JP2020173547A - Image inspection system, image inspection device and image inspection program - Google Patents

Abstract

To provide an image inspection system, an image inspection device, and an image inspection program for accurately identifying a component including a defect from an image of the component.SOLUTION: A plurality of first segmented images and a plurality of second segmented images are generated by dividing a plurality of training images for a first size and a second size. Feature amounts corresponding to a plurality of types of parameters are extracted from each of the plurality of first normal segmented images corresponding to a normal state among the plurality of first segmented images and the plurality of second normal segmented images corresponding to the normal state among the plurality of second segmented images, and a first parameter in which the similar relation of the feature amounts between the plurality of first normal segmented images and the plurality of second normal segmented images satisfies a predetermined condition is identified from the plurality of types of parameters. Machine learning of the plurality of first segmented images and the plurality of second segmented images is performed by using the feature amount corresponding to the identified first parameter.SELECTED DRAWING: Figure 6

Description

The present invention relates to an image inspection system, an image inspection device and an image inspection program.

For example, a business operator that provides a service to a user (hereinafter, also simply referred to as a business operator) constructs an information processing system for providing the service to the user. Specifically, the business operator constructs, for example, an information processing system that identifies parts including defects (for example, stains) from images of mass-produced parts.

In such an information processing system, for example, a supervised learning model (hereinafter, also simply referred to as a learning model) is created by performing machine learning on a feature amount extracted from an image to be learned (hereinafter, also referred to as a learning image). Generate. Then, the information processing system inputs the feature amount extracted from the image to be inspected (hereinafter, also referred to as the inspection image) into the learning model to indicate whether or not the component corresponding to the inspection image contains a defect. Get the score.

As a result, the business operator can identify the part including the defect from each mass-produced part by referring to the acquired score (see, for example, Patent Documents 1 to 3).

JP 2015-041164 Japanese Unexamined Patent Publication No. 2010-067068 Japanese Unexamined Patent Publication No. 2010-102690

Here, when generating the learning model as described above, the business operator prepares, for example, a learning image of a part including a defect for each size of the defect. Then, the information processing system generates a learning model for each defect size by using, for example, learning images for each defect size. As a result, the information processing system can accurately determine whether or not the component corresponding to the inspection image contains a defect.

However, the business operator may not be able to sufficiently prepare each of the learning images corresponding to each size of the defect. Therefore, the information processing system cannot comprehensively generate a learning model for each defect size, and cannot accurately determine whether or not the component corresponding to the inspection image contains a defect. There is.

Therefore, in one aspect, it is an object of the present invention to provide an image inspection system, an image inspection device, and an image inspection program that can accurately identify a part containing a defect from an image of the part.

In one aspect of the embodiment, in an image inspection system having a plurality of information processing devices that perform machine learning of a plurality of learned images, each of the plurality of information processing devices performs the plurality of learned images for each first size. A divided image generation unit that generates a plurality of first divided images by dividing and generates a plurality of second divided images by dividing the plurality of learning images into second sizes different from the first size. From each of the plurality of first normally divided images corresponding to the normal state of the plurality of first divided images and the plurality of second normally divided images corresponding to the normal state of the plurality of second divided images. , The feature amount extraction unit that extracts the feature amount corresponding to the plurality of types of parameters, and the feature between the plurality of first normal division images and the plurality of second normal division images from the plurality of types of parameters. The plurality of first divided images and the plurality of second divided images are obtained by using a parameter specifying unit that specifies a first parameter whose quantity similarity satisfies a predetermined condition and a feature amount corresponding to the specified first parameter. It has a machine learning execution unit that performs machine learning of divided images.

According to one aspect, it is possible to accurately identify a part containing a defect from an image of the part.

FIG. 1 is a diagram illustrating a configuration of the information processing system 10. FIG. 2 describes a specific example in the case of generating a learning model for each defect size. FIG. 3 is a diagram illustrating a hardware configuration of the information processing device 1. FIG. 4 is a block diagram of the function of the information processing device 1. FIG. 5 is a flowchart illustrating an outline of the image inspection process according to the first embodiment. FIG. 6 is a flowchart illustrating an outline of the image inspection process according to the first embodiment. FIG. 7 is a flowchart illustrating the details of the image inspection process according to the first embodiment. FIG. 8 is a flowchart illustrating the details of the image inspection process according to the first embodiment. FIG. 9 is a flowchart illustrating the details of the image inspection process according to the first embodiment. FIG. 10 is a flowchart illustrating the details of the image inspection process according to the first embodiment. FIG. 11 is a flowchart illustrating the details of the image inspection process according to the first embodiment. FIG. 12 is a flowchart illustrating the details of the image inspection process according to the first embodiment. FIG. 13 is a flowchart illustrating the details of the image inspection process according to the first embodiment. FIG. 14 is a diagram illustrating a specific example of the process of S22. FIG. 15 is a diagram illustrating a specific example of the process of S23. FIG. 16 is a diagram illustrating a specific example of the processing of S34 and S35. FIG. 17 is a diagram illustrating a specific example of the processing of S34 and S35. FIG. 18 is a diagram illustrating a specific example of the distribution corresponding to the luminance histogram. FIG. 19 is a diagram illustrating a specific example of the distribution corresponding to the gradient histogram. FIG. 20 is a diagram illustrating details of the image inspection process according to the first embodiment.

[Information processing system configuration]
First, the configuration of the information processing system 10 will be described. FIG. 1 is a diagram illustrating a configuration of the information processing system 10.

As shown in FIG. 1, the information processing system 10 (hereinafter, also referred to as an image inspection system 10) includes, for example, one or more information processing devices 1 (hereinafter, also referred to as an image inspection device 1) and one or more operations. Including terminal 5.

The operation terminal 5 is, for example, a PC (Personal Computer), and is a terminal for a business operator to input necessary information or the like. Specifically, the operation terminal 5 inputs, for example, a learning image or an inspection image of a part.

The information processing device 1 generates a learning model by, for example, performing machine learning on a feature amount extracted from a learning image input via an operation terminal 5. Then, for example, when the information processing device 1 receives the inspection image input via the operation terminal 5, the feature amount extracted from the inspection image is input to the learning model. After that, the information processing apparatus 1 determines whether or not the component corresponding to the inspection image contains a defect based on the score output from the learning model with the input of the feature amount.

Here, when generating the learning model as described above, the business operator prepares, for example, a learning image of a part including a defect for each size of the defect. Then, the information processing device 1 generates a learning model for each defect size by using, for example, learning images for each defect size. Hereinafter, a specific example of generating a learning model for each defect size will be described.

[Specific example of learning model for each defect size]
FIG. 2 describes a specific example in the case of generating a learning model for each defect size.

Specifically, the learning image IM11 shown in FIG. 2A is an image including the component IM12. The component IM 12 shown in FIG. 2 (A) includes a defect IM 13. Further, the learning image IM21 shown in FIG. 2C is an image including the component IM22. The component IM22 shown in FIG. 2C includes a defect IM23.

That is, the learning image IM11 shown in FIG. 2A is an image of the component IM12 including the defect IM13 having a relatively small size, and the learning image IM21 shown in FIG. 2C is a defect IM23 having a relatively large size. It is an image about the component IM22 including.

Specifically, when the information processing apparatus 1 learns about the learning image IM11 shown in FIG. 2A, for example, the defect IM13 is out of a plurality of predetermined sizes as the size of the defect that generates the learning model. (For example, the size closest to the size of the defect IM13) corresponding to is specified. Then, as shown in FIG. 2B, the information processing device 1 divides the learning image IM11 for each window size WS11 corresponding to the specified size to generate a plurality of divided images. After that, the information processing device 1 performs machine learning by using the feature amounts extracted from each of the generated plurality of divided images.

Further, when the information processing apparatus 1 learns about the learning image IM21 shown in FIG. 2C, for example, the information processing device 1 corresponds to the defect IM23 among a plurality of predetermined sizes of the defects for generating the learning model. The size to be processed (for example, the size closest to the size of the defect IM23) is specified. Then, as shown in FIG. 2D, the information processing device 1 divides the learning image IM21 for each window size WS21 corresponding to the specified size to generate a plurality of divided images. After that, the information processing device 1 performs machine learning by using the feature amounts extracted from each of the generated plurality of divided images.

As a result, the information processing device 1 can generate a learning model for each defect size. Therefore, the information processing device 1 can use the learning model properly according to the situation, and can accurately determine whether or not the component corresponding to the inspection image contains a defect. ..

However, the operator may not be able to sufficiently prepare learning images for each size of the defect. Therefore, the information processing apparatus 1 cannot comprehensively generate a learning model for each defect size, and cannot accurately determine whether or not the component corresponding to the inspection image contains a defect. In some cases.

Therefore, the information processing device 1 in the present embodiment generates a plurality of divided images (hereinafter, also referred to as first divided images) by dividing a plurality of learning images for each first size. Further, the information processing device 1 generates a plurality of divided images (hereinafter, also referred to as second divided images) by dividing the plurality of learned images into second sizes different from the first size.

Then, the information processing device 1 includes a plurality of divided images corresponding to the normal state of the plurality of first divided images (hereinafter, also referred to as first normal divided images) and a normal state of the plurality of second divided images. From each of the plurality of divided images corresponding to (hereinafter, also referred to as the second normal divided image), the feature quantities corresponding to the plurality of types of parameters are extracted. Further, in the information processing device 1, a parameter (hereinafter, first) in which the similarity relationship between the plurality of first normally divided images and the plurality of second normally divided images satisfies a predetermined condition from a plurality of types of parameters. (Also called a parameter).

After that, the information processing device 1 performs machine learning of the plurality of first divided images and the plurality of second divided images by using the feature quantities corresponding to the specified first parameter.

That is, the information processing apparatus 1 specifies the first parameter corresponding to the feature amount that does not depend on the size of the defect, among the parameters that can extract the feature amount from the first divided image and the second divided image. Then, the information processing device 1 performs machine learning of the learning image by using the feature amount corresponding to the specified first parameter.

As a result, the information processing apparatus 1 can accurately identify defects of a size for which a sufficient number of training images have not been prepared by using the generated learning model. Therefore, the information processing apparatus 1 can accurately determine whether or not the component corresponding to the inspection image contains a defect.

[Hardware configuration of information processing system]
Next, the hardware configuration of the information processing system 10 will be described. FIG. 3 is a diagram illustrating a hardware configuration of the information processing device 1.

As shown in FIG. 3, the information processing device 1 includes a CPU 101 which is a processor, a memory 102, an external interface (I / O unit) 103, and a storage medium 104. The parts are connected to each other via the bus 105.

The storage medium 104 has, for example, a program storage area (not shown) for storing a program 110 for performing a process of identifying a component containing a defect from an image of the component (hereinafter, also referred to as an image inspection process). Further, the storage medium 104 has, for example, a storage unit 130 (hereinafter, also referred to as an information storage area 130) for storing information used when performing an image inspection process. The storage medium 104 may be, for example, an HDD (Hard Disk Drive) or an SSD (Solid State Drive).

The CPU 101 executes the program 110 loaded from the storage medium 104 into the memory 102 to perform the image inspection process.

Further, the external interface 103 communicates with, for example, the operation terminal 5.

[Information processing system functions]
Next, the function of the information processing system 10 will be described. FIG. 4 is a block diagram of the function of the information processing device 1.

As shown in FIG. 4, the information processing device 1 is divided into an information receiving unit 111 and an information management unit 112 by, for example, the hardware such as the CPU 101 and the memory 102 and the program 110 organically collaborating with each other. Various functions including an image generation unit 113, a feature amount extraction unit 114, a parameter identification unit 115, a machine learning execution unit 116, a score acquisition unit 117, an abnormality determination unit 118, and an information output unit 119 are realized. ..

Further, for example, as shown in FIG. 4, the information processing device 1 stores the learning image 131, the first divided image 132, the second divided image 133, and the learning model 134 in the information storage area 130.

First, a function for performing machine learning (generation of the learning model 134) of the learning image 131 will be described.

The information receiving unit 111 receives, for example, a plurality of learning images 131 (learning images about parts) transmitted from the operation terminal 5.

The information management unit 112 stores, for example, a plurality of learning images 131 received by the information receiving unit 111 in the information storage area 130.

The divided image generation unit 113 generates a plurality of first divided images 132 by, for example, dividing a plurality of learning images 131 stored in the information storage area 130 for each first size. Further, the divided image generation unit 113 generates a plurality of second divided images 133 by, for example, dividing the plurality of learning images 131 stored in the information storage area 130 for each second size. Then, the information management unit 112 stores, for example, the plurality of first divided images 132 and the plurality of second divided images 133 generated by the divided image generation unit 113 in the information storage area 130.

The feature amount extraction unit 114 includes a plurality of first normal divided images 132a among the plurality of first divided images 132 generated by the divided image generation unit 113, and a plurality of second divided images 133 generated by the divided image generation unit 113. The feature quantities corresponding to the plurality of types of parameters are extracted from each of the plurality of second normally divided images 133a.

The parameter specifying unit 115 is similar in the feature amount between the plurality of first normal division images 132a and the plurality of second normal division images 133a from the plurality of types of parameters obtained by the feature amount extraction unit 114. Identify a first parameter whose relationship satisfies a predetermined condition.

The machine learning execution unit 116 performs machine learning of the plurality of first divided images 132 and the plurality of second divided images 133 by using the feature quantities corresponding to the first parameter specified by the parameter specifying unit 115. Then, the machine learning execution unit 116 generates the learning model 134 in the information storage area 130.

Next, the function for identifying the component including the defect by using the learning model 134 will be described.

The information receiving unit 111 receives, for example, an inspection image (inspection image of a part) transmitted from the operation terminal 5.

The divided image generation unit 113 generates a plurality of divided images by dividing the inspection image received by the information receiving unit 111 into predetermined sizes. Specifically, the divided image generation unit 113 divides the inspection image received by the information receiving unit 111 for each size specified in advance by the business operator.

The feature amount extraction unit 114 extracts the feature amount corresponding to the first parameter from the plurality of divided images generated by the divided image generation unit 113.

The score acquisition unit 117 outputs the score output from the learning model 134 with the input of the feature amount corresponding to the first parameter extracted by the feature amount extraction unit 114 for each of the plurality of divided images generated by the divided image generation unit 113. To get. That is, the score acquisition unit 117 acquires the score for each of the plurality of divided images generated by the divided image generation unit 113 by using the learning model 134 generated in the information storage area 130.

The abnormality determination unit 118 determines that, of the plurality of divided images generated by the divided image generation unit 113, the divided image whose score acquired by the score acquisition unit 117 is equal to or less than the threshold value is in an abnormal state.

For example, when the abnormality determination unit 118 determines that the divided image in the abnormal state is included, the information output unit 119 outputs information indicating that the inspection image received by the information receiving unit 111 is in the abnormal state. Output to 5.

[Outline of the first embodiment]
Next, the outline of the first embodiment will be described. 5 and 6 are flowcharts illustrating an outline of the image inspection process according to the first embodiment.

As shown in FIG. 5, the information processing device 1 waits until the machine learning execution timing (NO in S1). The machine learning execution timing may be, for example, the timing at which the business operator inputs information to the information processing apparatus 1 to start the image inspection process.

Then, when the machine learning execution timing is reached (YES in S1), the information processing device 1 generates a plurality of first divided images 132 by dividing the plurality of learning images 131 for each first size (S2). .. Further, in this case, the information processing apparatus 1 generates a plurality of second divided images 133 by dividing the plurality of learning images 131 for each second size (S3).

Subsequently, the information processing apparatus 1 includes a plurality of first normal divided images 132a corresponding to the normal state of the plurality of first divided images 132, and a plurality of first normally divided images 132a corresponding to the normal state of the plurality of second divided images 133. The feature quantities corresponding to the plurality of types of parameters are extracted from each of the second normal divided images 133a (S4).

Further, as shown in FIG. 6, the information processing apparatus 1 has a predetermined similarity relationship between the plurality of first normal division images 132a and the plurality of second normal division images 133a from a plurality of types of parameters. The first parameter that satisfies the condition is specified (S5).

After that, the information processing apparatus 1 performs machine learning of the plurality of first divided images 132 and the plurality of second divided images 133 by using the feature quantities corresponding to the first parameter specified in the process of S5 (S6). ..

That is, the information processing apparatus 1 specifies the first parameter corresponding to the feature amount that does not depend on the size of the defect, among the parameters that can extract the feature amount from the first divided image 132 and the second divided image 133. .. Then, the information processing device 1 performs machine learning of the learning image by using the feature amount corresponding to the specified first parameter.

As a result, the information processing apparatus 1 can accurately identify defects of a size for which a sufficient number of training images 131 have not been prepared by using the generated learning model. Therefore, the information processing apparatus 1 can accurately determine whether or not the component corresponding to the inspection image contains a defect.

[Details of the first embodiment]
Next, the details of the first embodiment will be described. 7 to 13 are flowcharts illustrating the details of the image inspection process according to the first embodiment. 14 to 20 are views for explaining the details of the image inspection process according to the first embodiment.

[Image storage processing]
First, a process for accumulating the learning image 131 (hereinafter, also referred to as an image accumulating process) will be described. FIG. 7 is a flowchart illustrating the image storage process.

As shown in FIG. 7, the information receiving unit 111 of the information processing device 1 waits until the learning image 131 is received (NO in S11). Specifically, the information receiving unit 111 waits until, for example, the learning image 131 is transmitted from the operation terminal 5.

Then, when the learning image 131 is received (YES in S11), the information management unit 112 of the information processing device 1 stores the learning image 131 received in the process of S11 in the information storage area 130 (S12).

[Machine learning process]
Next, a process for performing machine learning of the learning image 131 (hereinafter, also referred to as a machine learning process) will be described. 8 to 11 are flowcharts illustrating the machine learning process.

As shown in FIG. 8, the information management unit 112 waits until the machine learning execution timing (NO in S21).

Then, when the machine learning execution timing is reached (YES in S21), the divided image generation unit 113 divides the plurality of learning images 131 stored in the information storage area 130 into each of the first sizes, so that the plurality of first images 131 are divided. The divided image 132 is generated (S22). Specifically, the divided image generation unit 113 divides at least a part of the plurality of learning images 131 stored in the information storage area 130. Hereinafter, a specific example of the processing of S22 will be described.

[Specific example of processing of S22]
FIG. 14 is a diagram illustrating a specific example of the process of S22. FIG. 14 (A) is a diagram illustrating a specific example when processing S22 with respect to the learning image IM31 included in the learning image 131, and FIG. 14 (B) is a diagram showing a learning image included in the learning image 131. It is a figure explaining the specific example at the time of performing the process of S22 with respect to IM32.

Specifically, as shown in FIG. 14A, for example, the divided image generation unit 113 divides the learning image IM31 into windows size WS31 (first size) to divide a plurality of first divided images 132. Generate. Further, the divided image generation unit 113 generates a plurality of first divided images 132 by dividing the learning image IM 32 for each window size WS31, for example, as shown in FIG. 14 (B).

Returning to FIG. 8, the divided image generation unit 113 generates a plurality of second divided images 133 by dividing the plurality of learning images 131 stored in the information storage area 130 for each second size (S23). Specifically, the divided image generation unit 113 divides at least a part of the plurality of learning images 131 stored in the information storage area 130. Hereinafter, a specific example of the processing of S23 will be described.

[Specific example of processing of S23]
FIG. 15 is a diagram illustrating a specific example of the process of S23. FIG. 15A is a diagram illustrating a specific example when the processing of S23 is performed on the learning image IM31 included in the learning image 131, and FIG. 15B is a diagram showing a learning image included in the learning image 131. It is a figure explaining the specific example at the time of performing the process of S23 with respect to IM32.

Specifically, as shown in FIG. 15A, for example, the divided image generation unit 113 divides the learning image IM31 into windows size WS32 (second size) to divide a plurality of second divided images 133. Generate. Further, the divided image generation unit 113 generates a plurality of second divided images 133 by dividing the learning image IM 32 for each window size WS 32, for example, as shown in FIG. 15 (B).

Returning to FIG. 8, the feature amount extraction unit 114 of the information processing apparatus 1 identifies a plurality of first normal divided images 132a from the plurality of first divided images 132 generated in the process of S22 (S24).

Specifically, the feature amount extraction unit 114 refers to, for example, the state information added by the business operator for each of the plurality of first divided images 132 generated in the process of S22, and the plurality of firsts generated in the process of S22. Among the divided images 132, the image associated with the state information indicating that each image is in the normal state is specified as the first normal divided image 132a.

Then, the feature amount extraction unit 114 identifies the plurality of second normal divided images 133a from the plurality of second divided images 133 generated in the process of S23 (S25).

Specifically, the feature amount extraction unit 114 refers to, for example, the state information added by the business operator for each of the plurality of second divided images 133 generated in the process of S23, and the plurality of second images generated in the process of S23. Among the divided images 133, the image associated with the state information indicating that each image is in the normal state is specified as the second normal divided image 133a.

In addition, for example, the business operator adds state information indicating that each image is in a normal state to an image that does not include a defect of a part among a plurality of first divided images 132 generated in the process of S22. It may be a thing. Further, for example, the business operator indicates that, among the plurality of first divided images 132 generated in the process of S22, each image is in an abnormal state with respect to an image including at least a part of a defect of a part. May be added.

Subsequently, as shown in FIG. 9, the feature amount extraction unit 114 sets a plurality of types of parameters from each of the plurality of first normal division images 132a and the plurality of second normal division images 133a generated by the processing of S24 or the like. The corresponding feature amount is extracted (S31).

Specifically, the feature amount extraction unit 114 extracts, for example, the feature amounts corresponding to the average brightness, the luminance dispersion, and the like of the first divided image 132 and the second divided image 133.

Then, the parameter specifying unit 115 of the information processing apparatus 1 generates a distribution for the feature amounts of the plurality of first normal division images 132a extracted in the process of S31 for each parameter for which the feature amount was extracted in the process of S31. (S32).

In addition, the parameter specifying unit 115 generates a distribution for the feature amounts of the plurality of second normally divided images 133a extracted in the process of S31 for each parameter for which the feature amount was extracted in the process of S31 (S33).

After that, the parameter specifying unit 115 calculates the difference between the shape of the distribution generated in the processing of S32 and the shape of the distribution generated in the processing of S33 for each parameter for which the feature amount has been extracted in the processing of S31 (S34). ).

Then, the parameter specifying unit 115 specifies as the first parameter a parameter corresponding to the feature amount for which the difference calculated in the processing of S34 satisfies a predetermined condition among the parameters obtained by extracting the feature amount in the processing of S31 (S35). ). Hereinafter, details of the processing of S34 and S35 will be described.

[Details of processing of S34 and S35]
FIG. 11 is a flowchart illustrating the details of the processes of S34 and S35.

As shown in FIG. 11, the parameter specifying unit 115 calculates the basic statistic of the distribution generated in the process of S32 for each parameter for which the feature amount has been extracted in the process of S31 (S101).

Specifically, the parameter specifying unit 115, for example, when the parameter whose feature amount is extracted in the process of S31 includes the brightness variance, the average of the brightness variance in the distribution generated in the process of S32 (hereinafter, simply average μ1). (Also also referred to as), the standard deviation of the brightness variance in the distribution generated by the processing of S32 (hereinafter, also simply referred to as the standard deviation ρ1), and the skewness of the histogram corresponding to the distribution generated by the processing of S32 (hereinafter, simply the skewness). (Also referred to as s1) and the kurtosis of the histogram corresponding to the distribution generated in the process of S32 (hereinafter, also simply referred to as kurtosis k1) are specified as basic statistics corresponding to the brightness variance.

Further, the parameter specifying unit 115 calculates the basic statistic of the distribution generated in the process of S33 for each parameter for which the feature amount has been extracted in the process of S31 (S102).

Specifically, the parameter specifying unit 115, for example, when the parameter whose feature amount is extracted in the process of S31 includes the brightness variance, the average of the brightness variance in the distribution generated in the process of S33 (hereinafter, simply average μ2). (Also also referred to as), the standard deviation of the brightness variance in the distribution generated by the processing of S33 (hereinafter, also simply referred to as the standard deviation ρ2), and the skewness of the histogram corresponding to the distribution generated by the processing of S33 (hereinafter, simply the skewness). (Also also referred to as s2) and the kurtosis of the histogram corresponding to the distribution generated in the process of S33 (hereinafter, also simply referred to as kurtosis k2) are specified as basic statistics corresponding to the brightness variance.

Then, the parameter specifying unit 115 generated the distribution generated in the processing of S32 and the processing of S33 from the basic statistics calculated in the processing of S101 and S102 for each parameter for which the feature amount was extracted in the processing of S31. The degree of shape similarity with the distribution is calculated (S103).

Specifically, the parameter specifying unit 115 calculates the degree of similarity by using, for example, the following equation (1).
Similarity = | μ1-μ2 | + | ρ1-ρ2 | + | s1-s2 | + | k1-k2 | (Equation 1)

Subsequently, the parameter specifying unit 115 specifies N parameters in descending order of similarity calculated in the process of S103 (S104).

Specifically, the parameter specifying unit 115 calculates N by using, for example, the following equations (2) and (3). The M in the following equations (2) and (3) is, for example, the learning image 131 used when generating the first divided image 132 in the processing of S22 (generating the second divided image 133 in the processing of S23). It may be the number of data of the learning image 131) used at the time.
N = M / 10 + 2 (M <80) (Equation 2)
N = 10 (M ≧ 80) (Equation 3)

After that, the parameter specifying unit 115 specifies N parameters specified in the process of S104 as the first parameter (S105).

In the processing of S35, the parameter specifying unit 115 includes, for example, the center of the distribution generated in the processing of S32 (for example, the median value, the average value or the mode value of the distribution) and the center of the distribution generated in the processing of S33. The parameter corresponding to the feature amount whose difference is equal to or less than the threshold value may be specified as the first parameter. Further, the parameter specifying unit 115 specifies, for example, a parameter corresponding to the feature amount in which the difference between the width of the distribution generated in the process of S32 and the width of the distribution generated in the process of S33 is equal to or less than the threshold value as the first parameter. It may be something to do. Hereinafter, a specific example of the processing of S34 and S35 when the difference in the center value of the distribution is used will be described.

[Specific example of processing of S34 and S35]
16 and 17 are diagrams for explaining specific examples of the processes of S34 and S35. FIG. 16A is a specific example for explaining the distribution of the first normal division image 132a when the average luminance is adopted as a parameter, and FIG. 16B is a second when the average luminance is adopted as a parameter. This is a specific example for explaining the distribution of the normally divided image 133a. Further, FIG. 17A is a specific example for explaining the distribution of the first normally divided image 132a when the luminance dispersion is adopted as the parameter, and FIG. 17B is the case where the luminance variance is adopted as the parameter. It is a specific example explaining the distribution of the 2nd normal division image 133a.

Specifically, in the example shown in FIG. 16, the median value in the histogram shown on the right side of FIG. 16 (A) coincides with the median value in the histogram shown on the right side of FIG. 16 (B). Therefore, in this case, the parameter specifying unit 115 determines, for example, that the difference between the median distribution of the first normally divided image 132a and the median of the second normally divided image 133a is equal to or less than the threshold value, and determines the average brightness. Specify as one parameter.

On the other hand, in the example shown in FIG. 17, the median value in the histogram shown on the right side of FIG. 17 (A) does not match the median value in the histogram shown on the right side of FIG. 17 (B). Therefore, when the parameter specifying unit 115 determines that the difference between the median distribution of the first divided image 132 and the median of the second divided image 133 exceeds the threshold value, for example, the luminance variance is used as the first parameter. Make a decision not to specify.

In the process of S31, the feature amount extraction unit 114 may extract the feature amount corresponding to the luminance histogram or the gradient histogram of the first divided image 132 or the second divided learning image 132, for example. Then, the parameter specifying unit 115 may generate a distribution for the feature amount corresponding to the luminance histogram or the gradient histogram in the processing of S32 and S33, for example. Hereinafter, a specific example of the distribution corresponding to the luminance histogram will be described.

[Specific example of distribution corresponding to luminance histogram]
FIG. 18 is a diagram illustrating a specific example of the distribution corresponding to the luminance histogram. FIG. 18A is a specific example of the luminance histogram for each of the first normal division images 132a, and FIGS. 18B and 18C are distributions corresponding to the luminance histogram shown in FIG. 18A. This is a specific example.

Specifically, the luminance histogram shown in FIG. 18 (A) shows, for example, a section showing the number of pixels having a brightness value between 0 and 63 (hereinafter, also referred to as a first section) and a luminance value of 64 to 127. A section indicating the number of pixels between the two (hereinafter, also referred to as a second section), a section indicating the number of pixels having a brightness value between 128 and 191 (hereinafter, also referred to as a third section), and a brightness value. Includes a section (hereinafter, also referred to as a fourth section) indicating the number of pixels in which is between 192 and 255.

Then, in the processing of S31, the feature amount extraction unit 114 sets the number of pixels corresponding to each of the first section, the second section, the third section, and the fourth section for each first normal division image 132a as a luminance histogram. The feature amount corresponding to is extracted.

After that, in the process of S32, the parameter specifying unit 115 aggregates the number of the first normal division image 132a corresponding to each of the number of pixels in the first section, for example, as shown in FIG. 18B. A distribution for the luminance histogram of the first normally divided image 132a is generated. Further, in the process of S33, the parameter specifying unit 115 obtains the luminance histogram of the second normally divided image 133a by, for example, totaling the number of the second normally divided images 133a corresponding to each of the number of pixels in the first section. Generates a distribution of (not shown).

Then, in the processing of S34 and S35, the parameter specifying unit 115 uses the distribution of the brightness histogram of the first normally divided image 132a and the distribution of the brightness histogram of the second normally divided image 133a to obtain the first parameter. To identify.

In the process of S31, the parameter specifying unit 115 counts the number of the first normally divided images 132a corresponding to each of the number of pixels in the fourth section, for example, as shown in FIG. 18C. The distribution of the luminance histogram of the first normally divided image 132a may also be generated.

Then, in the processing of S34 and S35, the parameter specifying unit 115 specifies the first parameter by using the distribution of the luminance histogram corresponding to the plurality of sections (for example, the first section and the fourth section), respectively. It may be.

Specifically, the parameter specifying unit 115 may, for example, calculate the difference in shape for each of the distributions corresponding to the plurality of sections. Then, the parameter specifying unit 115 may specify the luminance histogram as the first parameter when, for example, at least one of the differences in shape corresponding to the plurality of sections satisfies a predetermined condition. Further, in this case, the parameter specifying unit 115 may specify, for example, only the luminance histogram corresponding to the section where the difference in shape satisfies the predetermined condition as the first parameter among the luminance histograms.

[Specific example of distribution corresponding to gradient histogram]
Next, a specific example of the distribution corresponding to the gradient histogram will be described. FIG. 19 is a diagram illustrating a specific example of the distribution corresponding to the gradient histogram. FIG. 19 (A) is a specific example of the gradient histogram for each of the first normal division images 132a, and FIGS. 19 (B) and 19 (C) show distributions corresponding to the gradient histogram shown in FIG. 19 (A). This is a specific example.

Specifically, the luminance histogram shown in FIG. 19A shows a section showing the number of pixels in which the luminance gradient is between 0 ° and 45 ° (hereinafter, also referred to as the first section) and a luminance gradient from 46 °. A section showing the number of pixels between 90 ° (hereinafter, also referred to as a second section) and a section showing the number of pixels having a brightness gradient between 91 ° and 135 ° (hereinafter, also referred to as a third section). And a section (hereinafter, also referred to as a fourth section) indicating the number of pixels whose luminance gradient is between 136 ° and 180 °.

Then, in the processing of S31, the feature amount extraction unit 114 sets the number of pixels corresponding to each of the first section, the second section, the third section, and the fourth section for each first normal division image 132a as a gradient histogram. The feature amount corresponding to is extracted.

After that, in the process of S32, the parameter specifying unit 115 aggregates the number of the first normal division image 132a corresponding to each of the number of pixels in the first section, for example, as shown in FIG. 19B. A distribution for the gradient histogram of the first normally divided image 132a is generated. Further, in the process of S33, the parameter specifying unit 115 obtains the gradient histogram of the second normally divided image 133a by, for example, totaling the number of the second normally divided images 133a corresponding to each of the number of pixels in the first section. Generates a distribution of (not shown).

Then, in the processing of S34 and S35, the parameter specifying unit 115 uses the distribution of the gradient histogram of the first normally divided image 132a and the distribution of the gradient histogram of the second normally divided image 133a to obtain the first parameter. To identify.

In the process of S31, the parameter specifying unit 115 counts the number of the first normally divided images 132a corresponding to each of the number of pixels in the third section, for example, as shown in FIG. 19C. The distribution of the gradient histogram of the first normally divided image 132a may also be generated.

Then, the parameter specifying unit 115 uses the distribution for the gradient histogram corresponding to the plurality of partitions (for example, the first partition and the third partition) in the processing of S34 and S35, respectively, as in the case described with reference to FIG. May be used to specify the first parameter.

Returning to FIG. 10, the feature amount extraction unit 114 identifies a plurality of first abnormally divided images 132b from the plurality of first divided images 132 generated in the process of S22 (S41). In addition, the feature amount extraction unit 114 identifies a plurality of second abnormally divided images 133b from the plurality of second divided images 133 generated in the process of S23 (S42).

Then, the feature amount extraction unit 114 extracts the feature amount corresponding to the first parameter specified in the process of S35 from the plurality of first abnormally divided images 132b and the plurality of second abnormally divided images 133b specified in the process of S41 or the like. Extract (S43).

Subsequently, the information management unit 112 includes a plurality of learning data in which the feature amount corresponding to the first parameter extracted in the process of S31 and the state information indicating that each image is in the normal state are associated with each other, and S43. A plurality of learning data in which the feature amount corresponding to the first parameter extracted in the process and the state information indicating that each image is in an abnormal state are associated with each other are generated (S44).

That is, the information management unit 112 brings the feature amounts corresponding to the first parameters extracted from each of the plurality of first normal division images 132a and the plurality of second normal division images 133a in the processing of S31, and each image into a normal state. Generate a plurality of training data associated with the state information indicating that there is. In addition, the information management unit 112 sets the feature amount corresponding to the first parameter extracted from each of the plurality of first abnormally divided images 132b and the plurality of second abnormally divided images 133b in the processing of S43, and each image into an abnormal state. A plurality of learning data associated with the state information indicating the existence are generated.

After that, the machine learning execution unit 116 of the information processing apparatus 1 generates the learning model 134 by performing machine learning using the plurality of learning data generated in the process of S44 (S45). Then, the information processing device 1 ends the machine learning process.

That is, the information processing apparatus 1 specifies the first parameter corresponding to the feature amount that does not depend on the size of the defect, among the parameters that can extract the feature amount from the first divided image 132 and the second divided image 133. .. Then, the information processing device 1 performs machine learning by using the feature amount corresponding to the specified first parameter.

As a result, the information processing apparatus 1 can accurately identify defects of a size for which a sufficient number of training images 131 have not been prepared, as shown in FIG. 20, by using the generated learning model 134. It will be possible. Therefore, the information processing apparatus 1 can accurately determine whether or not the component corresponding to the inspection image contains a defect.

In the processing of S34 and S35, for example, the parameter specifying unit 115 specifies the number of samples for each section in the histogram of the distribution corresponding to each parameter for each parameter for which the feature amount is extracted in the processing of S31. It may create a vector of the number of samples whose dimension is the number of partitions. Then, the parameter specifying unit 115 may specify the first parameter by, for example, calculating the cosine similarity between the vectors corresponding to each parameter. Further, the parameter specifying unit 115 may specify the first parameter by, for example, calculating the similarity between the vectors corresponding to each parameter by the histogram intersection method.

[Abnormality identification processing]
Next, a process for identifying an inspection image of a part containing a defect (hereinafter, also referred to as an abnormality identification process) will be described. 12 and 13 are flowcharts illustrating the abnormality identification process.

The information receiving unit 111 waits until the inspection image (new image) is received (NO in S51). Specifically, the information receiving unit 111 waits until the inspection image is transmitted from the operation terminal 5.

Then, when the inspection image is received (YES in S51), the divided image generation unit 113 generates the divided image by dividing the inspection image received in the process of S51 into predetermined sizes (S52). In this case, the divided image generation unit 113 may use, for example, the geometric mean of the first size used in the processing of S24 and the second size used in the processing of S25 as a predetermined size.

Subsequently, the feature amount extraction unit 114 extracts the feature amount corresponding to the first parameter specified in the process of S35 from the divided image generated in the process of S52 (S53).

After that, the score acquisition unit 117 acquires the score output from the learning model 134 with the input of the feature amount corresponding to the first parameter extracted in the process of S53 for each divided image generated in the process of S52 (. S54).

Then, as shown in FIG. 13, the abnormality determination unit 118 of the information processing apparatus 1 identifies a score below the threshold value from the score acquired in the process of S54 (S61).

As a result, when it is determined that a score equal to or lower than the threshold value exists (YES in S62), the abnormality determination unit 118 abnormalizes the divided image corresponding to the score specified in the processing of S61 among the divided images generated in the processing of S52. It is determined that the state is present (S63).

Further, in this case, the abnormality determination unit 118 determines that the inspection image received in the process of S51 is in an abnormal state (S64). That is, the abnormality determination unit 118 determines that the component corresponding to the inspection image received in the process of S51 contains a defect.

After that, the information output unit 119 of the information processing device 1 outputs, for example, information indicating that the inspection image received in the process of S51 is in an abnormal state to the operation terminal 5 (S64).

Then, the information processing device 1 ends the abnormality identification process. The information processing device 1 also ends the abnormality identification process in the same manner (NO in S62) even when it is determined that there is no score equal to or lower than the threshold value in the process of S62.

As a result, the information processing apparatus 1 can accurately determine whether or not the component corresponding to the inspection image contains a defect.

The above embodiments can be summarized as follows.

(Appendix 1)
An image inspection system having a plurality of information processing devices that perform machine learning of a plurality of learning images.
Each of the plurality of information processing devices
A plurality of first divided images are generated by dividing the plurality of training images into each first size, and a plurality of second by dividing the plurality of learning images into second sizes different from the first size. A split image generator that generates split images,
From each of the plurality of first normally divided images corresponding to the normal state of the plurality of first divided images and the plurality of second normally divided images corresponding to the normal state of the plurality of second divided images. , A feature amount extraction unit that extracts feature amounts corresponding to multiple types of parameters,
From the plurality of types of parameters, a parameter specifying unit that specifies a first parameter in which the similarity relationship between the plurality of first normal divided images and the plurality of second normally divided images satisfies a predetermined condition. ,
It has a machine learning execution unit that performs machine learning of the plurality of first divided images and the plurality of second divided images by using the feature amount corresponding to the specified first parameter.
An image inspection system characterized by that.

(Appendix 2)
In Appendix 1,
The feature amount extraction unit includes a plurality of first abnormally divided images corresponding to the abnormal state of the plurality of first divided images, and a plurality of second images corresponding to the abnormal state of the plurality of second divided images. The feature quantities corresponding to the first parameter are extracted from each of the abnormally divided images.
The machine learning execution unit indicates that the feature quantities corresponding to the first parameter extracted from each of the plurality of first normal division images and the plurality of second normal division images and that each image is in a normal state. Each image includes a plurality of learning data associated with state information, a feature amount corresponding to the first parameter extracted from each of the plurality of first abnormally divided images and the plurality of second abnormally divided images, and each image. Machine learning is performed with a plurality of learning data associated with state information indicating that the state is abnormal.
An image inspection system characterized by that.

(Appendix 3)
In Appendix 1,
The parameter specifying unit specifies as the first parameter a parameter corresponding to a feature amount in which the difference in the shape of distribution between the plurality of first normal divided images and the plurality of second normally divided images satisfies a predetermined condition. To do,
An image inspection system characterized by that.

(Appendix 4)
In Appendix 3,
The parameter specifying unit specifies as the first parameter a parameter corresponding to a feature amount in which the difference in the center of distribution between the plurality of first normal divided images and the plurality of second normally divided images is equal to or less than a threshold value. To do,
An image inspection system characterized by that.

(Appendix 5)
In Appendix 3,
The parameter specifying unit specifies as the first parameter a parameter corresponding to a feature amount in which the difference in the width of the distribution between the plurality of first normal divided images and the plurality of second normally divided images is equal to or less than a threshold value. To do,
An image inspection system characterized by that.

(Appendix 6)
In Appendix 2,
The machine learning execution unit generates a learning model by performing the machine learning,
When a new image is received, the divided image generation unit generates a plurality of divided images by dividing the new image into predetermined sizes.
The feature amount extraction unit extracts each feature amount corresponding to the first parameter from the plurality of divided images, and further
Each of the plurality of information processing devices
A score acquisition unit that acquires a score output from the learning model with the input of a feature amount corresponding to the first parameter for each of the plurality of divided images.
Among the plurality of divided images, the divided image having the acquired score equal to or lower than the threshold value has an abnormality determining unit for determining that the divided image is in an abnormal state.
An image inspection system characterized by that.

(Appendix 7)
In Appendix 6, further
Each of the plurality of information processing devices
When it is determined that the divided image in the abnormal state is included in the plurality of divided images, it has an information output unit that outputs information indicating that the new image is in the abnormal state.
An image inspection system characterized by that.

(Appendix 8)
A plurality of first divided images are generated by dividing a plurality of training images into each first size, and a plurality of second divisions are performed by dividing the plurality of training images into second sizes different from the first size. A split image generator that generates an image and
From each of the plurality of first normally divided images corresponding to the normal state of the plurality of first divided images and the plurality of second normally divided images corresponding to the normal state of the plurality of second divided images. , A feature amount extraction unit that extracts feature amounts corresponding to multiple types of parameters,
From the plurality of types of parameters, a parameter specifying unit that specifies a first parameter in which the similarity relationship between the plurality of first normal divided images and the plurality of second normally divided images satisfies a predetermined condition. ,
It has a machine learning execution unit that performs machine learning of the plurality of first divided images and the plurality of second divided images by using the feature amount corresponding to the specified first parameter.
An image inspection device characterized in that.

(Appendix 9)
In Appendix 8,
The feature amount extraction unit includes a plurality of first abnormally divided images corresponding to the abnormal state of the plurality of first divided images, and a plurality of second images corresponding to the abnormal state of the plurality of second divided images. The feature quantities corresponding to the first parameter are extracted from each of the abnormally divided images.
The machine learning execution unit indicates that the feature quantities corresponding to the first parameter extracted from each of the plurality of first normal division images and the plurality of second normal division images and that each image is in a normal state. Each image includes a plurality of learning data associated with state information, a feature amount corresponding to the first parameter extracted from each of the plurality of first abnormally divided images and the plurality of second abnormally divided images, and each image. Machine learning is performed with a plurality of learning data associated with state information indicating that the state is abnormal.
An image inspection device characterized in that.

(Appendix 10)
In Appendix 9,
The machine learning execution unit generates a learning model by performing the machine learning,
When a new image is received, the divided image generation unit generates a plurality of divided images by dividing the new image into predetermined sizes.
The feature amount extraction unit extracts each feature amount corresponding to the first parameter from the plurality of divided images, and further
A score acquisition unit that acquires a score output from the learning model with the input of a feature amount corresponding to the first parameter for each of the plurality of divided images.
Among the plurality of divided images, the divided image having the acquired score equal to or lower than the threshold value has an abnormality determining unit for determining that the divided image is in an abnormal state.
An image inspection device characterized in that.

(Appendix 11)
A plurality of first divided images are generated by dividing a plurality of training images into each first size, and a plurality of second divisions are performed by dividing the plurality of training images into second sizes different from the first size. Generate an image and
From each of the plurality of first normally divided images corresponding to the normal state of the plurality of first divided images and the plurality of second normally divided images corresponding to the normal state of the plurality of second divided images. , Extract the features corresponding to multiple types of parameters
From the plurality of types of parameters, a first parameter is specified in which the similarity relationship between the plurality of first normal division images and the plurality of second normal division images satisfies a predetermined condition.
By using the feature quantity corresponding to the specified first parameter, machine learning of the plurality of first divided images and the plurality of second divided images is performed.
An image inspection program characterized by having a computer perform processing.

(Appendix 12)
In Appendix 11, further
From each of the plurality of first abnormally divided images corresponding to the abnormal state of the plurality of first divided images and the plurality of second abnormally divided images corresponding to the abnormal state of the plurality of second divided images. , Each feature quantity corresponding to the first parameter is extracted.
Let the computer do the work
In the process of performing machine learning, the feature amount corresponding to the first parameter extracted from each of the plurality of first normal division images and the plurality of second normal division images, and the fact that each image is in a normal state. A plurality of learning data associated with the indicated state information, a feature amount corresponding to the first parameter extracted from each of the plurality of first abnormally divided images and the plurality of second abnormally divided images, and each image. Performs machine learning with a plurality of training data associated with state information indicating that is in an abnormal state.
An image inspection program characterized by that.

(Appendix 13)
In Appendix 12,
In the process of performing machine learning, a learning model is generated by performing the machine learning, and further,
When a new image is received, a plurality of divided images are generated by dividing the new image into predetermined sizes.
The feature quantities corresponding to the first parameter are extracted from the plurality of divided images, respectively.
For each of the plurality of divided images, a score output from the learning model is acquired with the input of the feature amount corresponding to the first parameter.
Among the plurality of divided images, it is determined that the divided image whose score is equal to or less than the threshold value is in an abnormal state.
An image inspection program characterized by having a computer perform processing.

1: Information processing device 5: Operation terminal NW: Network 10: Information processing system

Claims (9)

An image inspection system having a plurality of information processing devices that perform machine learning of a plurality of learning images.
Each of the plurality of information processing devices
A plurality of first divided images are generated by dividing the plurality of training images into each first size, and a plurality of second by dividing the plurality of learning images into second sizes different from the first size. A split image generator that generates split images,
From each of the plurality of first normally divided images corresponding to the normal state of the plurality of first divided images and the plurality of second normally divided images corresponding to the normal state of the plurality of second divided images. , A feature amount extraction unit that extracts feature amounts corresponding to multiple types of parameters,
From the plurality of types of parameters, a parameter specifying unit that specifies a first parameter in which the similarity relationship between the plurality of first normal divided images and the plurality of second normally divided images satisfies a predetermined condition. ,
It has a machine learning execution unit that performs machine learning of the plurality of first divided images and the plurality of second divided images by using the feature amount corresponding to the specified first parameter.
An image inspection system characterized by that. In claim 1,
The feature amount extraction unit includes a plurality of first abnormally divided images corresponding to the abnormal state of the plurality of first divided images, and a plurality of second images corresponding to the abnormal state of the plurality of second divided images. The feature quantities corresponding to the first parameter are extracted from each of the abnormally divided images.
The machine learning execution unit indicates that the feature quantities corresponding to the first parameter extracted from each of the plurality of first normal division images and the plurality of second normal division images and that each image is in a normal state. Each image includes a plurality of learning data associated with state information, a feature amount corresponding to the first parameter extracted from each of the plurality of first abnormally divided images and the plurality of second abnormally divided images, and each image. Machine learning is performed with a plurality of learning data associated with state information indicating that the state is abnormal.
An image inspection system characterized by that. In claim 1,
The parameter specifying unit specifies as the first parameter a parameter corresponding to a feature amount in which the difference in the shape of distribution between the plurality of first normal divided images and the plurality of second normally divided images satisfies a predetermined condition. To do,
An image inspection system characterized by that. In claim 3,
The parameter specifying unit specifies as the first parameter a parameter corresponding to a feature amount in which the difference in the center of distribution between the plurality of first normal divided images and the plurality of second normally divided images is equal to or less than a threshold value. To do,
An image inspection system characterized by that. In claim 3,
The parameter specifying unit specifies as the first parameter a parameter corresponding to a feature amount in which the difference in the width of the distribution between the plurality of first normal divided images and the plurality of second normally divided images is equal to or less than a threshold value. To do,
An image inspection system characterized by that. In claim 2,
The machine learning execution unit generates a learning model by performing the machine learning,
When a new image is received, the divided image generation unit generates a plurality of divided images by dividing the new image into predetermined sizes.
The feature amount extraction unit extracts each feature amount corresponding to the first parameter from the plurality of divided images, and further
Each of the plurality of information processing devices
A score acquisition unit that acquires a score output from the learning model in response to input of a feature amount corresponding to the first parameter for each of the plurality of divided images.
Among the plurality of divided images, the divided image having the acquired score equal to or lower than the threshold value has an abnormality determining unit for determining that the divided image is in an abnormal state.
An image inspection system characterized by that. In claim 6, further
Each of the plurality of information processing devices
When it is determined that the divided image in the abnormal state is included in the plurality of divided images, it has an information output unit that outputs information indicating that the new image is in the abnormal state.
An image inspection system characterized by that. A plurality of first divided images are generated by dividing a plurality of training images into each first size, and a plurality of second divisions are performed by dividing the plurality of training images into second sizes different from the first size. A split image generator that generates an image and
From each of the plurality of first normally divided images corresponding to the normal state of the plurality of first divided images and the plurality of second normally divided images corresponding to the normal state of the plurality of second divided images. , A feature amount extraction unit that extracts feature amounts corresponding to multiple types of parameters,
From the plurality of types of parameters, a parameter specifying unit that specifies a first parameter in which the similarity relationship between the plurality of first normal divided images and the plurality of second normally divided images satisfies a predetermined condition. ,
It has a machine learning execution unit that performs machine learning of the plurality of first divided images and the plurality of second divided images by using the feature amount corresponding to the specified first parameter.
An image inspection device characterized in that. A plurality of first divided images are generated by dividing a plurality of training images into each first size, and a plurality of second divisions are performed by dividing the plurality of training images into second sizes different from the first size. Generate an image and
From each of the plurality of first normally divided images corresponding to the normal state of the plurality of first divided images and the plurality of second normally divided images corresponding to the normal state of the plurality of second divided images. , Extract the features corresponding to multiple types of parameters
From the plurality of types of parameters, a first parameter is specified in which the similarity relationship between the plurality of first normal division images and the plurality of second normal division images satisfies a predetermined condition.
Machine learning of the plurality of first-divided images and the plurality of second-divided images is performed by using the feature quantities corresponding to the specified first parameter.
An image inspection program characterized by having a computer perform processing.

Images