resource amount estimation unit that analyzes the aggregate and estimates the amount of metal resources contained in the aggregate;
a database unit that stores and accumulates the image capturing results from the image capturing unit and the estimation results from the resource amount estimation unit in association with each other;
a data acquisition unit that acquires image data of the collection from the image capturing unit or the database unit;
a feature extraction unit that extracts image complexity from the image data and sets it as a feature of the image data;
a similarity determination unit that uses the feature amount as an index to determine whether one piece of image data acquired by the data acquisition unit regarding the analysis target in the resource amount estimation unit is similar to another piece of image data acquired by the data acquisition unit from the database unit for comparison with the one piece of image data;
an evaluation information output unit that extracts from the database information corresponding to image data that has been determined to have a high similarity to the one image data by the similarity determination unit, and outputs the extracted information as evaluation information about the collection that is the analysis target of the resource amount estimation unit;
Equipped with
the complexity of the image extracted by the feature extraction unit is an index representing the uniformity of pixel values constituting the image;
the feature extraction unit calculates a coefficient of variation of the complexity using a predetermined arithmetic expression including an arithmetic expression for dividing a variance value of a distribution of a predetermined number of pixels in a plot space of the complexity of each pixel value by an average value of the predetermined number of pixels, and sets the calculation result as the feature. an image capturing step of capturing an image of the collection of waste electronic substrates with an image capturing unit;
a resource amount estimation step of analyzing the aggregate to estimate the amount of metal resources contained in the aggregate;
a data storage step of storing and accumulating the image capturing result in the image capturing step and the estimation result in the resource amount estimation step in a database unit in association with each other;
a data acquisition step of acquiring image data of the collection from the image capturing unit or the database unit;
a feature extraction step of extracting image complexity from the image data and setting the image complexity as a feature of the image data;
a similarity determination step of determining whether one image data item acquired in the data acquisition step for the analysis target in the resource amount estimation step is similar to another image data item acquired from the database unit in the data acquisition step for comparison with the one image data item, using the feature amount as an index;
an evaluation information output step of retrieving, from the database unit, correspondence information of image data determined to have a high similarity to the one image data in the similarity determination step, and outputting the information as evaluation information about the collection to be analyzed in the resource amount estimation step;
Equipped with
The complexity of the image extracted in the feature extraction step is an index representing the uniformity of pixel values constituting the image,
the feature extraction step calculates a coefficient of variation of the complexity using a predetermined arithmetic expression including an arithmetic expression for dividing a variance value of a distribution of a predetermined number of pixels in a plot space of the complexity of each pixel value by an average value of the predetermined number of pixels, and the calculation result is used as the feature.