Panel zero-shot defect detection method and system, device, and storage medium

Abstract

The present disclosure relates to the technical field of industrial defect detection, and provides a panel zero-shot defect detection method and system, a device, and a storage medium. The process flow of the method comprises: performing feature extraction on a panel image on the basis of a deep learning network, so as to obtain feature vectors of all feature points of the panel image; performing similarity calculation on all the feature points on the basis of the feature vectors of all the feature points of the panel image, so as to obtain similarity values of feature point pairs; performing anomaly calculation on all the feature points on the basis of the similarity values of the feature point pairs, so as to construct a feature anomaly map regarding the feature points; and performing anomaly region segmentation on the panel image on the basis of the feature anomaly map of the feature points, so as to obtain a defect region of the panel image. The present disclosure requires no model training, enables direct detection of defect regions, achieves high accuracy and efficiency in defect detection, requires no manual labeling, has low detection costs, and is suitable for low-incidence, zero-shot defect detection.

Description

A method, system, device and storage medium for zero-sample defect detection of panels

[0001] Cross-reference to related applications

[0002] This disclosure claims priority to Chinese Patent Application No. 2024114266292, filed on October 14, 2024, entitled "A Method, System, Device and Storage Medium for Zero-Sample Defect Detection of Panels", the entire contents of which are incorporated herein by reference. Technical Field

[0003] This disclosure relates to the field of industrial defect detection technology, and more specifically, to a method, system, device, and storage medium for zero-sample defect detection of panels. Background Technology

[0004] In the panel manufacturing process, for panel images with regular features, such as Array, CF, and OLED, existing technologies mainly capture anomalies through sliding matching. This method requires confirming the periodic pixel values ​​of the matching and correcting for issues such as image periodic tilt, which can easily lead to over-detection. Meanwhile, existing technologies can also use supervised deep learning training to learn defective regions from panel images. This method requires labeling a large number of training samples, which is costly due to manual labeling.

[0005] Application content

[0006] To address the problem of over-detection when performing defect detection on panel images with regular features, this disclosure provides a panel zero-sample defect detection method, system, device, and storage medium.

[0007] This disclosure provides a method for detecting zero-sample defects in a panel, the method comprising the following steps:

[0008] Feature extraction of panel images is performed based on deep learning networks to obtain feature vectors of all feature points in the panel images;

[0009] The similarity of all feature points is calculated based on the feature vectors of all feature points in the panel image to obtain the similarity value of the feature point pairs.

[0010] Anomaly calculations are performed on all feature points based on the similarity values ​​of feature point pairs to construct a feature anomaly map of the feature points.

[0011] The panel image is segmented into defective regions based on the feature anomaly map of feature points.

[0012] In the above embodiments, this disclosure is based on the cyclical characteristics of the panel manufacturing process. First, it uses a neural network to extract deep semantic features, calculates the similarity between each feature point and the remaining feature points, then determines the outlier value of the current feature point to form a feature anomaly map, and finally uses a statistical algorithm to statistically analyze the feature anomaly map to obtain the abnormal region, that is, to detect the defect region in the panel image. No model training is required, and defect region detection can be performed directly. The accuracy and efficiency of defect detection are high, no manual annotation is required, and the detection cost is low. It is suitable for defect detection with low occurrence rate and zero sample size.

[0013] As some optional embodiments of this disclosure, the feature vector of all feature points of the panel image is represented as: G = {F} ij |i=1,...,w g j = 1, ..., h g}

[0014] Where G represents the set of eigenvectors of all feature points, and F... ij Let i represent the feature vector, where i represents the width corresponding to the location of the feature point, and j represents the height corresponding to the location of the feature point.

[0015] In the above embodiments, this disclosure automatically extracts hierarchical and abstract feature vectors from the original panel image using models such as convolutional neural networks. These feature vectors are then passed and optimized layer by layer to form a set of all feature vectors of the current panel image.

[0016] As some optional embodiments of this disclosure, the feature vector includes image information of the region surrounding the feature point; wherein, the image information of the region surrounding the feature point includes color, texture and shape.

[0017] In the above embodiments, this disclosure can quickly and accurately detect anomalies in feature points by analyzing the image information of the area surrounding the feature points.

[0018] As some optional embodiments of this disclosure, the similarity is calculated using distance metrics and vector similarity methods.

[0019] In the above embodiments, this disclosure uses distance measurement and vector similarity measurement to calculate similarity. The distance measurement methods include, but are not limited to, Euclidean distance, Manhattan distance, Minkowski distance, etc. The most similar feature point pairs can be quickly obtained through similarity calculation.

[0020] As some optional implementations of this disclosure, the process of constructing a feature anomaly map about feature points by performing anomaly calculation on all feature points based on the similarity value of feature point pairs is as follows:

[0021] Based on the similarity value between each feature point and the remaining feature points, similarity matching is performed on the feature points to obtain the k feature point pairs with the highest similarity matching values;

[0022] The outlier value of each feature point is calculated based on the similarity values ​​of the k feature point pairs with the highest similarity matching values, wherein the outlier value is the mean of the similarity values ​​of the k feature point pairs;

[0023] An image is constructed based on the outliers of all feature points in the panel image to obtain a feature anomaly map of all feature points.

[0024] In the above embodiments, this disclosure constructs a feature anomaly map based on the outlier values ​​of all feature points. The feature anomaly map can reflect the abnormal areas of the panel image, thereby achieving defect detection.

[0025] As some optional embodiments of this disclosure, the formula for calculating the outlier is:

[0026] Among them, s ij F represents outliers of feature points, k represents the number of feature point pairs, and F represents the number of feature point pairs. ij Let f represent the eigenvector, and let f represent the eigenvector of the feature point with respect to the remaining feature points.

[0027] In the above embodiments, the outlier value of the feature point disclosed herein is the mean of the similarity values ​​of the k most similar feature point pairs. Using the mean value method can reduce the interference of noise on the feature point anomaly detection. At the same time, by selecting only the similarity values ​​of the k most similar feature point pairs related to the feature point for outlier value calculation, the parameters can be simplified and the efficiency of anomaly detection can be improved.

[0028] As some optional embodiments of this disclosure, the feature anomaly map is upsampled to the panel image size using linear interpolation and noise points are removed using Gaussian filtering.

[0029] In the above embodiments, this disclosure upsamples the feature anomaly map to the size of the panel image, which facilitates the localization of the anomaly region, and the Gaussian filtering method can prevent noise points from interfering with anomaly detection.

[0030] As some optional embodiments of this disclosure, the process of segmenting abnormal regions of a panel image based on feature anomaly maps of feature points to obtain defective regions of the panel image is as follows:

[0031] The mean-standard-deviation statistical method was used to detect all anomalous feature points in the feature anomaly map;

[0032] Based on the location information of all abnormal feature points in the feature anomaly map, mark the defect area in the panel image.

[0033] In the above embodiments, the present disclosure uses the mean standard deviation statistical method to quickly obtain abnormal feature points, and then marks the defect area from the panel image.

[0034] This disclosure also provides a panel zero-sample defect detection system, the system comprising:

[0035] The feature point extraction unit extracts features from the panel image based on a deep learning network to obtain feature vectors of all feature points in the panel image.

[0036] A similarity calculation unit performs similarity calculation on all feature points based on the feature vectors of all feature points in the panel image to obtain the similarity value of feature point pairs.

[0037] An anomaly calculation unit performs anomaly calculations on all feature points based on the similarity values ​​of feature point pairs to construct a feature anomaly map about the feature points.

[0038] An abnormal region segmentation unit performs abnormal region segmentation on the panel image based on the feature abnormal map of feature points to obtain the defective regions of the panel image.

[0039] This disclosure also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor performs the panel zero-sample defect detection method.

[0040] This disclosure also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the panel zero-sample defect detection method.

[0041] The beneficial effects of this disclosure are as follows: Based on the cyclical characteristics of the panel manufacturing process, this disclosure first extracts deep semantic features by using a neural network, calculates the similarity between each feature point and the remaining feature points, then determines the outlier value of the current feature point to form a feature anomaly map, and finally uses a statistical algorithm to statistically analyze the feature anomaly map to obtain the abnormal region, that is, to detect the defect region in the panel image. No model training is required, and defect region detection can be performed directly. The accuracy and efficiency of defect detection are high, no manual annotation is required, and the detection cost is low. It is suitable for defect detection with low occurrence rate and zero sample size. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of this disclosure and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 is a schematic diagram of the computer device structure of the hardware operating environment described in the embodiments of this disclosure;

[0044] Figure 2 is a flowchart of the panel zero-sample defect detection method according to an embodiment of this disclosure;

[0045] Figure 3 is a schematic diagram of the panel image described in an embodiment of this disclosure;

[0046] Figure 4 is a schematic diagram of the feature anomaly map described in an embodiment of this disclosure;

[0047] Figure 5 is a structural block diagram of the panel zero-sample defect detection system according to an embodiment of this disclosure. Detailed Implementation

[0048] It should be understood that the specific embodiments described herein are merely illustrative of this disclosure and are not intended to limit this disclosure.

[0049] To address the issue of over-detection when performing defect detection on panel images with regular features, this disclosure provides a panel zero-sample defect detection method, system, device, and storage medium. Before introducing the specific technical solutions of this disclosure, the hardware operating environment involved in the embodiments of this disclosure will be described first.

[0050] Please refer to Figure 1, which is a schematic diagram of the computer device structure of the hardware operating environment involved in the embodiments of this disclosure.

[0051] As shown in Figure 1, the computer device may include: a processor, such as a central processing unit (CPU), a communication bus, a user interface, a network interface, and memory. The communication bus is configured to enable communication between these components. The user interface may include a display screen and an input unit such as a keyboard; optionally, the user interface may also include a standard wired interface or a wireless interface. The network interface may be a wired interface or a wireless interface (such as a Wi-Fi interface). The memory may be high-speed random access memory (RAM) or stable non-volatile memory (NVM), such as a disk storage device; optionally, the memory may also be a storage device independent of the aforementioned processor.

[0052] Those skilled in the art will understand that the structure shown in Figure 1 does not constitute a limitation on the computer device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0053] As shown in Figure 1, a memory, which serves as a storage medium, may include an operating system module, a network communication module, a user interface module, and a data storage module.

[0054] In the computer device shown in Figure 1, the network interface is mainly configured to communicate with the network server; the user interface is mainly configured to interact with the user; the processor and memory in the computer device of this disclosure can be set in the computer device, and the computer device calls the panel zero-sample defect detection system stored in the data storage module through the processor and executes the panel zero-sample defect detection method provided in the embodiments of this disclosure.

[0055] Based on the hardware environment of the foregoing embodiments, embodiments of this disclosure provide a panel zero-sample defect detection method. Please refer to Figure 2, which is a flowchart of the panel zero-sample defect detection method. The method flow is as follows:

[0056] (1) Input the panel image to be detected into a deep learning network, and then perform feature extraction on the panel image based on the deep learning network to obtain the feature vectors of all feature points of the panel image, and form the feature set of the current panel image, G={F ij |i=1,...,w g j = 1, ..., h g}; where G represents the set of eigenvectors of all feature points, F ij Let i represent a feature vector of dimension m, where i represents the width corresponding to the location of the feature point, and j represents the height corresponding to the location of the feature point; wherein, the deep neural network includes, but is not limited to, convolutional neural networks (CNN), deep residual networks (Res-Net), feature pyramid networks (FPN), etc.; optionally, embodiments of this disclosure use convolutional neural networks to extract features from panel images.

[0057] In the embodiments of this disclosure, the feature points of the panel image are usually points that are relatively prominent or have regularity in the panel image, such as corner points, edge points, and extreme points.

[0058] In this embodiment of the disclosure, a feature vector of dimension m is a vector containing m values ​​or elements that describe the measurement of a feature point on certain features or attributes. Optionally, the feature vector includes image information of the region surrounding the feature point, wherein the image information of the region surrounding the feature point includes, but is not limited to, color, texture, and shape.

[0059] (2) Calculate the similarity of all feature points based on the feature vectors of all feature points in the panel image to obtain the similarity value of the feature point pairs.

[0060] In this embodiment of the disclosure, the similarity calculation method includes distance measurement and vector similarity measurement. The distance measurement method includes, but is not limited to, Euclidean distance, Manhattan distance, Minkowski distance, etc.; the vector similarity measurement method includes, but is not limited to, cosine similarity, etc. Optionally, this embodiment of the disclosure uses Euclidean distance to calculate the similarity of all feature points. Optionally, the feature point pair represents the combination of each feature point and each of the remaining feature points.

[0061] Optionally, the Euclidean distance is calculated using the following formula:

[0062] Where D(X,Y) represents the Euclidean distance between eigenvectors X and Y, X=(x1,x2,...,x m ) and Y = (y1, y2, ..., y m ) represents two feature vectors of dimension m.

[0063] (3) Perform anomaly calculation on all feature points based on the similarity values ​​of feature point pairs to construct a feature anomaly map of the feature points.

[0064] In this embodiment of the disclosure, the process for calculating anomalies of all feature points is as follows:

[0065] (3.1) Perform similarity matching on the feature points based on the similarity value between each feature point and the remaining feature points to obtain the k feature point pairs with the highest similarity matching value; wherein, the similarity matching is to quickly find the k most similar feature point pairs by performing a fast approximate nearest neighbor search.

[0066] (3.2) Calculate the outlier value of each feature point based on the similarity values ​​of the k feature point pairs with the highest similarity matching values, wherein the outlier value is the mean of the similarity values ​​of the k feature point pairs.

[0067] (3.3) Construct an image based on the outliers of all feature points in the panel image to obtain a feature anomaly map of all feature points, where the size of the feature anomaly map is w. g *h g .

[0068] Optionally, the formula for calculating the outlier is as follows:

[0069] Among them, s ij F represents outliers of feature points, k represents the number of feature point pairs, and F represents the number of feature point pairs. ij Let f represent the eigenvector, and let f represent the eigenvector of the feature point with respect to the remaining feature points.

[0070] In this embodiment of the disclosure, the construction process of the feature anomaly map is as follows:

[0071] ① Initialize vertices: Each vertex corresponds to a feature point.

[0072] ② Construct undirected edges and set weights: For any two vertices in the feature anomaly graph, construct an undirected edge and set the weight of the edge. The weight of the undirected edge is the similarity between the corresponding feature point pairs.

[0073] ③ Construct self-looping edges and set weights: For vertices in the feature anomaly graph, set their weights to the anomaly values ​​of the feature points.

[0074] (4) Based on the feature anomaly map of feature points, the panel image is segmented into anomaly regions to obtain the defect regions of the panel image.

[0075] In this embodiment of the disclosure, the process of segmenting the abnormal region is as follows:

[0076] (4.1) The mean and standard deviation statistical method is used to detect all abnormal feature points in the feature anomaly map; Optionally, a parameter value p is set, and when the pixel value of the corresponding feature point is greater than μ + pσ, the feature point is determined to be abnormal, where μ is the mean of the pixel values ​​of all feature points and σ is the standard deviation of the pixel values ​​of all feature points; Optionally, the parameter value p can be set according to the actual situation, and the specific value of the parameter value p is not limited in this embodiment.

[0077] (4.2) Mark the defect areas in the panel image based on the location information of all abnormal feature points in the feature anomaly map. The feature anomaly map is upsampled to the size of the panel image using linear interpolation and noise points are removed using Gaussian filtering. Please refer to Figure 3, which is a schematic diagram of the panel image, and Figure 4, which is a schematic diagram of the feature anomaly map.

[0078] Furthermore, in order to further improve the accuracy of defect detection, this embodiment of the present disclosure, based on the above-mentioned defect detection, performs secondary verification on the feature anomaly map, which can greatly improve the accuracy of defect detection.

[0079] In this embodiment of the disclosure, the process for secondary verification of the feature anomaly map is as follows:

[0080] (4.3) Construct a feature anomaly map of the normal panel image based on the normal panel image; wherein, the construction process of the feature anomaly map of the normal panel image can refer to the construction process of the feature anomaly map of the panel image to be detected described above.

[0081] (4.4) The pixel values ​​of the feature anomaly map corresponding to the normal panel image and the feature anomaly map corresponding to the image to be detected are subtracted to obtain the final anomaly MAP map; wherein the anomaly MAP map no longer contains background interference of the panel image.

[0082] (4.5) Mark the defect area in the panel image based on the location information of all feature points of the anomalies in the final anomaly MAP.

[0083] In this embodiment of the disclosure, an abnormal MAP map without background interference can be obtained by subtracting the pixel values ​​of the feature abnormal map corresponding to the normal panel image and the feature abnormal map corresponding to the image to be detected. The abnormal MAP map can be used to more accurately determine and locate abnormalities, thereby enabling the acquisition of defect areas from the panel image to be detected.

[0084] In summary, based on the cyclical characteristics of the panel manufacturing process, this embodiment first extracts deep semantic features using a neural network, calculates the similarity between each feature point and the remaining feature points, then determines the outlier value of the current feature point to form a feature anomaly map, and finally uses a statistical algorithm to statistically analyze the feature anomaly map to obtain the abnormal region, i.e., detect the defect region in the panel image. This embodiment does not require model training and can directly detect defect regions. The accuracy and efficiency of defect detection are high, no manual annotation is required, and the detection cost is low. It is suitable for defect detection with low incidence and zero samples.

[0085] Furthermore, in one embodiment, based on the same inventive concept as the foregoing embodiments, this disclosure provides a panel zero-sample defect detection system. Please refer to Figure 5, which is a structural block diagram of the panel zero-sample defect detection system. The system corresponds one-to-one with the method described in Embodiment 1. The system includes:

[0086] The feature point extraction unit extracts features from the panel image based on a deep learning network to obtain feature vectors of all feature points in the panel image.

[0087] A similarity calculation unit performs similarity calculation on all feature points based on the feature vectors of all feature points in the panel image to obtain the similarity value of feature point pairs.

[0088] An anomaly calculation unit performs anomaly calculations on all feature points based on the similarity values ​​of feature point pairs to construct a feature anomaly map about the feature points.

[0089] An abnormal region segmentation unit performs abnormal region segmentation on the panel image based on the feature abnormal map of feature points to obtain the defective regions of the panel image.

[0090] It should be noted that each unit in the panel zero-sample defect detection system in this embodiment corresponds one-to-one with each step in the panel zero-sample defect detection method in the aforementioned embodiment. Therefore, the specific implementation method and the technical effects achieved in this embodiment can be referred to the implementation method of the panel zero-sample defect detection method, and will not be repeated here.

[0091] Furthermore, in one embodiment, this disclosure also provides a computer device including a processor, a memory, and a computer program stored in the memory, the computer program being executed by the processor to implement the methods in the foregoing embodiments.

[0092] Furthermore, in one embodiment, this disclosure also provides a computer storage medium storing a computer program that is executed by a processor to implement the methods described in the foregoing embodiments.

[0093] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a device including one or any combination of the above-mentioned memories. The computer may be a variety of computing devices, including smart terminals and servers.

[0094] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0095] As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0096] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0097] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0098] The sequence numbers of the embodiments disclosed above are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0099] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a multimedia terminal device (which may be a mobile phone, computer, television receiver, or network device, etc.) to execute the methods described in the various embodiments of this disclosure.

[0100] The above are merely preferred embodiments of this disclosure and do not limit the patent scope of this disclosure. Any equivalent structural or procedural transformations made using the content of this disclosure and its drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this disclosure. Industrial applicability

[0101] This disclosure provides a panel zero-sample defect detection method, system, device, and storage medium. It can directly detect defect areas without model training, and has high accuracy and efficiency in defect detection. It does not require manual annotation, has low detection cost, and is suitable for defect detection with low occurrence rate and zero samples.

Claims

1. A method of panel zero-shot defect detection, the method comprising: The method includes the following steps: Feature extraction of panel images is performed based on deep learning networks to obtain feature vectors of all feature points in the panel images; The similarity of all feature points is calculated based on the feature vectors of all feature points in the panel image to obtain the similarity value of the feature point pairs. Anomaly calculations are performed on all feature points based on the similarity values ​​of feature point pairs to construct a feature anomaly map of the feature points. The panel image is segmented into defective regions based on the feature anomaly map of feature points.

2. The method of claim 1, wherein, The feature vector of all feature points of the panel image is represented as: G = {F ij |i = 1,..., w g ,j = 1,..., h g} where G represents a set of feature vectors of all feature points, F ij represents a feature vector, i represents a width corresponding to a position of a feature point, and j represents a height corresponding to the position of the feature point.

3. The method of claim 1, wherein, The feature vector contains image information of the region surrounding the feature point, including color, texture, and shape.

4. The method of claim 1, wherein, The similarity is calculated using both distance metrics and vector similarity methods.

5. The method of claim 1, wherein, The process of calculating anomalies for all feature points based on the similarity values ​​of feature point pairs to construct a feature anomaly map for the feature points is as follows: Based on the similarity value between each feature point and the remaining feature points, similarity matching is performed on the feature points to obtain the k feature point pairs with the highest similarity matching values; The outlier value of each feature point is calculated based on the similarity values ​​of the k feature point pairs with the highest similarity matching values, wherein the outlier value is the mean of the similarity values ​​of the k feature point pairs; An image is constructed based on the outliers of all feature points in the panel image to obtain a feature anomaly map of all feature points.

6. The method of panel zero-shot defect detection of claim 5, wherein, The formula for calculating the abnormal value is: where s ij represents the abnormal value of the feature point, k represents the number of feature point pairs, F ij represents the feature vector, and f represents the feature vector of the feature point pair remaining feature point.

7. The method of claim 5, wherein the method further comprises: The feature anomaly map is upsampled to the panel image size using linear interpolation, and noise points are removed using Gaussian filtering.

8. The method of claim 1, wherein, The process of segmenting abnormal regions in a panel image based on feature point-based feature anomaly maps to obtain the defective regions of the panel image is as follows: The mean-standard-deviation statistical method was used to detect all anomalous feature points in the feature anomaly map; Based on the location information of all abnormal feature points in the feature anomaly map, mark the defect area in the panel image.

9. A panel zero-shot defect detection system, comprising: The system includes: The feature point extraction unit extracts features from the panel image based on a deep learning network to obtain feature vectors of all feature points in the panel image. A similarity calculation unit performs similarity calculation on all feature points based on the feature vectors of all feature points in the panel image to obtain the similarity value of feature point pairs. An anomaly calculation unit performs anomaly calculations on all feature points based on the similarity values ​​of feature point pairs to construct a feature anomaly map about the feature points. An abnormal region segmentation unit performs abnormal region segmentation on the panel image based on the feature abnormal map of feature points to obtain the defective regions of the panel image.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the panel zero-sample defect detection method according to any one of claims 1-8.

11. A computer readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the panel zero-sample defect detection method according to any one of claims 1-8.

Images