Learning program, learning method and learning device

Abstract

To provide a learning program, learning method and learning device capable of improving accuracy of a learned model generated by machine learning using pseudo images.SOLUTION: A learning program causes a computer to execute: an image acquisition step of acquiring a captured non-defective product image and a pseudo-generated defective image including a defective portion; a feature acquisition step of acquiring a first feature of the non-defective product image and a second feature of a region other than the defective portion in the defective image; a determination step of determining whether the first feature and the second feature acquired in the feature acquisition step are similar; and a learning step of executing machine learning of a learning model using the defective image having the second feature when it is determined in the determination step that the first feature and the second feature are similar.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to a learning program, a learning method, and a learning device. [Background technology]

[0002] Patent Document 1 discloses a learning data generation device that includes a defective part data storage unit that stores defective part data, which is image data of defective part images; a good product data storage unit that stores good product data, which is image data of good product images; a learning data generation unit that generates learning data, which is image data of learning images in which good product images are combined with defective part images, based on the defective part data, good product data, and generation parameters; a learning data storage unit that stores the learning data; and a generation parameter setting unit that sets the generation parameters. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2020-27424 Summary of the Invention [Problem to be solved by the invention]

[0004] One method for inspecting an item involves inputting an image containing the item into a trained model generated by machine learning, and inspecting the item according to the prediction results output from the trained model. In order to improve prediction accuracy, it is preferable to generate a trained model using a large amount of training data. However, in an actual inspection process, defective products are not often generated, making it difficult to obtain many defective images (actual images) of defective products. Therefore, it takes time to obtain many defective images, and therefore it takes time to generate a trained model. Therefore, as described above, pseudo images of defective images are generated, and machine learning of the training model is performed using the pseudo images as training data.

[0005] However, the quality (features) of the generated pseudo-images are not necessarily equivalent to the quality of the real images actually captured. If the quality of the pseudo-images differs from the quality of the real images, the accuracy of the trained model will not improve even if machine learning of the training model is performed using the pseudo-images.

[0006] The present disclosure aims to provide a learning program, a learning method, and a learning device that can improve the accuracy of a trained model generated by machine learning using pseudo images. [Means for solving the problem]

[0007] A learning program according to one embodiment of the present disclosure is a learning program that is executed by at least one processor and performs machine learning of a learning model, and causes a computer to execute the following steps: an image acquisition step of acquiring photographed images of good products and pseudo-generated images of defective products that include defective parts; a feature acquisition step of acquiring first features of the good product images and second features of areas of the defective images other than the defective parts; a determination step of determining whether the first feature and the second feature acquired in the feature acquisition step are similar; and a learning step of performing machine learning of a learning model using defective images having the second feature if it is determined in the determination step that the first feature and the second feature are similar. [Effects of the Invention]

[0008] According to the present disclosure, it is possible to improve the accuracy of a trained model generated by machine learning using pseudo images. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram illustrating an example of the hardware configuration of a computer used in a prediction system according to an embodiment. [Figure 2] FIG. 2 is a diagram illustrating an example of a functional configuration of the prediction system. [Figure 3] FIG. 3 is a diagram illustrating an example of the feature acquisition process in the feature acquisition unit. [Figure 4] FIG. 4 is a diagram illustrating an example of a neural network. [Figure 5] FIG. 5 is a diagram showing an example of distribution in the feature space. [Figure 6] FIG. 6 is a flowchart showing an example of a learning method and generation of a trained model as a processing flow. [Figure 7] FIG. 7 is a flowchart showing an example of a prediction method as a processing flow. DETAILED DESCRIPTION OF THE INVENTION

[0010] [Description of the embodiments of the present disclosure] First, the contents of the embodiments of the present disclosure will be listed and described. At least some of the embodiments described below may be combined in any combination.

[0011] (1) A learning program according to one aspect of the present disclosure is a learning program that is executed by at least one processor and performs machine learning of a learning model, and causes a computer to execute the following steps: an image acquisition step of acquiring photographed images of good products and pseudo-generated images of defective products that include defective parts; a feature acquisition step of acquiring first features of the images of good products and second features of areas of the defective images other than the defective parts; a determination step of determining whether the first feature and the second feature acquired in the feature acquisition step are similar; and a learning step of performing machine learning of a learning model using defective images that have the second feature if it is determined in the determination step that the first feature and the second feature are similar.

[0012] A learning program according to one aspect of the present disclosure acquires a first feature of a good product image and a second feature of a defective product image, and determines whether the first feature and the second feature are similar. If the learning program determines that the first feature and the second feature are similar, it performs machine learning of a learning model using a defective product image having the second feature. In this way, the learning program performs machine learning of a learning model using a defective product image (pseudo image) in which the features of an area other than the defective part (the good product area) are similar to those of an actually captured good product image (real image). In other words, the learning program does not use defective images with different features from good product images in the machine learning of the learning model. Therefore, the learning program can perform machine learning of a learning model using defective images with the same quality as good product images, thereby improving the accuracy of a trained model generated by machine learning. Therefore, the learning program can improve the accuracy of a trained model generated by machine learning using pseudo images.

[0013] (2) In the learning program of (1) above, in the feature acquisition step, the defective part of the defective image may be masked, and the features of the area other than the defective part in the defective image may be acquired. With this configuration, the features of the area other than the defective part in the defective image can be acquired more reliably.

[0014] (3) In the learning program of (1) or (2) above, the feature acquisition step may involve inputting a non-defective image and a defective image to an input layer of a neural network including an input layer, one or more intermediate layers, and an output layer, and acquiring the first feature and the second feature. In this configuration, the first feature and the second feature can be acquired by the neural network.

[0015] (4) In the learning program of (3) above, in the feature acquisition step, the output of the intermediate layer or the output layer may be acquired as the first feature and the second feature. The output of the intermediate layer and the output layer have different features, such as different feature dimensions. Therefore, the features to be acquired can be selected depending on the desired performance.

[0016] (5) In the learning program of (1) or (2) above, the feature acquisition step may involve performing image processing on the non-defective image and the defective image to acquire the first feature and the second feature. In this configuration, the first feature and the second feature can be acquired by image processing.

[0017] (6) A learning method according to one aspect of the present disclosure is a learning method executed by at least one processor and performs machine learning of a learning model, and includes: an image acquisition step of acquiring photographed images of good products and pseudo-generated images of defective products that include defective parts; a feature acquisition step of acquiring first features of the images of good products and second features of areas of the defective images other than the defective parts; a determination step of determining whether the first and second features acquired in the feature acquisition step are similar; and a learning step of performing machine learning of a learning model using defective images that have the second features if it is determined in the determination step that the first and second features are similar.

[0018] In a learning method according to one aspect of the present disclosure, a first feature of a good product image and a second feature of a defective product image are acquired, and it is determined whether the first feature and the second feature are similar. If it is determined that the first feature and the second feature are similar, the learning method performs machine learning of a learning model using a defective product image having the second feature. In this way, the learning method performs machine learning of a learning model using a defective product image (pseudo image) in which the features of an area other than the defective part (the good product area) are similar to those of an actually captured good product image (real image). In other words, the learning method does not use defective images with different features from the good product image in the machine learning of the learning model. Therefore, the learning method can perform machine learning of a learning model using defective images of the same quality as good product images, thereby improving the accuracy of a trained model generated by machine learning. Therefore, the learning method improves the accuracy of a trained model generated by machine learning using pseudo images.

[0019] (7) A learning device according to one aspect of the present disclosure is a learning device that includes at least one processor and performs machine learning of a learning model, and includes: an image acquisition unit that acquires photographed images of good products and pseudo-generated images of defective products that include defective parts; a feature acquisition unit that acquires first features of the good product images and second features of areas of the defective images other than the defective parts; a determination unit that determines whether the first feature and the second feature acquired by the feature acquisition unit are similar; and a learning unit that performs machine learning of a learning model using defective images that have the second feature when the determination unit determines that the first feature and the second feature are similar.

[0020] A learning device according to one aspect of the present disclosure acquires a first feature of a good product image and a second feature of a defective product image and determines whether the first feature and the second feature are similar. If the learning device determines that the first feature and the second feature are similar, it performs machine learning of a learning model using a defective product image having the second feature. In this way, the learning device performs machine learning of a learning model using a defective product image (pseudo image) in which the features of an area other than the defective part (the good product area) are similar to those of an actually captured good product image (real image). In other words, the learning device does not use defective images with different features from good product images in the machine learning of the learning model. Therefore, the learning device can perform machine learning of a learning model using defective images of the same quality as good product images, thereby improving the accuracy of the trained model generated by machine learning. Therefore, the learning device can improve the accuracy of the trained model generated by machine learning using pseudo images.

[0021] [Details of the embodiments of the present disclosure] Specific examples of embodiments of the present disclosure will be described below with reference to the drawings. The present disclosure is not limited to these examples, but is defined by the claims, and is intended to include all modifications within the meaning and scope of the claims. In the description of the drawings, the same elements are given the same reference numerals, and duplicate explanations will be omitted.

[0022] [System Overview] The prediction system 1 according to this embodiment is a computer system that predicts the presence or absence of defects in, for example, the inspection of an item. The inspection of the item is, for example, an appearance inspection of the item. In this embodiment, the result predicted by the prediction system 1 is referred to as the "prediction result."

[0023] The prediction system 1 uses machine learning to predict the presence or absence of defects. Machine learning is a technique for autonomously finding laws or rules by iteratively (repeatedly) learning based on given information. The prediction system 1 performs machine learning using a machine learning model. For example, the prediction system 1 may perform machine learning using a gradient boosting decision tree (GBDT), machine learning using multiple regression analysis, or machine learning using a convolutional neural network (CNN) configured to include a convolutional layer and a pooling layer. A convolutional neural network is a type of deep learning that uses a neural network with a multilayer structure.

[0024] The prediction system 1 trains a machine learning model by repeating learning, and acquires this machine learning model as a trained model. This corresponds to the learning phase. In the learning phase, the prediction system 1 functions as a learning device. The trained model is a machine learning model that is predicted to be optimal for predicting the presence or absence of defects. The prediction system 1 processes input data using the trained model and outputs a prediction result of the presence or absence of defects, which corresponds to the operation phase (prediction phase). In the operation phase, the prediction system 1 functions as a prediction device.

[0025] Trained models are portable between computer systems. Therefore, a trained model generated in one computer system can be used in another computer system. Of course, one computer system may execute both the generation and use of trained models. That is, the prediction system 1 may execute both the learning phase and the operation phase, or may not execute either the learning phase or the operation phase. In this embodiment, the prediction system 1 executes both the learning phase and the operation phase.

[0026] In the learning phase, the prediction system 1 uses training data. The training data is image data. The image data may include images of good products and images of defective products. A good product image is an image that does not contain defective parts. A defective image is an image that contains defective parts. The prediction system 1 generates a trained model by performing machine learning using the training data. In the operation phase, the prediction system 1 obtains prediction results by providing input data to the trained model. The input data is image data captured during an inspection process, etc.

[0027] [System Configuration] FIG. 1 is a diagram illustrating an example of a general hardware configuration of a computer 100 constituting the prediction system 1. For example, the computer 100 includes a processor 101, a main memory 102, an auxiliary memory 103, a communication control unit 104, an input device 105, and an output device 106. The processor 101 executes an operating system and application programs. The main memory 102 is composed of, for example, ROM and RAM. The auxiliary memory 103 is composed of, for example, a hard disk or flash memory, and generally stores a larger amount of data than the main memory 102. The communication control unit 104 is composed of, for example, a network card or a wireless communication module. The input device 105 is composed of, for example, a keyboard, a mouse, a touch panel, etc. The output device 106 is composed of, for example, a monitor and a speaker.

[0028] Each functional element of the prediction system 1 is realized by a learning program 110 and a prediction program 120 pre-stored in the auxiliary storage unit 103. Specifically, each functional element is realized by loading the learning program 110 or the prediction program 120 onto the processor 101 or the main storage unit 102 and executing the learning program 110 or the prediction program 120. The processor 101 operates the communication control unit 104, the input device 105, or the output device 106 in accordance with the learning program 110 or the prediction program 120, and reads and writes data from and to the main storage unit 102 or the auxiliary storage unit 103. Data or databases required for processing are stored in the main storage unit 102 or the auxiliary storage unit 103.

[0029] The learning program 110 and the prediction program 120 may be provided by being fixedly recorded on a tangible recording medium such as a CD-ROM, a DVD-ROM, a semiconductor memory, etc. Alternatively, the learning program 110 and the prediction program 120 may be provided via a communication network as a data signal superimposed on a carrier wave.

[0030] The prediction system 1 may be configured with one computer 100 or multiple computers 100. When multiple computers 100 are used, these computers 100 are connected via a communication network such as the Internet or an intranet, thereby logically constructing a single prediction system 1.

[0031] Fig. 2 is a diagram showing an example of the functional configuration of the prediction system 1. As shown in Fig. 2, the prediction system 1 includes an image generation unit 10, an image acquisition unit 11, a feature acquisition unit 12, a determination unit 13, a learning unit 14, a storage unit 15, a prediction unit 16, and a database 20.

[0032] The image generation unit 10 is a component that generates a pseudo-image. The image generation unit 10 can access a database that stores image data. The database 20 can be used to train a learning model. The database 20 may be, for example, a component of the prediction system 1, or may be built in a computer system separate from the prediction system 1. The prediction system 1 and the database 20 may be connected via a communication network, or both the prediction system 1 and the database 20 may be built in a single computer.

[0033] The method for preparing the image data to be stored in the database 20 is not limited. For example, the image data may be stored in the database 20 by an operator's input work, or may be automatically collected and stored in the database 20 by the prediction system 1 or another computer system. The database 20 stores images of good products and images of defective products as image data. The images of good products may include real images of good products actually taken in an inspection process or the like, and pseudo-good images generated by the image generation unit 10. The images of defective products may include real images of defective products actually taken in an inspection process or the like, and pseudo-failure images generated by the image generation unit 10.

[0034] The image generating unit 10 acquires an actual defect image from the database 20 and generates a pseudo-defect image. The image generating unit 10 generates the pseudo-defect image by, for example, changing the appearance of a defective part included in the actual defect image or adding noise. The image generating unit 10 generates a pseudo-defect image based on the actual defect image by changing at least one of the shape, size, position, and color of the defective part. The properties of the defective part, such as the shape, size, position, and color, can be set arbitrarily or can be selected randomly. The image generating unit 10 generates the pseudo-defect image using, for example, a neural network. The image generating unit 10 stores the generated pseudo-defect image in the database 20. The image generating unit 10 may acquire a non-defective product image from the database 20 and generate a pseudo-defective product image.

[0035] The image acquisition unit 11 is a functional element that acquires image data. The image acquisition unit 11 acquires image data as teacher data (learning data) from the database 20. The image acquisition unit 11 acquires images of good products and images of defective products as image data. The image acquisition unit 11 acquires pseudo-defective images as defective images. The image acquisition unit 11 outputs the image data to the feature acquisition unit 12.

[0036] The feature acquisition unit 12 is a functional element that acquires features from image data. Based on the image data output from the image acquisition unit 11, the feature acquisition unit 12 acquires a first feature of the non-defective image and a second feature of the area other than the defective part in the pseudo-defective image.

[0037] 3 is a diagram showing an example of feature acquisition processing in the feature acquisition unit. As shown in FIG. 3, in this embodiment, the feature acquisition unit 12 covers the defective parts DP in the pseudo-fault image G with a mask M. The feature acquisition unit 12 acquires features of the areas (non-defective areas) other than the defective parts DP in the pseudo-fault image G in which the defective parts DP are covered with the mask. The feature acquisition unit 12 inputs each of the non-defective image and the pseudo-fault image into a neural network, and acquires a first feature and a second feature.

[0038] FIG. 4 is a diagram showing an example of a neural network NW. As shown in FIG. 4, the neural network NW is composed of, for example, a first layer which is an input layer L1, a second layer, a third layer, and a fourth layer which are intermediate layers (hidden layers) L2, and a fifth layer which is an output layer L3. The first layer outputs an input value x=(x0, x1, x2, ... xp) having p parameters as elements to the second layer as is. Each of the second, third, and fourth layers converts the total input into an output using an activation function and passes the output to the next layer. The fifth layer also converts the total input into an output using an activation function, and this output is the output value y=(y0, y1, ... , yp) of the neural network NW having q parameters as elements. q )

[0039] The feature acquisition unit 12 inputs each of the good product image and the pseudo-fault image into the input layer L1 of the neural network NW and acquires a first feature and a second feature. For the pseudo-fault image, the feature acquisition unit 12 inputs the area other than the faulty part (the part without the mask M) into the neural network NW. The feature acquisition unit 12 acquires the output of the intermediate layer L2 or the output layer L3 as the first feature and the second feature. The first feature and the second feature acquired by the feature acquisition unit 12 are represented in a feature space (see FIG. 5).

[0040] FIG. 5 is a diagram showing an example of distribution in feature space. In FIG. 5, first features are indicated by black circles (●) and second features are indicated by white circles (◯). FIG. 5 shows a first distribution D1 of the first features and a second distribution D2 of the second features. The first distribution D1 is, for example, a 95% distribution range of a normal distribution for a plurality of first features. The second distribution D2 is, for example, a 95% distribution range of a normal distribution for a plurality of second features.

[0041] The determination unit 13 is a functional element that determines the similarity between the first feature and the second feature. The determination unit 13 determines whether the first feature and the second feature acquired by the feature acquisition unit 12 are similar. In this embodiment, the determination unit 13 determines that the first feature and the second feature are similar if the second feature is present in the first distribution D1.

[0042] The learning unit 14 is a functional element that performs machine learning. The learning unit 14 performs machine learning of the neural network NW. The learning unit 14 performs machine learning of the learning model using images including pseudo-fault images having a second feature belonging to the first distribution D1 in the determination unit 13. The learning unit 14 inputs learning data to the input layer L1 and repeatedly (iteratively) performs a learning process on the neural network NW to generate a trained model. The learning unit 14 stores the trained model obtained by repeating the learning process a predetermined number of times in the memory unit 15.

[0043] The prediction unit 16 is a functional element that predicts the presence or absence of a defect using a trained model. The prediction unit 16 inputs input data to the trained model. In response to inputting the input data to the neural network NW of the trained model, the prediction unit 16 acquires a prediction result including an output value output from the neural network NW.

[0044] In this embodiment, the neural network NW receives the pixel values ​​of each pixel of image data as input data and outputs information indicating a prediction result. The input layer L1 of the neural network NW is provided with neurons equal to the number of pixels in the image. The output layer L3 of the neural network NW is provided with neurons for outputting information related to the prediction result of the target object. The presence or absence of a defect can be predicted based on the output values ​​(likelihoods) of the neurons in the output layer.

[0045] [Operation of the prediction system] (Learning phase) The learning method and the method for generating a trained model will be described with reference to Fig. 6. Fig. 6 is a flowchart showing an example of the learning method and the generation of a trained model as a processing flow S1. The processing flow S1 corresponds to the learning phase.

[0046] In step S11, the image acquisition unit 11 acquires image data from the database 20. In step S12, the feature acquisition unit 12 acquires the first feature and the second feature.

[0047] In step S13, the determination unit 13 determines whether the first feature and the second feature are similar. If the first feature and the second feature are similar (step S13: YES), the learning unit 14 performs machine learning using the training image in step S14. If the first feature and the second feature are not similar (step S13: NO), the process ends.

[0048] In step S15, the learning unit 14 determines whether to terminate learning. If the termination condition for machine learning is satisfied, the learning unit 14 terminates learning. If the termination condition is not satisfied, the learning unit 14 continues machine learning. The termination condition is set arbitrarily. For example, the termination condition may be set based on an error, or may be set based on the number of training data to be processed, i.e., the number of times learning is performed.

[0049] If learning is to continue (step S15: NO), the learning unit 14 acquires the next training image and executes the processes from step S14 onwards for that training image. If learning is to end (step S15: YES), the learning unit 14 acquires a trained model in step S16. Thus, in the learning phase, the prediction system 1 generates a trained model by performing machine learning using training data.

[0050] (Operational phase) The prediction method will be described with reference to Fig. 7. Fig. 7 is a flowchart showing an example of the prediction method as a processing flow S2. The processing flow S2 corresponds to the operation phase and is an example of the prediction method according to the present disclosure.

[0051] In step S21, the prediction unit 16 acquires input data. In step S22, the prediction unit 16 inputs the input data into a trained model and outputs a prediction result obtained by the trained model. The method for outputting the prediction result by the prediction unit 16 is not particularly limited. For example, the prediction unit 16 may output the prediction result to the output device 106, store it in a predetermined database, or transmit it to another computer system.

[0052] [effect] As described above, the prediction system 1 according to this embodiment acquires a first feature of a good product image and a second feature of a pseudo-failure image and determines whether the first feature and the second feature are similar. If the prediction system 1 determines that the first feature and the second feature are similar, it performs machine learning of the learning model using the pseudo-failure image having the second feature. In this way, the prediction system 1 performs machine learning of the learning model using pseudo-failure images in which the features of the non-defective area (the good product area) are similar to those of an actually captured good product image (actual image). In other words, the learning program does not use pseudo-failure images with different features from good product images in the machine learning of the learning model. Therefore, the prediction system 1 can perform machine learning of the learning model using pseudo-failure images of similar quality to good product images, thereby improving the accuracy of the trained model generated by machine learning. Therefore, the prediction system 1 can improve the accuracy of the trained model generated by machine learning using pseudo images. As a result, the prediction system 1 can improve prediction accuracy.

[0053] In the prediction system 1 according to this embodiment, the feature acquisition unit 12 masks the defective parts of the pseudo-fault image and acquires the features of the areas other than the defective parts in the pseudo-fault image. With this configuration, the features of the areas other than the defective parts in the pseudo-fault image can be acquired more reliably.

[0054] In the prediction system 1 according to this embodiment, the feature acquisition unit 12 inputs a non-defective image and a pseudo-defective image to the input layer of a neural network NW including an input layer L1, one or more intermediate layers L2, and an output layer L3, and acquires the first feature and the second feature. In this configuration, the neural network NW can acquire the first feature and the second feature.

[0055] In the prediction system 1 according to this embodiment, the feature acquisition unit 12 may acquire the output of the intermediate layer L2 or the output layer L3 as the first feature and the second feature. The output of the intermediate layer L2 and the output of the output layer L3 have different features, such as different feature dimensions. Therefore, the features to be acquired can be selected depending on the desired performance.

[0056] Although the embodiments of the present disclosure have been described above, the present disclosure is not necessarily limited to the above-described embodiments, and various modifications are possible without departing from the spirit of the present disclosure.

[0057] In the above embodiment, the image generating unit 10 generates the pseudo-fault image using a neural network as an example. However, the method of generating the pseudo-fault image by the image generating unit 10 is not limited to this.

[0058] In the above embodiment, the neural network NW is described as being composed of a first layer, which is an input layer L1, a second layer, a third layer, and a fourth layer, which are intermediate layers (hidden layers) L2, and a fifth layer, which is an output layer L3. However, the neural network may include one or more intermediate layers.

[0059] In addition to the above embodiment, the learning unit 14 may select pseudo-fault images to be used for machine learning of the learning model. The learning unit 14 determines whether or not a pseudo-fault image satisfies a predetermined condition from among a plurality of pseudo-fault images, and uses only the pseudo-fault images that satisfy the predetermined condition for machine learning of the learning model.

[0060] The learning unit 14 may select a pseudo-fault image using a generative adversarial network (GAN). The learning unit 14 inputs the pseudo-fault image to the generative adversarial network and obtains an output result (class) of the generative adversarial network. If the output result indicates that the pseudo-fault image is genuine, the learning unit 14 determines that the pseudo-fault image satisfies a predetermined condition. If the output result indicates that the pseudo-fault image is fake, the learning unit 14 determines that the pseudo-fault image does not satisfy the predetermined condition.

[0061] The learning unit 14 may also input a virtual image and a real image to a generative adversarial network and determine whether or not a pseudo-fault image satisfies a predetermined condition based on the output result of the generative adversarial network. A virtual image is an image that may exist between feature points of a pseudo-fault image and feature points of a real image in a feature space including feature points of the pseudo-fault image and the real-fault image. In the generative adversarial network, the learning unit 14 inputs the virtual image and the real-fault image to a discriminator and determines whether or not the virtual image satisfies a predetermined condition based on the output result of the discriminator. If the output result indicates that the virtual image is genuine, the learning unit 14 determines that the pseudo-fault image satisfies the predetermined condition. If the output result indicates that the virtual image is fake, the learning unit 14 determines that the pseudo-fault image does not satisfy the predetermined condition.

[0062] The learning unit 14 may also input each of the real defective image and the virtual image into a discrimination model and determine whether the pseudo-fault image satisfies a predetermined condition based on the output result of the discrimination model. The discrimination model is a trained model that determines whether an image is a pseudo-fault image. If the output results of the discrimination model are the same for each of the real defective image and the virtual image, the learning unit 14 determines that the pseudo-fault image satisfies the predetermined condition. If the output results of the discrimination model are different for each of the real defective image and the virtual image, the learning unit 14 determines that the pseudo-fault image does not satisfy the predetermined condition. [Explanation of symbols]

[0063] 1. Prediction system 10...Image generation unit 11...Image acquisition unit 12...Feature acquisition unit 13…Judgment section 14…Study Department 15...Storage section 16...Prediction Department 20...Database 100...Computer 101...Processor 102...Main memory section 103…Auxiliary storage unit 104...Communication control unit 105...Input device 106...Output device 110...Study Program 120...Prediction program D1…first distribution D2…Second distribution DP…Defective part G...pseudo-fault image L1...input layer L2…middle layer L3: Output layer M...Mask NW...neural network S1...Processing flow S2...Processing flow S11...Step S12...Step S13...Step S14...Step S15...Step S16...Step S21...Step S22...Step

Claims

1. A learning program that is executed on at least one processor and performs machine learning of a learning model, an image acquisition step of acquiring a photographed image of a non-defective product and a pseudo-generated image of a defective product including a defective portion; a feature acquiring step of acquiring a first feature of the non-defective image and a second feature of an area other than the defective portion in the defective image; a determination step of determining whether the first feature and the second feature acquired in the feature acquisition step are similar to each other; a learning step of performing machine learning of the learning model using the defective image having the second feature when it is determined in the determining step that the first feature and the second feature are similar; A learning program that causes a computer to execute the following.

2. 2. The learning program according to claim 1, wherein the feature acquiring step masks the defective portion of the defective image and acquires features of an area other than the defective portion in the defective image.

3. 3. The learning program according to claim 1, wherein the feature acquisition step inputs the non-defective image and the defective image to an input layer of a neural network including an input layer, one or more intermediate layers, and an output layer, and acquires the first feature and the second feature.

4. The learning program according to claim 3 , wherein in the feature acquisition step, outputs of the intermediate layer or the output layer are acquired as the first feature and the second feature.

5. 3. The learning program according to claim 1, wherein the feature acquisition step performs image processing on the non-defective image and the defective image to acquire the first feature and the second feature.

6. A learning method for performing machine learning on a learning model, the method being executed on at least one processor, comprising: an image acquisition step of acquiring a photographed image of a non-defective product and a pseudo-generated image of a defective product including a defective portion; a feature acquiring step of acquiring a first feature of the non-defective image and a second feature of an area other than the defective portion in the defective image; a determination step of determining whether the first feature and the second feature acquired in the feature acquisition step are similar to each other; a learning step of performing machine learning of the learning model using the defective image having the second feature when it is determined in the determination step that the first feature and the second feature are similar.

7. A learning device that includes at least one processor and performs machine learning on a learning model, an image acquisition unit that acquires a photographed image of a non-defective product and a pseudo-generated image of a defective product that includes a defective portion; a feature acquisition unit that acquires a first feature of the non-defective image and a second feature of a region other than the defective portion in the defective image; a determination unit that determines whether the first feature and the second feature acquired by the feature acquisition unit are similar to each other; a learning unit that, when the judgment unit determines that the first feature and the second feature are similar, performs machine learning of the learning model using the defective image having the second feature.

Images