Industrial image anomaly detection-oriented anomaly score graph optimization method and system

Abstract

The invention provides an abnormal score graph optimization method and system for industrial image anomaly detection, which further optimizes an abnormal score graph extracted by an existing anomaly detection method based on unsupervised or self-supervised learning, and designs a foreground noise suppression module and a background anomaly suppression module. Obtaining an abnormal score graph to be optimized through an existing method; through foreground noise suppression processing, reducing an abnormal score of pseudo anomalies on the foreground region, and suppressing the pseudo anomaly influence of the foreground region in an anomaly detection process; an optimization mark related to an image foreground or an object area is generated through a background anomaly suppression module, an anomaly score graph only pays attention to an abnormal area of the image foreground or the object area, the pseudo-anomaly influence of a background area is suppressed, the method does not depend on product categories, extra training is not needed, and the method is suitable for large-scale popularization and application. The method is not limited to a specific unsupervised or self-supervised learning anomaly detection method, and can be used as a plug-in for plug-and-play.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition, and in particular relates to an anomaly score map optimization method and system for industrial image anomaly detection. Background Art

[0002] Automatic anomaly detection methods based on artificial intelligence have been widely used in the industrial field. One common solution is based on unsupervised / self-supervised learning methods, which model a large number of normal samples to obtain the distribution or representation of normal images. In actual application, the difference between the sample to be detected and the modeled normal distribution or representation is calculated to obtain an anomaly score map for identifying / locating abnormal areas.

[0003] However, since no real abnormal regions / samples have been seen during the training process, for the "pseudo-abnormal regions" in the samples to be detected that are different from normal regions but not real abnormal regions, such abnormality detection methods based on unsupervised or self-supervised learning are often greatly limited in recognition / localization and cannot distinguish them from real abnormal regions. The specific situation is as follows: 1) For the foreground or object regions on the image, there are often pseudo-abnormal regions caused by noise, which will also have high abnormality scores on the abnormality score map, which will affect the final detection / localization results. 2) For some pseudo-abnormal regions in the background area of ​​the image, they will also have high abnormality scores on the abnormality score map. However, the problem of abnormal detection / localization focuses on the foreground or object in the image, and these pseudo-abnormalities on the background will also affect the final detection / localization results. Summary of the invention

[0004] The purpose of the present invention is to provide an anomaly score map optimization method for industrial image anomaly detection. For the anomaly score map extracted by the existing anomaly detection method based on unsupervised or self-supervised learning, the anomaly scores of pseudo-anomaly areas on the foreground and background of the sample to be detected in the anomaly score map are suppressed respectively through foreground noise suppression and background anomaly suppression, and then the anomaly score map is further optimized to improve the ability to identify / locate abnormal areas. At the same time, the present invention does not depend on the specific product category, does not require additional training, is not limited to a specific anomaly detection method based on unsupervised or self-supervised learning, and can be used as a plug-in in a plug-in form.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is: an abnormal score map optimization method for industrial image anomaly detection, comprising the following steps: For the initial anomaly score map to be optimized, suppress the influence of pseudo anomalies in the foreground area during anomaly detection to obtain an anomaly score map that suppresses foreground noise; Background anomalies are suppressed on the industrial image to be detected to obtain a label map for distinguishing the background and foreground areas. Based on the label map for distinguishing the background and foreground areas, the outliers in the background area in the anomaly score map are removed to obtain an optimized anomaly score map.

[0006] Furthermore, obtaining an initial anomaly score map to be optimized includes: For the industrial image data to be detected, an initial anomaly score map to be optimized is obtained through the existing anomaly detection method based on unsupervised or self-supervised learning; The anomaly detection method based on unsupervised or self-supervised learning obtains a distribution or representation model of a normal image by modeling and learning normal samples, and obtains an anomaly score map by calculating the difference between the sample to be detected and the modeled normal distribution or representation.

[0007] Furthermore, for the initial anomaly score map to be optimized, suppressing the influence of pseudo-anomalies in the foreground area during anomaly detection to obtain an anomaly score map that suppresses foreground noise includes: smoothing the anomaly scores of local areas of the initial anomaly score map to be optimized by using a Gaussian kernel of set size and variance to reduce the anomaly scores of pseudo-anomaly areas on the foreground or object, and obtaining an anomaly score map that suppresses foreground noise.

[0008] Furthermore, background anomalies are suppressed for the industrial image to be detected to obtain a label map for distinguishing the background and foreground areas. Based on the label map for distinguishing the background and foreground areas, outliers in the background area in the anomaly score map are removed, and the optimized anomaly score map includes: Use pre-designed text prompt words that are independent of the category, and input the images to be detected into the trained semantic segmentation model to obtain multi-label images related to the foreground prompt words. And the multi-label graph related to the object prompt word ; The value of the relevant area in the multi-labeled image is 1, and the value of the irrelevant area is 0; the multi-labeled image is merged by taking the union method to obtain the initial non-background area label, in which the non-background area value is 1 and the background area value is 0; The initial non-background area is processed as follows: the marked area that is not a foreground or object is removed, the hole noise area is removed, and a dilation operation is performed to obtain a marked map for distinguishing the background and foreground areas; The marker map used to distinguish the background and foreground areas is multiplied element by element with the anomaly score map that suppresses foreground noise to obtain the optimized anomaly score map.

[0009] Furthermore, based on the technical concept of the above method, the present invention also provides an anomaly score map optimization system for industrial image anomaly detection, including a foreground noise suppression module and a background anomaly suppression module. The foreground noise suppression module is used to suppress the pseudo-anomaly influence of the foreground area in the anomaly detection process for the initial anomaly score map to be optimized, and obtain an anomaly score map with suppressed foreground noise; the background anomaly suppression module is used to suppress background anomalies in the industrial image to be detected to obtain a labeled map for distinguishing between background and foreground areas. Based on the labeled map for distinguishing between background and foreground areas, the outliers in the background area in the anomaly score map are removed to obtain an optimized anomaly score map.

[0010] Furthermore, the foreground noise suppression module input is the initial anomaly score map to be optimized , the output is an anomaly score map that suppresses foreground noise , calculated as follows:

[0011] in, is the Gaussian filter function, and the Gaussian kernel size is , with a variance of 5.

[0012] Furthermore, the background anomaly suppression module includes a non-background marked area extraction unit and a background anomaly elimination unit; the non-background marked area extraction unit is provided with a pre-trained semantic segmentation model, the industrial image to be detected is used as the input of the semantic segmentation model, and the pre-designed text prompt words that are not related to the category are used to input the images to be detected into the trained semantic segmentation model respectively, and the multi-label images related to the foreground prompt words are obtained respectively. And the multi-label graph related to the object prompt word ; The value of the relevant area in the multi-labeled image is 1, and the value of the irrelevant area is 0; the multi-labeled image is merged by taking the union method to obtain the initial non-background area label, in which the non-background area value is 1 and the background area value is 0; The initial non-background area is processed as follows: the non-foreground or object marked area is removed, the hole noise area is removed, and the dilation operation is performed to obtain the marked map used to distinguish the background and foreground areas. ;

[0013] The background anomaly removal unit uses the extracted marker map for distinguishing the background and foreground areas. , a marker map used to distinguish between background and foreground areas Anomaly score map with suppressed foreground noise Perform element-by-element multiplication to obtain the optimized anomaly score map

[0014] .

[0015] When the anomaly score map optimization method for industrial image anomaly detection of the present invention is applied, the anomaly score map output by the anomaly detection method is optimized to obtain an optimized anomaly score map, and the optimized anomaly score map is used as the input of the industrial image anomaly recognition and positioning method.

[0016] The present invention can also provide a computer device, including a processor and a memory, the memory is used to store a computer executable program, the processor reads the computer executable program from the memory and executes it, and when the processor executes the computer executable program, the anomaly score map optimization method for industrial image anomaly detection described in the present invention can be implemented.

[0017] At the same time, a computer-readable storage medium is provided, in which a computer program is stored. When the computer program is executed by a processor, the anomaly score map optimization method for industrial image anomaly detection according to the present invention can be implemented.

[0018] Compared with the prior art, the present invention has at least the following beneficial effects: The present invention provides an anomaly score map optimization method for industrial image anomaly detection. By suppressing foreground noise and background anomaly, the anomaly scores of pseudo-anomaly areas on the foreground and background of the sample to be detected in the anomaly score map can be suppressed respectively, thereby solving the shortcomings of the anomaly score map extracted by the existing anomaly detection method based on unsupervised or self-supervised learning, and improving the ability to identify / locate abnormal areas. The method provided by the present invention is easy to use and can be used as a plug-in in a plug-in form. It is not limited to a specific anomaly detection method based on unsupervised or self-supervised learning, and can effectively gain benefits for anomaly score maps extracted by existing methods; the anomaly score map obtained by the upstream anomaly detection method can be directly used as input, and the output result can be directly used as input for downstream anomaly positioning or identification methods.

[0019] The method provided by the present invention itself does not require additional training process, is independent of specific product categories, and does not require separate processing of single-category product images. It can be widely used in actual industrial scenarios with a variety of product categories.

[0020] Figure content Figure 1 The present invention is a schematic diagram of an implementable process.

[0021] Figure 2 This is the architecture diagram of the anomaly score graph optimization method for industrial image anomaly detection proposed in the present invention.

[0022] Figure 3This is the framework diagram of the anomaly score map extraction module.

[0023] Figure 4 This is the framework diagram of the foreground noise suppression module.

[0024] Figure 5 This is the framework diagram of the non-background marked area extraction unit in the background anomaly suppression module.

[0025] Figure 6 This is the framework diagram of the background anomaly elimination unit in the background anomaly suppression module.

[0026] Figure 7 This is the effect diagram of the anomaly score map optimization method for industrial image anomaly detection proposed in the present invention. DETAILED DESCRIPTION

[0027] The technical solution of the present invention is described in detail below with reference to the accompanying drawings.

[0028] This paper proposes an anomaly score map optimization method for industrial image anomaly detection, which further optimizes the anomaly score map extracted by the existing anomaly detection method based on unsupervised or self-supervised learning, thereby improving the ability to identify / locate abnormal areas. Figure 2 As shown, in order to reduce the influence of pseudo-abnormal areas existing in the sample to be detected, the present invention designs a foreground noise suppression module and a background abnormality suppression module, and obtains the abnormality score map to be optimized through the abnormality score map extraction module composed of the existing abnormality detection method based on unsupervised or self-supervised learning; then the abnormality score of the pseudo-abnormal on the foreground area is reduced by the foreground noise suppression module, thereby suppressing the pseudo-abnormal influence of the foreground area during the abnormality detection process; finally, the background abnormality suppression module generates an optimization mark related to the image foreground or object area, and further optimizes the abnormality score map so that it only focuses on the abnormal area in the image foreground or object area, thereby suppressing the pseudo-abnormal influence of the background area. The present invention does not depend on the product category, does not require additional training, is not limited to a specific abnormality detection method based on unsupervised or self-supervised learning, and can be used as a plug-in.

[0029] As attached Figure 1 As shown, an anomaly score map optimization method for industrial image anomaly detection specifically includes the following steps: S1, obtain the anomaly score map to be optimized As attached Figure 3 As shown in the figure, for the industrial image data to be detected, the initial anomaly score map to be optimized is obtained through the anomaly score map extraction module, which is recorded as ; The anomaly score map extraction module is used to implement the existing anomaly detection method based on unsupervised or self-supervised learning, but is not limited to a specific method; the anomaly detection method based on unsupervised or self-supervised learning is to obtain a distribution or representation model of a normal image by modeling and learning a large number of normal samples, and to obtain an anomaly score map for identifying and locating abnormal areas by calculating the difference between the sample to be detected and the modeled normal distribution or representation.

[0030] S2, after foreground noise suppression For the initial anomaly score map to be optimized , through the designed foreground noise suppression module, the pseudo-anomaly influence of the foreground area in the anomaly detection process is suppressed, and the anomaly score map of suppressed foreground noise is obtained .

[0031] As attached Figure 4 As shown, the foreground noise suppression module includes a Gaussian filter, which smoothes the anomaly score of the local area through a Gaussian kernel with a set size and variance, thereby reducing the anomaly score of the pseudo-anomaly area on the foreground or object. The input of the foreground noise suppression module is the initial anomaly score map to be optimized , the output is an anomaly score map that suppresses foreground noise , the calculation of the whole process is as follows:

[0032] in, is a Gaussian filter function. In the specific implementation, the Gaussian kernel size can be set to , with a variance of 5.

[0033] S3, optimized for background anomaly suppression For the industrial image to be detected described in S1 above and the anomaly score map for suppressing foreground noise described in S2 above , and further optimized by the background anomaly suppression module, so that the obtained anomaly score map We pay more attention to the anomalies in the foreground or object area, and suppress the influence of pseudo anomalies in the background area. The optimized anomaly score is The background anomaly suppression module includes a non-background marking area extraction unit and a background anomaly elimination unit; As attached Figure 5 As shown, the industrial image to be detected in S1 is input into the non-background marked area extraction unit, and a marked image for distinguishing the background and foreground areas is output. ; The non-background marked area extraction unit is provided with a pre-trained semantic segmentation model Ground-SAM, which is based on the paper Ren T, Liu S, Zeng A, et al. Groundedsam: Assembling open-world models for diverse visual tasks[J]. arXiv preprintarXiv:2401.14159, 2024. The specific steps of the marking process are as follows: First, we use pre-designed text prompt words that are not related to categories, such as foreground and object, and input the images to be detected into the semantic segmentation model Ground-SAM to obtain multi-label images related to the foreground prompt words. And the multi-label graph related to the object prompt word ; The value of the relevant area in the multi-label map is 1, and the value of the irrelevant area is 0; Then, the above multi-labeled images are merged by taking the union method to obtain the initial non-background area label, where the non-background area value is 1 and the background area value is 0.

[0034] Finally, the initial non-background area is further marked. The entire marking process includes the following image operations: 1) : Remove small marked areas, which are usually marked areas that are not foreground or objects; 2) : Remove small hole noise areas, which usually come from the error interference caused by the semantic segmentation model itself; 3) :Use the dilation operation to further ensure that all non-background areas are within the marker. The above image processing operations can be implemented using standard functions in the open-cv library under Python.

[0035] Finally, we get a label map for distinguishing the background and foreground areas. , the calculation process is as follows:

[0036] As attached Figure 6 As shown, the background anomaly elimination unit uses the extracted marker image for distinguishing the background and foreground areas. , further reducing the anomaly score in the graph The anomaly score in the background area. Specifically, the marker map used to distinguish between background and foreground areas Anomaly score map with suppressed foreground noise Perform element-by-element multiplication to remove outliers in the background area of ​​the anomaly score map, so that the optimized anomaly score map Focusing on the foreground or object area where the real anomaly exists, the calculation process is as follows:

[0037] S4, obtain the optimized anomaly score map Use the anomaly score map obtained in step S3 Replace the original anomaly score map to be optimized , used to improve identification / location of abnormal areas.

[0038] As attached Figure 7 As shown in the figure, the anomaly score map optimization method for industrial image anomaly detection proposed by the present invention is given, showing the optimization effects on normal samples, small anomaly area samples and large anomaly area samples respectively. Relative to the original anomaly score map to be optimized The anomaly scores of the pseudo-anomaly areas in the foreground and background areas are effectively suppressed, while the anomaly scores of the real anomaly areas are less affected, which shows that the anomaly score map optimization method for industrial image anomaly detection proposed in the present invention has good performance.

[0039] In addition, the present invention is independent of product categories, does not require additional training, is not limited to a specific anomaly detection method based on unsupervised or self-supervised learning, and can be used as a plug-in in a plug-in-and-play manner.

[0040] In summary, the present invention proposes an anomaly score map optimization method for industrial image anomaly detection, mainly designs a foreground noise suppression module and a background anomaly suppression module. The present invention uses the anomaly score map extracted by the existing anomaly detection method based on unsupervised or self-supervised learning as input, and then the foreground noise suppression module suppresses the pseudo-anomaly influence of the foreground area in the anomaly detection process by reducing the anomaly score of the pseudo-anomaly on the foreground area; the background anomaly suppression module generates an optimization mark related to the image foreground or object area, so that the optimized anomaly score map only focuses on the image foreground or object area, thereby suppressing the pseudo-anomaly influence of the background area. In addition, the present invention is independent of product category, does not require additional training, is not limited to a specific anomaly detection method based on unsupervised or self-supervised learning, and can be used as a plug-in. The present invention further optimizes the anomaly score map extracted by the anomaly detection method based on unsupervised or self-supervised learning, thereby improving the ability to identify / locate abnormal areas.

[0041] Embodiment 2, based on the technical concept of the above method, the present invention also provides an anomaly score map optimization system for industrial image anomaly detection, including a foreground noise suppression module and a background anomaly suppression module. The foreground noise suppression module is used to suppress the pseudo-anomaly influence of the foreground area in the anomaly detection process for the initial anomaly score map to be optimized, and obtain an anomaly score map with suppressed foreground noise; the background anomaly suppression module is used to suppress background anomalies in the industrial image to be detected to obtain a labeled map for distinguishing between background and foreground areas. Based on the labeled map for distinguishing between background and foreground areas, the outliers in the background area in the anomaly score map are removed to obtain an optimized anomaly score map.

[0042] The input of the foreground noise suppression module is the initial anomaly score map to be optimized , the output is an anomaly score map that suppresses foreground noise , calculated as follows:

[0043] in, is the Gaussian filter function, and the Gaussian kernel size is , with a variance of 5.

[0044] The background anomaly suppression module includes a non-background marking area extraction unit and a background anomaly elimination unit; The non-background labeled area extraction unit is provided with a pre-trained semantic segmentation model. The industrial image to be detected is used as the input of the semantic segmentation model. The pre-designed text prompt words that are not related to the category are used to input the images to be detected into the trained semantic segmentation model respectively, and the multi-label images related to the foreground prompt words are obtained respectively. And the multi-label graph related to the object prompt word ; The value of the relevant area in the multi-labeled image is 1, and the value of the irrelevant area is 0; the multi-labeled image is merged by taking the union method to obtain the initial non-background area label, in which the non-background area value is 1 and the background area value is 0; The initial non-background area is processed as follows: the non-foreground or object marked area is removed, the hole noise area is removed, and the dilation operation is performed to obtain the marked map used to distinguish the background and foreground areas. ;

[0045] The background anomaly removal unit uses the extracted marker map for distinguishing the background and foreground areas. , a marker map used to distinguish between background and foreground areas Anomaly score map with suppressed foreground noise Perform element-by-element multiplication to obtain the optimized anomaly score map

[0046] .

[0047] On the other hand, the present invention provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the anomaly score map optimization method for industrial image anomaly detection described in the present invention can be implemented.

[0048] The present invention may also provide a computer device, including a processor and a memory, the memory being used to store a computer executable program, the processor reading the computer executable program from the memory and executing it, and the processor being able to implement the anomaly score map optimization method for industrial image anomaly detection described in the present invention when executing the computer executable program.

[0049] The computer device may be a laptop computer, a desktop computer or a workstation.

[0050] The processor can be a central processing unit (CPU), a graphics processing unit (GPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), or an off-the-shelf field-programmable gate array (FPGA).

[0051] The memory of the present invention may be an internal storage unit of a laptop computer, a desktop computer or a workstation, such as a memory or a hard disk; or an external storage unit, such as a mobile hard disk or a flash memory card.

[0052] Computer-readable storage media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer-readable storage media may include: read-only memory (ROM), random access memory (RAM), solid-state drive (SSD) or optical disk, etc. Among them, random access memory may include resistance random access memory (ReRAM) and dynamic random access memory (DRAM).

[0053] The present invention is described by way of embodiments, and various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention for those skilled in the art. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application belong to the protection scope of the present invention.

Claims

1. An anomaly score map optimization method for industrial image anomaly detection, characterized in that: The following steps are involved: For the initial anomaly score map to be optimized, suppress the influence of pseudo anomalies in the foreground area during anomaly detection to obtain an anomaly score map that suppresses foreground noise; Background anomalies are suppressed on the industrial image to be detected to obtain a label map for distinguishing the background and foreground areas. Based on the label map for distinguishing the background and foreground areas, the outliers in the background area in the anomaly score map are removed to obtain an optimized anomaly score map.

2. The anomaly score map optimization method for industrial image anomaly detection according to claim 1, characterized in that: Obtaining the initial anomaly score map to be optimized includes: For the industrial image data to be detected, an initial anomaly score map to be optimized is obtained through the existing anomaly detection method based on unsupervised or self-supervised learning; The anomaly detection method based on unsupervised or self-supervised learning obtains a distribution or representation model of a normal image by modeling and learning normal samples, and obtains an anomaly score map by calculating the difference between the sample to be detected and the modeled normal distribution or representation.

3. The anomaly score map optimization method for industrial image anomaly detection according to claim 1, characterized in that: For the initial anomaly score map to be optimized, suppressing the influence of pseudo anomalies in the foreground area during anomaly detection to obtain an anomaly score map that suppresses foreground noise includes: smoothing the anomaly scores of local areas of the initial anomaly score map to be optimized through a Gaussian kernel with a set size and variance, reducing the anomaly scores of pseudo anomaly areas on the foreground or object, and obtaining an anomaly score map that suppresses foreground noise.

4. The anomaly score map optimization method for industrial image anomaly detection according to claim 1, characterized in that: Background anomaly suppression is performed on the industrial image to be detected to obtain a label map for distinguishing the background and foreground areas. Based on the label map for distinguishing the background and foreground areas, the outliers in the background area in the anomaly score map are removed, and the optimized anomaly score map includes: Use pre-designed text prompt words that are independent of the category, and input the images to be detected into the trained semantic segmentation model to obtain multi-label images related to the foreground prompt words. And the multi-label graph related to the object prompt word ; The value of the relevant area in the multi-labeled image is 1, and the value of the irrelevant area is 0; the multi-labeled image is merged by taking the union method to obtain the initial non-background area label, in which the non-background area value is 1 and the background area value is 0; The initial non-background area is processed as follows: the marked area that is not a foreground or object is removed, the hole noise area is removed, and a dilation operation is performed to obtain a marked map for distinguishing the background and foreground areas; The marker map used to distinguish the background and foreground areas is multiplied element by element with the anomaly score map that suppresses foreground noise to obtain the optimized anomaly score map.

5. An anomaly score map optimization system for industrial image anomaly detection, characterized in that: It includes a foreground noise suppression module and a background anomaly suppression module. The foreground noise suppression module is used to suppress the pseudo-anomaly influence of the foreground area in the anomaly detection process for the initial anomaly score map to be optimized, and obtain an anomaly score map suppressing the foreground noise; The background anomaly suppression module is used to suppress background anomalies of the industrial image to be detected to obtain a label map for distinguishing the background and foreground areas. Based on the label map for distinguishing the background and foreground areas, the outliers in the background area in the anomaly score map are removed to obtain an optimized anomaly score map.

6. The anomaly score map optimization system for industrial image anomaly detection according to claim 5, characterized in that: The input of the foreground noise suppression module is the initial anomaly score map to be optimized , the output is an anomaly score map that suppresses foreground noise , calculated as follows: in, is the Gaussian filter function.

7. The anomaly score map optimization system for industrial image anomaly detection according to claim 5, characterized in that: The background anomaly suppression module includes a non-background marking area extraction unit and a background anomaly elimination unit; The non-background labeled area extraction unit is provided with a pre-trained semantic segmentation model. The industrial image to be detected is used as the input of the semantic segmentation model. The pre-designed text prompt words that are not related to the category are used to input the images to be detected into the trained semantic segmentation model respectively, and the multi-label images related to the foreground prompt words are obtained respectively. And the multi-label graph related to the object prompt word ; The value of the relevant area in the multi-labeled image is 1, and the value of the irrelevant area is 0; the multi-labeled image is merged by taking the union method to obtain the initial non-background area label, in which the non-background area value is 1 and the background area value is 0; The initial non-background area is processed as follows: the non-foreground or object marked area is removed, the hole noise area is removed, and the dilation operation is performed to obtain the marked map used to distinguish the background and foreground areas. ; The background anomaly removal unit uses the extracted marker map for distinguishing the background and foreground areas. , a marker map used to distinguish between background and foreground areas Anomaly score map with suppressed foreground noise Perform element-by-element multiplication to obtain the optimized anomaly score map 。 8. The application of the anomaly score map optimization method for industrial image anomaly detection according to any one of claims 1 to 4, characterized in that: The anomaly score map output by the anomaly detection method is optimized to obtain an optimized anomaly score map, and the optimized anomaly score map is used as the input of the industrial image anomaly recognition and positioning method.

9. A computer device, characterized in that: It includes a processor and a memory, the memory is used to store a computer executable program, the processor reads part or all of the computer executable program from the memory and executes it, and when the processor executes part or all of the computer executable program, the anomaly score map optimization method for industrial image anomaly detection according to any one of claims 1 to 7 can be implemented.

10. A computer-readable storage medium, characterized in that: A computer program is stored in a computer-readable storage medium. When the computer program is executed by a processor, the anomaly score map optimization method for industrial image anomaly detection according to any one of claims 1 to 7 can be implemented.

Images