TWI762193B - Image defect detection method, image defect detection device, electronic device and storage media - Google Patents

Abstract

The present application provides an image defect detection method, an image defect detection device, an electronic device and storage medium. The method includes: entering flawless sample training image into a self-encoder, calculating a first submersion feature through encoding layers of the self-encoder, entering the first submersion feature into decoding layers of the self-encoder to obtain first reconstruction images, and calculating first reconstruction error using an error function, entering the first subliminal feature into a deep learning model and a Gauss hybrid model to obtain a first probability distribution and a second probability distribution, calculating a dispersion between the first probability distribution and the second probability distribution, and calculating a total loss according to the first reconstruction error and the dispersion, setting a threshold based on the total loss, obtaining a testing sample image and calculating a total error to determine whether the testing sample image is a defective image.

Description

Image defect detection method, device, electronic device and storage medium

The invention relates to the field of image detection, provides the technical field of defect detection and computer vision, and in particular relates to an image defect detection method, device, electronic equipment and storage medium.

In order to improve the quality of industrial products, certain defects are usually detected on industrial products before they are packaged. Because the current flaw detection method based on Gaussian mixture model usually has a huge amount of calculation, the program speed cannot be optimized, and the execution time cannot be reduced.

In view of the above content, it is necessary to propose an image defect detection method, device, electronic device and storage medium to improve the efficiency of defect image detection.

A first aspect of the present application provides an image defect detection method, which includes: acquiring a training image of a flawless sample; inputting the training image of a flawless sample into an autoencoder, and obtaining it by calculating the coding layer of the autoencoder The first latent feature of the flawless sample training image; inputting the first latent feature into the decoding layer of the self-encoder and calculating the first reconstructed image of the flawless sample training image, and Compute the flawless sample training using a preset error function image and the first reconstructed image to obtain the first reconstruction error between the flawless sample training image and the first reconstructed image; input the first latent feature into the deep learning model and generate a Calculate the first probability distribution of the first latent feature; input the first latent feature into a Gaussian mixture model and calculate the second probability distribution of the first latent feature; calculate the first probability distribution and the the Kulbeck-Leibler divergence between the second probability distributions; a total loss is obtained from the first reconstruction error and the Kulbec-Leibler divergence, and the self-optimization is based on the total loss encoder, the deep learning model and the Gaussian mixture model, and set a threshold according to the total loss; obtain a test sample image, input the test sample image into the autoencoder, and use the autoencoder The coding layer of the encoder obtains the second latent feature of the test sample image, and the second latent feature is input into the decoding layer of the self-encoder to calculate the second reconstructed image of the test sample image. , and use the preset error function to calculate the second reconstruction error between the test sample image and the second reconstructed image, input the second latent feature into the trained deep learning model and Calculate a third probability distribution of the second latent feature, and calculate a total error according to the third probability distribution and the second reconstruction error; when the total error is greater than or equal to the threshold, determine the test The sample image is a flawed image, and when the total error is less than the threshold, it is determined that the test sample image is a flawless image.

Preferably, calculating the first latent feature of the flawless sample training image by the coding layer of the self-encoder includes: performing vectorization processing on the flawless sample training image to obtain the flawless sample training image. The feature vector of the flawed sample training image; The first latent feature is obtained by operating on the feature vector of the flawless sample training image by using the encoding layer in the autoencoder.

Preferably, the inputting the first latent feature into the decoding layer of the self-encoder and calculating the first reconstructed image of the flawless sample training image comprises: using all the data in the self-encoder The decoding layer operates on the first latent feature; and performs restoration processing on the vector obtained after the operation to obtain the first reconstructed image.

Preferably, the inputting the first latent feature into a deep learning model and calculating the first probability distribution of the first latent feature comprises: inputting the first latent feature into the deep learning model; One or more of a convolution layer, a pooling layer, and at least one hidden layer in the deep learning model operate on the first latent feature to obtain the first probability distribution.

計算所述第一概率分佈和所述第二概率分佈的庫爾貝克-萊布勒散度，其中，D _KL (P||Q)為所述第一概率分佈和所述第二概率分佈的庫爾貝克-萊布勒散度，P(i)為所述第二概率分佈，Q(i)為所述第一概率分佈。 Preferably, the calculating the Kulbec-Leibler divergence between the first probability distribution and the second probability distribution comprises: according to the formula

Calculate the Kulbec-Leibler divergence of the first probability distribution and the second probability distribution, wherein D _KL ( P || Q ) is the difference between the first probability distribution and the second probability distribution Kulbec-Leibler divergence, P ( i ) is the second probability distribution, and Q ( i ) is the first probability distribution.

Preferably, the total loss is obtained according to the first reconstruction error and the Kulbec-Leibler divergence, and the autoencoder, the deep learning model and the Gaussian are optimized according to the total loss The mixture model includes: calculating the product of the first reconstruction error and the Kulbeck-Leibler divergence to obtain the total loss; adjusting the autoencoder, the deep learning model and the Gaussian mixture The parameters of the model minimize the total loss.

Preferably, the calculating a total error according to the third probability distribution and the second reconstruction error includes: summing the third probability distribution and the second reconstruction error to obtain a total error.

A second aspect of the present application provides an image defect detection device, the device includes: a training image acquisition module for acquiring flawless sample training images; a first latent feature acquisition module for The flawed sample training image is input into the self-encoder, and the first latent feature of the flawless sample training image is calculated by the coding layer of the self-encoder; the first reconstruction error acquisition module is used to The first latent feature is input into the decoding layer of the self-encoder and the first reconstructed image of the flawless sample training image is obtained by calculation, and a preset error function is used to calculate the flawless sample training image and all samples. The first reconstructed image obtains the first reconstruction error between the flawless sample training image and the first reconstructed image; the first probability distribution calculation module is used to calculate the first latent image. The feature is input into the deep learning model and the first probability distribution of the first latent feature is obtained by calculation; the second probability distribution calculation module is used to input the first latent feature into the Gaussian mixture model and calculate the first latent feature. The second probability distribution of A total loss is obtained from the first reconstruction error and the Kulbec-Leibler divergence, and the autoencoder, the deep learning model and the Gaussian mixture model are optimized according to the total loss, and according to the The total loss is set as a threshold; the total error calculation module is used to obtain the test sample image, input the test sample image into the self-encoder, and calculate the test sample image by the coding layer of the self-encoder The second latent feature of the image, input the second latent feature into the decoding layer of the self-encoder and calculate the second reconstructed image of the test sample image, and use the preset error function to calculate The test sample image and the For the second reconstruction error between the second reconstructed images, the second latent feature is input into the trained deep learning model and the third probability distribution of the second latent feature is calculated to obtain, according to the third probability distribution and the second reconstruction error to calculate a total error; a judgment module, configured to determine that the test sample image is a defective image when the total error is greater than or equal to the threshold, and when the total error is less than or equal to the threshold When the threshold value is set, it is determined that the test sample image is a flawless image.

A third aspect of the present application provides an electronic device, the electronic device comprising: a memory storing at least one instruction; and a processor executing the instruction stored in the memory to implement the defect detection method.

A fourth aspect of the present application provides a computer storage medium on which a computer program is stored, and when the computer program is executed by a processor, the image defect detection method is implemented.

In the present invention, by training the autoencoder, the deep learning model and the Gaussian mixture model at the same time, the deep learning model and the Gaussian mixture model can output the same probability distribution prediction, so that the deep learning model can be used to replace the Gaussian mixture model. Achieve greater efficiency in detection.

30: Image defect detection device

301: Train image acquisition module

302: The first latent feature acquisition module

303: The first reconstruction error acquisition module

304: The first probability distribution calculation module

305: Second probability distribution calculation module

306: Divergence calculation module

307: Model training module

308: Total error calculation module

309: Judgment Module

6: Electronic equipment

61: Memory

62: Processor

63: Computer Programs

S11~S19: Steps

FIG. 1 is a flowchart of an image defect detection method according to an embodiment of the present invention.

FIG. 2 is a structural diagram of an image defect detection apparatus according to an embodiment of the present invention.

FIG. 3 is a schematic diagram of an electronic device in an embodiment of the present invention.

In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments may be combined with each other in the case of no conflict.

In the following description, many specific details are set forth in order to facilitate a full understanding of the present invention, and the described embodiments are only some, but not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. The terms used herein in the description of the present invention are for the purpose of describing specific embodiments only, and are not intended to limit the present invention.

Preferably, the image defect detection method of the present invention is applied in one or more electronic devices. The electronic device is a device that can automatically perform numerical calculations and/or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application specific integrated circuits (ASICs) , Programmable Gate Array (Field-Programmable Gate Array, FPGA), Digital Signal Processor (Digital Signal Processor, DSP), embedded devices, etc.

The electronic device may be a computing device such as a desktop computer, a notebook computer, a tablet computer, and a cloud server. The electronic device can perform human-computer interaction with the user by means of a keyboard, a mouse, a remote control, a touch pad or a voice control device.

Example 1

FIG. 1 is a flowchart of an image defect detection method in an embodiment of the present invention. The image defect detection method is applied in electronic equipment. According to different requirements, the order of the steps in the flowchart can be changed, and some steps can be omitted.

Referring to FIG. 1 , the image defect detection method specifically includes the following steps.

In step S11, a training image of a flawless sample is obtained.

Step S12: Input the flawless sample training image into an auto-encoder, and calculate the first latent feature of the flawless sample training image through the coding layer of the auto-encoder.

In at least one embodiment of the present invention, calculating the first latent feature of the flawless sample training image by the coding layer of the autoencoder includes: performing a vector to obtain the feature vector of the flawless sample training image; use the coding layer in the self-encoder to operate on the feature vector of the flawless sample training image to obtain the first latent features.

Step S13, input the first latent feature into the decoding layer of the auto-encoder and calculate the first reconstructed image of the flawless sample training image, and use a preset error function to calculate the flawless The sample training image and the first reconstructed image obtain a first reconstruction error between the flawless sample training image and the first reconstructed image.

Preferably, the inputting the first latent feature into the decoding layer of the self-encoder and calculating the first reconstructed image of the flawless sample training image comprises: using all the data in the self-encoder The decoding layer operates on the first latent feature; and performs restoration processing on the vector obtained after the operation to obtain the first reconstructed image.

Step S14, inputting the first latent feature into a deep learning model and calculating a first probability distribution of the first latent feature.

In at least one embodiment of the present invention, the inputting the first latent feature into a deep learning model and calculating the first probability distribution of the first latent feature includes: inputting the first latent feature into the depth learning model; the first probability distribution is obtained by operating on the first latent feature by one or more of a convolution layer, a pooling layer and at least one hidden layer in the deep learning model.

Step S15, the first latent feature is input into a Gaussian mixture model and a second probability distribution of the first latent feature is obtained by calculation.

In at least one embodiment of the present invention, inputting the first latent feature into a Gaussian mixture model and calculating a second probability distribution of the first latent feature includes: inputting the first latent feature into the Gaussian mixture model; The second probability distribution is obtained by fitting the probability distribution of the first latent feature by the Gaussian mixture model.

其中，x _i 表示第一潛特徵對應的向量，t=1，2，3...，M，M為第一潛特徵的維度，α _k 為第k個高斯分佈的權重，μ _k ，σ _k 分別時第k個高斯分佈的均值和方差，N(x _i |μ _k ,σ _k )表示向量x _i 符合均值為μ _k 且方差為σ _k 的正態分佈，K至少為3。 Specifically, the Gaussian mixture model is

Among them, x _i represents the vector corresponding to the first latent feature, t=1, 2, 3..., M, M is the dimension of the first latent feature, α _k is the weight of the kth Gaussian distribution, μ _k , σ _k is the mean and variance of the k-th Gaussian distribution, respectively, N ( x _i |μ _k ,σ _k ) means that the vector x _i conforms to a normal distribution with mean μ _k and variance σ _k , and K is at least 3.

Step S16: Calculate the Kulbeck-Leebler divergence between the first probability distribution and the second probability distribution.

計算所述第一概率分佈和所述第二概率分佈的庫爾貝克-萊布勒散度，其中，D _KL (P||Q)為所述第一概率分佈和所述第二概率分佈的庫爾貝克-萊布勒散度，P(i)為所述第二概率分佈，Q(i)為所述第一概率分佈。 Preferably, the calculating the Kulbec-Leibler divergence between the first probability distribution and the second probability distribution comprises: according to the formula

Calculate the Kulbec-Leibler divergence of the first probability distribution and the second probability distribution, wherein D _KL ( P || Q ) is the difference between the first probability distribution and the second probability distribution Kulbec-Leibler divergence, P ( i ) is the second probability distribution, and Q ( i ) is the first probability distribution.

Step S17: Obtain a total loss according to the first reconstruction error and the Kulbec-Leibler divergence, and optimize the autoencoder, the deep learning model and the Gaussian mixture model according to the total loss , and set a threshold based on the total loss.

In at least one embodiment of the present invention, a total loss is obtained according to the first reconstruction error and the Kurbeck-Leebler divergence, and the self-encoder, the self-encoder are optimized according to the total loss The deep learning model and the Gaussian mixture model include: calculating the product of the first reconstruction error and the Kulbec-Leebler divergence to obtain the total loss; The total loss is minimized by adjusting parameters of the autoencoder, the deep learning model and the Gaussian mixture model.

In at least one embodiment of the present invention, the purpose of adjusting the parameters of the autoencoder, the deep learning model and the Gaussian mixture model to minimize the total loss is to optimize the autoencoder, the The deep learning model and the Gaussian mixture model are configured so that the deep learning model and the Gaussian mixture model have the same probability distribution according to the reconstructed image generated by the autoencoder.

In step S18, a test sample image is acquired, the test sample image is input into the self-encoder, the second latent feature of the test sample image is calculated by the coding layer of the self-encoder, and the second latent feature of the test sample image is obtained. The second latent feature is input to the decoding layer of the auto-encoder to obtain a second reconstructed image of the test sample image, and the preset error function is used to calculate the test sample image and the first reconstructed image of the test sample image. For the second reconstruction error between the two reconstructed images, the second latent feature is input into the trained deep learning model and the third probability distribution of the second latent feature is obtained by calculation. According to the third probability distribution and the second reconstruction error to calculate the total error.

In at least one embodiment of the present invention, the calculating a total error according to the third probability distribution and the second reconstruction error includes: summing the third probability distribution and the second reconstruction error to obtain total error.

Step S19, when the total error is greater than or equal to the threshold, determine that the test sample image is a flawed image, and when the total error is less than the threshold, determine that the test sample image is a flawless image picture.

In the present invention, by training the autoencoder, the deep learning model and the Gaussian mixture model at the same time, the deep learning model and the Gaussian mixture model can output the same probability distribution prediction, so that the deep learning model can be used to replace the Gaussian mixture model. Achieve greater efficiency in detection.

Example 2

FIG. 2 is a structural diagram of an image defect detection apparatus 30 in an embodiment of the present invention.

In some embodiments, the image flaw detection apparatus 30 operates in an electronic device. The image defect detection apparatus 30 may include a plurality of functional modules composed of program code segments. The program codes of each program section in the image defect detection apparatus 30 can be stored in the memory and executed by at least one processor to perform the image defect detection function.

In this embodiment, the image defect detection apparatus 30 can be divided into a plurality of functional modules according to the functions performed by the image defect detection apparatus 30 . Referring to FIG. 2, the image defect detection device 30 may include a training image acquisition module 301, a first latent feature acquisition module 302, a first reconstruction error acquisition module 303, and a first probability distribution calculation module 304 , a second probability distribution calculation module 305 , a divergence calculation module 306 , a model training module 307 , a total error calculation module 308 and a judgment module 309 . The module referred to in the present invention refers to a series of computer program segments that can be executed by at least one processor and can perform fixed functions, and are stored in a memory. In some embodiments, the functions of each module will be described in detail in subsequent embodiments.

The training image acquisition module 301 acquires flawless sample training images.

The first latent feature acquisition module 302 inputs the flawless sample training image into an autoencoder, and calculates the first latent feature of the flawless sample training image through the coding layer of the autoencoder.

In at least one embodiment of the present invention, the first latent feature obtaining module 302 calculates the first latent feature of the flawless sample training image through the coding layer of the autoencoder, comprising: The flawless sample training image is vectorized to obtain the feature vector of the flawless sample training image; The first latent feature is obtained by operating on the feature vector of the flawless sample training image by using the encoding layer in the autoencoder.

The first reconstruction error acquisition module 303 inputs the first latent feature into the decoding layer of the self-encoder and calculates the first reconstructed image of the flawless sample training image, and uses a preset Calculate the error function of the flawless sample training image and the first reconstructed image to obtain the first reconstruction error between the flawless sample training image and the first reconstructed image.

In at least one embodiment of the present invention, the first reconstruction error obtaining module 303 inputs the first latent feature into the decoding layer of the auto-encoder and calculates the first image of the flawless sample training image. A reconstructed image includes: using the decoding layer in the self-encoder to operate on the first latent feature; and performing restoration processing on the vector obtained after the operation to obtain the first reconstructed image.

The first probability distribution calculation module 304 inputs the first latent feature into a deep learning model and calculates the first probability distribution of the first latent feature.

In at least one embodiment of the present invention, the first probability distribution calculation module 304 inputs the first latent feature into a deep learning model and calculates to obtain the first probability distribution of the first latent feature includes: The first latent feature is input into the deep learning model; the first latent feature is operated by one or more of the convolutional layer, the pooling layer and at least one hidden layer in the deep learning model, and the first latent feature is obtained. a probability distribution.

The second probability distribution calculation module 305 inputs the first latent feature into a Gaussian mixture model and calculates a second probability distribution of the first latent feature.

In at least one embodiment of the present invention, the second probability distribution calculation module 305 inputs the first latent feature into a Gaussian mixture model and obtains the second probability distribution of the first latent feature by calculating: The first latent feature is input into the Gaussian mixture model; the probability distribution of the first latent feature is fitted by the Gaussian mixture model to obtain the second probability distribution.

其中，x _i 表示第一潛特徵對應的向量，t=1，2，3...，M，M為第一潛特徵的維度，α _k 為第k個高斯分佈的權重，μ _k ，σ _k 分別時第k個高斯分佈的均值和方差，N(x _i |μ _k ,σ _k )表示向量x _i 符合均值為μ _k 且方差為σ _k 的正態分佈，K至少為3。 Specifically, the Gaussian mixture model is

Among them, x _i represents the vector corresponding to the first latent feature, t=1, 2, 3..., M, M is the dimension of the first latent feature, α _k is the weight of the kth Gaussian distribution, μ _k , σ _k is the mean and variance of the k-th Gaussian distribution, respectively, N ( x _i |μ _k ,σ _k ) means that the vector x _i conforms to a normal distribution with mean μ _k and variance σ _k , and K is at least 3.

The divergence calculation module 306 calculates the Kulbec-Leebler divergence between the first probability distribution and the second probability distribution.

計算所述第一概率分佈和所述第二概率分佈的庫爾貝克-萊布勒散度，其中，D _KL (P||Q)為所述第一概率分佈和所述第二概率分佈的庫爾貝克-萊布勒散度，P(i)為所述第二概率分佈，Q(i)為所述第一概率分佈。 Preferably, the calculating the Kulbec-Leibler divergence between the first probability distribution and the second probability distribution includes: according to the formula

Calculate the Kulbec-Leibler divergence of the first probability distribution and the second probability distribution, wherein D _KL ( P || Q ) is the difference between the first probability distribution and the second probability distribution Kulbec-Leibler divergence, P ( i ) is the second probability distribution, and Q ( i ) is the first probability distribution.

The model training module 307 obtains a total loss according to the first reconstruction error and the Kulbec-Leibler divergence, and optimizes the autoencoder, the deep learning model and all the methods according to the total loss. The Gaussian mixture model is described, and the threshold is set according to the total loss.

In at least one embodiment of the present invention, the model training module 307 obtains a total loss according to the first reconstruction error and the Kulbec-Leibler divergence, and optimizes the self-contained loss according to the total loss. The encoder, the deep learning model and the Gaussian mixture model include: calculating the product of the first reconstruction error and the Kulbeck-Leebler divergence to obtain the total loss; adjusting the self-encoding The total loss is minimized by the parameters of the engine, the deep learning model and the Gaussian mixture model.

In at least one embodiment of the present invention, the purpose of adjusting the parameters of the autoencoder, the deep learning model and the Gaussian mixture model to minimize the total loss is to optimize the autoencoder, the the deep learning model and the Gaussian mixture model, so that the deep learning model has the same probability distribution as the reconstructed image generated by the Gaussian mixture model from the autoencoder.

The total error calculation module 308 obtains a test sample image, inputs the test sample image into the auto-encoder, and calculates the second latent value of the test sample image through the coding layer of the auto-encoder. feature, input the second latent feature into the decoding layer of the self-encoder and calculate the second reconstructed image of the test sample image, and use the preset error function to calculate the test sample image The second reconstruction error between the image and the second reconstructed image, the second latent feature is input into the trained deep learning model and the third probability distribution of the second latent feature is obtained by calculation. A total error is calculated from the third probability distribution and the second reconstruction error.

In at least one embodiment of the present invention, the total error calculation module 308 calculating the total error according to the third probability distribution and the second reconstruction error includes: calculating the third probability distribution and the second The reconstruction errors are summed to obtain the total error.

The judging module 309 determines that the test sample image is a defective image when the total error is greater than or equal to the threshold, and determines that the test sample image is a defective image when the total error is less than the threshold. Flawless image.

In the present invention, by training the autoencoder, the deep learning model and the Gaussian mixture model at the same time, the deep learning model and the Gaussian mixture model can output the same probability distribution prediction, so that the deep learning model can be used to replace the Gaussian mixture model. Achieve greater efficiency in detection.

Example 3

FIG. 3 is a schematic diagram of an electronic device 6 in an embodiment of the present invention.

The electronic device 6 includes a memory 61 , a processor 62 and a computer program 63 stored in the memory 61 and executable on the processor 62 . When the processor 62 executes the computer program 63, the steps in the above-mentioned embodiment of the image defect detection method are implemented, such as the steps shown in FIG. 1 . S11~S19. Alternatively, when the processor 62 executes the computer program 63, the functions of each module/unit in the above-mentioned embodiment of the image defect detection apparatus are realized, for example, the modules 301-309 in FIG. 2 .

Exemplarily, the computer program 63 can be divided into one or more modules/units, and the one or more modules/units are stored in the memory 61 and executed by the processor 62 , to complete the present invention. The one or more modules/units may be a series of computer program instruction segments capable of performing specific functions, and the instruction segments are used to describe the execution process of the computer program 63 in the electronic device 6 . For example, the computer program 63 can be divided into the training image acquisition module 301 in FIG. 2 , the first latent feature acquisition module 302 , the first reconstruction error acquisition module 303 , and the first probability distribution calculation module 304 , the second probability distribution calculation module 305, the divergence calculation module 306, the model training module 307, the total error calculation module 308 and the judgment module 309. For the specific functions of each module, refer to Embodiment 2.

In this embodiment, the electronic device 6 may be a computing device such as a desktop computer, a notebook, a palmtop computer, a server, and a cloud terminal device. Those skilled in the art can understand that the schematic diagram is only an example of the electronic device 6, and does not constitute a limitation to the electronic device 6, and may include more or less components than the one shown, or combine some components, or different Components such as the electronic device 6 may also include input and output devices, network access devices, bus bars, and the like.

The processor 62 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs) ), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor 62 can also be any conventional processor, etc. The processor 62 is the control center of the electronic device 6, and uses various interfaces and lines to connect the entire electronic device 6. various parts.

The memory 61 can be used to store the computer programs 63 and/or modules/units, and the processor 62 runs or executes the computer programs and/or modules/units stored in the memory 61, And call the data stored in the memory 61 to realize various functions of the electronic device 6 . The memory 61 may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; storage data The area may store data (such as audio data, phone book, etc.) created according to the use of the electronic device 6, and the like. In addition, the memory 61 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, Smart Media Card (SMC), Secure Digital (Secure Digital, SD) card, flash memory card (Flash Card), at least one disk memory device, flash memory device, or other volatile solid state memory device.

If the modules/units integrated in the electronic device 6 are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the present invention realizes all or part of the processes in the methods of the above embodiments, and can also be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a computer-readable storage medium, When the computer program is executed by the processor, the steps of the above method embodiments can be implemented. Wherein, the computer program includes computer program code, and the computer program code may be in the form of original code, object code, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, a recording medium, a flash drive, a removable hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM, Read-Only Memory); Only Memory), random access memory (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium, etc.

In the several embodiments provided by the present invention, it should be understood that the disclosed apparatus and method may be implemented in other manners. For example, the device embodiments described above are only illustrative. For example, the division of the modules is only a logical function division, and other division methods may be used in actual implementation.

In addition, each functional module in each embodiment of the present invention may be integrated in the same processing module, or each module may exist physically alone, or two or more modules may be integrated in the same module. The above-mentioned integrated modules can be implemented in the form of hardware, or can be implemented in the form of hardware plus software function modules.

It will be apparent to those skilled in the art that the present invention is not limited to the details of the above-described exemplary embodiments, but that the present invention may be embodied in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments are to be regarded in all respects as illustrative and not restrictive, and the scope of the present invention is defined by the appended claims rather than the foregoing description, and is therefore intended to fall within the scope of the claims. All changes within the meaning and range of the equivalents of , are included in the present invention. Any reference sign in a claim should not be construed as limiting the claim to which it relates. Furthermore, it is clear that the word "comprising" does not exclude other modules or steps, and the singular does not exclude the plural. Multiple modules or electronic devices stated in the electronic device claim may also be implemented by the same module or electronic device through software or hardware. The terms first, second, etc. are used to denote names and do not denote any particular order.

To sum up, the present invention complies with the requirements of an invention patent, and a patent application can be filed in accordance with the law. However, the above descriptions are only the preferred embodiments of the present invention, and for those who are familiar with the techniques of this case, equivalent modifications or changes made in accordance with the creative spirit of this case shall be included in the scope of the following patent application.

S11~S19: Steps

Claims (10)

An image flaw detection method, applied in electronic equipment, wherein the method comprises: acquiring a flawless sample training image; inputting the flawless sample training image into an autoencoder, The encoding layer calculates and obtains the first latent feature of the flawless sample training image; inputs the first latent feature into the decoding layer of the self-encoder and calculates to obtain the first reconstruction of the flawless sample training image image, and use the preset error function to calculate the flawless sample training image and the first reconstructed image to obtain the first error between the flawless sample training image and the first reconstructed image. a reconstruction error; input the first latent feature into a deep learning model and calculate the first probability distribution of the first latent feature; input the first latent feature into a Gaussian mixture model and calculate the first latent feature a second probability distribution of features; calculating the Kulbec-Leibler divergence between the first probability distribution and the second probability distribution; The total loss is obtained from the Buller divergence, the autoencoder, the deep learning model and the Gaussian mixture model are optimized according to the total loss, and a threshold is set according to the total loss; the test sample image is obtained, and the The test sample image is input into the self-encoder, the second latent feature of the test sample image is calculated by the coding layer of the self-encoder, and the second latent feature is input into the decoding of the self-encoder layer and calculate the second reconstructed image of the test sample image, and use the preset error function to calculate the second reconstruction between the test sample image and the second reconstructed image error, input the second latent feature into the trained deep learning model and calculate the obtained the third probability distribution of the second latent feature, and calculate the total error according to the third probability distribution and the second reconstruction error; when the total error is greater than or equal to the threshold, determine the test sample image For a flawed image, when the total error is less than the threshold, the test sample image is determined to be a flawless image. The image flaw detection method according to claim 1, wherein calculating the first latent feature of the flawless sample training image by the coding layer of the self-encoder comprises: converting the flawless sample The training image is vectorized to obtain the feature vector of the flawless sample training image; the coding layer in the self-encoder is used to operate on the feature vector of the flawless sample training image, The first latent feature is obtained. The image flaw detection method according to claim 1, wherein the first latent feature is input into the decoding layer of the auto-encoder and the first reconstruction image of the flawless sample training image is obtained by calculation The image includes: using the decoding layer in the self-encoder to operate on the first latent feature; and performing restoration processing on the vector obtained after the operation to obtain the first reconstructed image. The image flaw detection method according to claim 1, wherein the inputting the first latent feature into a deep learning model and calculating the first probability distribution of the first latent feature comprises: The feature is input into the deep learning model; the first latent feature is operated by one or more of the convolution layer, the pooling layer and at least one hidden layer in the deep learning model to obtain the first probability distribution . The image flaw detection method according to claim 1, wherein the calculating the Kulbec-Leebler divergence between the first probability distribution and the second probability distribution comprises: according to the formula

Calculate the Kulbec-Leibler divergence of the first probability distribution and the second probability distribution, wherein D _KL ( P || Q ) is the difference between the first probability distribution and the second probability distribution Kulbec-Leibler divergence, P ( i ) is the second probability distribution, and Q ( i ) is the first probability distribution. The image defect detection method according to claim 1, wherein a total loss is obtained according to the first reconstruction error and the Kulbec-Leibler divergence, and the self-contained loss is optimized according to the total loss. The encoder, the deep learning model and the Gaussian mixture model include: calculating the product of the first reconstruction error and the Kulbeck-Leebler divergence to obtain the total loss; adjusting the self-encoding The total loss is minimized by the parameters of the engine, the deep learning model and the Gaussian mixture model. The image defect detection method according to claim 1, wherein the calculating the total error according to the third probability distribution and the second reconstruction error comprises: comparing the third probability distribution and the second reconstruction error Sum the construction errors to get the total error. An image defect detection device, wherein the device comprises: a training image acquisition module for acquiring flawless sample training images; a first latent feature acquisition module for acquiring the flawless sample training images Input the self-encoder, and calculate the first latent feature of the flawless sample training image by the coding layer of the self-encoder; the first reconstruction error acquisition module is used to input the first latent feature The decoding layer of the auto-encoder obtains the first reconstructed image of the flawless sample training image, and uses a preset error function to calculate the flawless sample training image and the first reconstructed image. The image obtains the first reconstruction error between the flawless sample training image and the first reconstructed image; the first probability distribution calculation module is used to input the first latent feature into the deep learning model and calculating the first probability distribution of the first latent feature; The second probability distribution calculation module is used to input the first latent feature into a Gaussian mixture model and calculate the second probability distribution of the first latent feature; the divergence calculation module is used to calculate the first probability the Kulbeck-Leibler divergence between the distribution and the second probability distribution; a model training module for obtaining a total loss, optimize the self-encoder, the deep learning model and the Gaussian mixture model according to the total loss, and set a threshold according to the total loss; the total error calculation module is used to obtain the test sample image, and the The test sample image is input into the self-encoder, the second latent feature of the test sample image is calculated by the coding layer of the self-encoder, and the second latent feature is input into the self-encoder The decoding layer and the second reconstructed image of the test sample image are obtained by calculation, and the preset error function is used to calculate the second reconstructed image between the test sample image and the second reconstructed image. Reconstruction error, input the second latent feature into the trained deep learning model and calculate the third probability distribution of the second latent feature, and calculate the total probability according to the third probability distribution and the second reconstruction error. error; a judgment module for determining that the test sample image is a defective image when the total error is greater than or equal to the threshold, and determining the test sample image when the total error is less than the threshold Like a flawless image. An electronic device, wherein the electronic device comprises: a memory that stores at least one instruction; and a processor that executes the instructions stored in the memory to realize the image according to any one of claim 1 to 7 Defect detection method. A storage medium having a computer program stored thereon, wherein: the computer program implements the image defect detection method according to any one of claim 1 to claim 7 when the computer program is executed by a processor.

Images