Defect identification method and system based on semi-supervised learning and change detection

Abstract

The invention provides a defect identification method and system based on semi-supervised learning and change detection, and relates to the field of defect detection.The method comprises the steps that single-time-phase images at different times under the same shooting condition are obtained, and the single-time-phase images are preprocessed to construct a sample data set; constructing a defect identification model, and performing semi-supervised learning and change detection on the defect identification model based on the sample data set to obtain a final defect identification model; wherein the defect identification model comprises an unsupervised model and a supervised model; and performing defect identification on the single-time-phase image by adopting the final defect identification model to obtain a defect identification result. According to the method, the defects are detected in a semi-supervised learning and change detection mode, and the defects are detected in a change detection mode by collecting part of samples in the hardware defects, so that the manpower is reduced, and the defect identification and detection accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of defect detection, and in particular to a defect recognition method and system based on semi-supervised learning and change detection. Background Art

[0002] In the industrial hardware production process, defects are usually detected by human visual inspection. However, manual inspection is costly and subject to human subjective factors, which can easily lead to over-inspection or missed inspections.

[0003] To save manpower and improve the detection rate of industrial hardware defects, traditional visual inspection can be used for defect detection. However, due to its sensitivity to light, traditional visual inspection cannot fully adapt to machines with fluctuating light sources. Later, supervised learning-based neural network methods were proposed for defect detection. However, this method requires a large amount of defect data for learning, which is not available in the industrial hardware field. Furthermore, some defects may have never occurred before, making some supervised learning methods immobile and unable to match new defects. Summary of the Invention

[0004] The present invention provides a defect recognition method and system based on semi-supervised learning and change detection, which solves the problem that existing industrial hardware defect detection lacks sample data and the supervised learning part cannot be migrated, resulting in the model's lack of new defect matching ability.

[0005] In a first aspect, an embodiment of the present invention provides a defect recognition method based on semi-supervised learning and change detection, the method comprising the following process: Acquire single-phase images at different times under the same shooting conditions, and preprocess the single-phase images to construct a sample data set; Constructing a defect recognition model, and performing semi-supervised learning and change detection on the defect recognition model based on a sample data set to obtain a final defect recognition model; wherein the defect recognition model includes an unsupervised model and a supervised model; The final defect recognition model is used to perform defect recognition on the single-phase image to obtain the defect recognition result.

[0006] In the above embodiment, the present invention combines unsupervised learning and change detection and applies it to defect detection in the hardware field. By comparing images taken at different times under the same shooting conditions, the differences in images are obtained to realize defect identification and detection.

[0007] As some optional implementations of the present application, the process of preprocessing a single-phase image is as follows: Denoising, cropping and scaling single-phase images taken at different times under the same shooting conditions, and feature labeling of partially denoised, cropped and scaled single-phase images; A sample dataset is constructed based on unlabeled single-phase images and labeled single-phase images.

[0008] In the above embodiment, the present invention facilitates subsequent training of the defect recognition model by preprocessing the single-phase image.

[0009] As some optional implementations of this application, the process of performing semi-supervised learning and change detection on the defect recognition model based on the sample dataset is as follows: The unlabeled single-phase images are input into the unsupervised model for unsupervised training. During the unsupervised training process, data augmentation, data perturbation, dual-phase exchange, spatial and channel feature exchange, and global feature enhancement extraction are performed on the unlabeled single-phase images; The labeled single-phase images are input into the supervised model for supervised training. During the supervised training process, the labeled single-phase images are subjected to random occlusion, data augmentation and bi-phase exchange.

[0010] In the above embodiment, the present invention combines data enhancement and data perturbation on a small number of labeled samples, and adds constraints on time changes, thereby generating a large number of dual-phase images from single-phase images, enriching the amount of sample data and reducing the cost of manual labeling.

[0011] As some optional implementations of the present application, the process of performing data augmentation, data perturbation, dual-phase exchange, spatial and channel feature exchange, and global feature enhancement extraction on unlabeled single-phase images is as follows: Perform data augmentation on unlabeled single-phase images to obtain enhanced images; Performing data perturbation and bi-temporal exchange on the enhanced image to obtain a distorted image; Perform feature extraction on the enhanced image and the distorted image to obtain a dual-phase feature image; Spatial feature exchange and channel feature exchange are performed on the dual-phase feature image, and global feature enhancement extraction is performed on the dual-phase feature image with spatial feature exchange and channel feature exchange to obtain resampled image features of the dual-phase feature image.

[0012] In the above embodiment, the present invention can more effectively obtain the contextual relationship and frequency domain relationship of the image by performing spatial feature exchange and channel feature exchange on the dual-phase feature image, and performing global feature enhancement extraction on the dual-phase feature image with spatial feature exchange and channel feature exchange.

[0013] As some optional implementations of the present application, the process of performing spatial feature exchange and channel feature exchange on a dual-temporal feature image is as follows: Feature extraction is performed on the bi-temporal feature image, and the image features are projected into a feature space. An attention mechanism is introduced to adaptively re-weight the spatial features related to the current position in the bi-temporal feature image using the distributed attention mechanism. Feature extraction is performed on the bi-phase feature image, the image features are projected into a feature channel, and an attention mechanism is introduced to adaptively reweight the features related to the current channel in the bi-phase feature image using the distributed attention mechanism.

[0014] In the above embodiment, the present invention faces a large number of pseudo dual-phase images, and the model needs to fully learn its global features as well as the respective channel features and spatial features. In order to enhance the global features, the dual-phase features are projected into a feature space, and the global common features are obtained, and then each channel of the dual-phase features is adaptively reweighted using distributed attention mapping. In order to fully obtain the channel features and spatial features, the extracted features are exchanged, so as to mutually learn and perceive the contextual information of the dual-phase features. And make the feature distributions of the two branches more similar, automatically achieving a certain degree of domain adaptation between the dual-phase fields.

[0015] As some optional implementations of the present application, the process of performing random occlusion, data enhancement, and dual-phase exchange on the labeled single-phase image is as follows: Randomly select single-phase images and corresponding labels to obtain pseudo-bi-phase images and corresponding labels; A bi-phase swap is performed on the pseudo bi-phase image to obtain a time-symmetric image, and change detection labels are generated on corresponding labels of the pseudo bi-phase image to obtain corresponding change detection labels.

[0016] In the above embodiment, the present invention applies change detection to the field of defect detection to determine the changes in defects under the same background. Change detection is a technology that identifies areas that have changed from one time point to another, thereby avoiding false detection of unseen characteristic defects.

[0017] In a second aspect, the present invention provides a defect recognition system based on semi-supervised learning and change detection, the system comprising: An image data acquisition unit, configured to acquire single-phase images taken at different times under the same shooting conditions, and preprocess the single-phase images to construct a sample data set; A defect recognition model unit, which is used to construct a defect recognition model and perform semi-supervised learning and change detection on the defect recognition model based on a sample data set to obtain a final defect recognition model; wherein the defect recognition model includes an unsupervised model and a supervised model; The defect recognition and detection unit is used to use the final defect recognition model to perform defect recognition on the single-phase image to obtain a defect recognition result.

[0018] In a third aspect, the present invention provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the defect identification method based on semi-supervised learning and change detection when executing the computer program.

[0019] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the defect identification method based on semi-supervised learning and change detection is implemented.

[0020] The beneficial effects of the present invention are as follows: the present invention uses a small number of single-phase images to enhance them into dual-phase images, solving the problems of small sample size and complex dual-phase data labeling; at the same time, unsupervised learning and change detection are combined and applied to defect detection in the hardware field. By comparing images at different times under the same shooting conditions, the differences in images are obtained, and defect identification and detection are realized. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For those skilled in the art, other relevant drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 is a schematic diagram of the computer device structure of the hardware operating environment described in an embodiment of the present invention; Figure 2 is a flowchart of a defect identification method based on semi-supervised learning and change detection according to an embodiment of the present invention; Figure 3 is a workflow diagram of the unsupervised model and the supervised model described in the embodiment of the present invention; Figure 4 This is a flowchart of the supervised model training described in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0024] In order to solve the problem of lack of sample data in existing industrial hardware defect detection and the lack of new defect matching ability of the model due to the inability to migrate the supervised learning part, this application provides a defect recognition method, system, device and storage medium based on semi-supervised learning and change detection. Before introducing the specific technical solution of this application, the hardware operating environment involved in the embodiment of this application is first introduced.

[0025] See also Figure 1 , Figure 1 This is a schematic diagram of the computer device structure of the hardware operating environment involved in the embodiment of the present application.

[0026] like Figure 1 As shown, 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 used to enable communication between these components. The user interface may include a display 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 optionally include a standard 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 disk storage. The memory may also be a storage device independent of the processor.

[0027] Those skilled in the art will understand that Figure 1 The structure shown in the figure does not constitute a limitation on the computer device, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0028] like Figure 1 As shown, the memory as a storage medium may include an operating system, a network communication module, a user interface module, and an electronic program module.

[0029] exist Figure 1In the computer device shown, the network interface is mainly used for data communication with the network server; the user interface is mainly used for data interaction with the user; the processor and memory in the computer device of the present application can be set in the computer device, and the computer device calls the computer program product stored in the electronic program module through the processor, and executes the defect identification method based on semi-supervised learning and change detection provided in the embodiment of the present application.

[0030] Based on the hardware environment of the aforementioned embodiment, the embodiment of this application provides a defect recognition method based on semi-supervised learning and change detection, please refer to Figure 2 , Figure 2 This is a flow chart of the defect recognition method based on semi-supervised learning and change detection. The method flow is as follows: (1) Obtain single-phase images at different times under the same shooting conditions and preprocess the single-phase images to construct a sample dataset.

[0031] In an embodiment of the present invention, the process of preprocessing a single-phase image is as follows: (1.1) Denoise, crop, and scale single-phase images taken at different times under the same shooting conditions, and perform feature annotation on partially denoised, cropped, and scaled single-phase images.

[0032] (1.2) Construct a sample dataset based on unlabeled single-phase images and labeled single-phase images.

[0033] (2) Constructing a defect recognition model, and performing semi-supervised learning and change detection on the defect recognition model based on a sample data set to obtain a final defect recognition model; wherein the defect recognition model includes an unsupervised model and a supervised model.

[0034] In an embodiment of the present invention, the process of performing semi-supervised learning and change detection on a defect recognition model based on a sample dataset is as follows: (2.1) The unlabeled single-phase images are input into the unsupervised model for unsupervised training. During the unsupervised training process, the unlabeled single-phase images are subjected to data augmentation, data perturbation, dual-phase exchange, spatial and channel feature exchange, and global feature enhancement extraction.

[0035] In the embodiment of the present invention, please refer to Figure 3 , Figure 3 The workflow diagram of the unsupervised model and the supervised model is as follows: (2.11) For N unlabeled single-phase images Perform data augmentation to obtain enhanced images .

[0036] In the embodiment of the present invention, the unlabeled single-phase image Data enhancement can increase data diversity and expand the amount of sample data. Specifically, the data enhancement method used in the embodiment of the present invention is a weak data enhancement method, including random flipping and translation.

[0037] (2.12) Enhanced image Perform data perturbation and dual-phase exchange to obtain distorted images .

[0038] In the embodiment of the present invention, the enhanced image Data perturbation can enhance the type and semantic expression of data and expand the amount of sample data. Specifically, the embodiment of the present invention adopts data perturbation as a strong data enhancement method, including contrast change, histogram equalization, rotation, sharpening and brightness change; preferably, the present invention randomly combines 2-3 enhancement methods to form an enhanced distorted image. .

[0039] In the embodiment of the present invention, the main purpose of the change detection is to detect "changes", thereby enhancing the image The temporal order of the input images is not important, and the order of the input images to the unsupervised training model does not affect the final change detection results. Therefore, a dual-phase exchange method is used. During the unsupervised training process, the two enhanced images are exchanged. , to increase the time symmetry constraint and enrich the types of sample data.

[0040] (2.13) Enhanced image and distorted image Feature extraction is performed to obtain a dual-phase feature image.

[0041] (2.14) Input the bi-phase feature image into the aggregation distribution model, perform spatial feature exchange and channel feature exchange on the bi-phase feature image through the aggregation distribution model, and perform global feature enhancement extraction on the bi-phase feature image with spatial feature exchange and channel feature exchange to obtain the resampled image features of the bi-phase feature image ; Specifically, we first perform multi-stage feature extraction on the bi-temporal feature images to obtain the corresponding features. We then perform one spatial feature exchange and two channel feature exchanges on the features. The basic idea behind these two methods is to project the bi-temporal features into a feature space and obtain global common features. Distributed attention mapping is then used to adaptively reweight each channel of the bi-temporal features.

[0042] Specifically, the process of exchanging spatial features and channel features for the dual-temporal feature image is as follows: (2.141) Feature extraction is performed on the bi-phase feature image, the image features are projected into a feature space, and an attention mechanism is introduced to adaptively re-weight the spatial features related to the current position in the bi-phase feature image using the distributed attention mechanism.

[0043] (2.142) Feature extraction is performed on the bi-phase feature image, the image features are projected into a feature channel, and an attention mechanism is introduced to adaptively reweight the features related to the current channel in the bi-phase feature image using the distributed attention mechanism.

[0044] In addition, global feature enhancement extraction is performed on the bi-temporal feature images with spatial feature exchange and channel feature exchange, including using two multi-layer perceptron decoders (MLP Decoders) to extract common features and using activation functions to obtain the final two attention maps. Then, the misalignment problem in feature extraction is solved through the two-stream alignment fusion network (FDAF Module), and the bi-temporal feature maps are resampled through the deformable field to obtain their respective corrected features. .

[0045] Specifically, the process of global feature enhancement extraction for the bi-temporal feature image with spatial feature exchange and channel feature exchange is as follows: (2.143) Performing element-wise addition on the features of the dual-phase feature image to obtain an element-wise added image, and performing global average pooling on the element-wise added image to obtain a global pooled feature image.

[0046] (2.144) Activate the global pooled feature image and perform dual-stream alignment fusion processing to obtain the resampled image features of the dual-phase feature image .

[0047] In the embodiment of the present invention, considering that there is no label in unsupervised learning, the resampled image features after the global feature enhancement model extraction model are Perform data perturbation again. At the same time, in order to avoid interference from some confidence thresholds, set confidence threshold filtering to obtain filtered resampled image features . And by calculating the loss function , thereby continuously reducing the confidence threshold filtered resampled image features With the original feature map The distance difference between the original feature map Unlabeled single-phase image The original feature map of .

[0048] (2.2) The labeled single-phase images are input into the supervised model for supervised training. During the supervised training process, the labeled single-phase images are randomly masked, data augmented, and subjected to dual-phase swapping.

[0049] In the embodiment of the present invention, please refer to Figure 3 , Figure 3 The workflow diagram of the unsupervised model and the supervised model is as follows: (2.21) For N single-phase images And the corresponding tags Perform two random selections to obtain a pseudo-bi-phase image and tags and pseudo-bi-phase images and , see Figure 4 , Figure 4 Flowchart of the supervised training.

[0050] (2.22) For pseudo-bi-phase images and tags and pseudo-bi-phase images and Perform data enhancement to generate more pseudo dual-phase images and And the corresponding tags and ; Specifically, the data enhancement method adopted in the embodiment of the present invention is a weak data enhancement method, including random flipping and translation.

[0051] (2.23) For pseudo-bi-phase images and Perform dual-phase exchange to obtain a time-symmetric image and , and for labels and Generate change detection labels to obtain corresponding change detection labels .

[0052] At the same time, the pseudo-bi-phase image and and time-symmetric images and Perform feature extraction to obtain the corresponding feature image; and input the feature image into the aggregation distribution model, and the aggregation distribution model is used to extract the pseudo dual-phase image. and and time-symmetric images and The spatial feature exchange and channel feature exchange are performed on the feature image of the spatial feature exchange and channel feature exchange, and the global feature enhancement extraction is performed on the feature image of the spatial feature exchange and channel feature exchange to obtain the resampled image feature of the pseudo dual-phase feature and , and by calculating the loss function and , thereby continuously reducing the resampled image features and with change detection tags gap.

[0053] Specifically, in the supervised data enhancement process, strong enhancement and weak enhancement are used to expand the sample data. And by increasing the exchange of data combinations at different times, the model's sensitivity to time changes is reduced. In the supervised learning process, first input N single-phase images And the corresponding tags By performing two random selections and pairing the two resulting image sets, we obtain a set of pseudo-bitemporal images. Secondly, to further increase the number of pseudo-bitemporal image sets, we employ data augmentation (strong and weak augmentation, etc.), while also using an XOR operation to generate pseudo change detection labels. Finally, we incorporate a temporal similarity constraint, further doubling the number of image pairs at the model input level.

[0054] (3) The final defect recognition model is used to perform defect recognition on the single-phase image to obtain the defect recognition result.

[0055] In the embodiment of the present invention, after supervised training and unsupervised training, the defect recognition model has the ability to recognize defects and can achieve fast and accurate defect recognition and detection.

[0056] In summary, the present invention transforms the problem of identifying defect types into the question of whether the detection area has "changed." By collecting some samples of hardware defects and using change detection to detect defects, the present invention reduces manpower and improves the accuracy of defect identification.

[0057] In addition, in one embodiment, based on the same inventive concept as the aforementioned embodiment, an embodiment of the present invention provides a defect recognition system based on semi-supervised learning and change detection. The system corresponds one-to-one with the method, and the system includes: An image data acquisition unit, configured to acquire single-phase images taken at different times under the same shooting conditions, and preprocess the single-phase images to construct a sample data set; A defect recognition model unit, which is used to construct a defect recognition model and perform semi-supervised learning and change detection on the defect recognition model based on a sample data set to obtain a final defect recognition model; wherein the defect recognition model includes an unsupervised model and a supervised model; The defect recognition and detection unit is used to use the final defect recognition model to perform defect recognition on the single-phase image to obtain a defect recognition result.

[0058] It should be noted that each unit in the defect recognition system based on semi-supervised learning and change detection in this embodiment corresponds one-to-one to each step in the defect recognition method based on semi-supervised learning and change detection in the aforementioned embodiment. Therefore, the specific implementation method and technical effects achieved in this embodiment can refer to the implementation method of the aforementioned defect recognition method based on semi-supervised learning and change detection, and will not be repeated here.

[0059] In addition, in one embodiment, the present application also provides a computer device, which includes a processor, a memory, and a computer program stored in the memory, and the computer program implements the method in the aforementioned embodiment when executed by the processor.

[0060] In addition, in one embodiment, the present application further provides a computer storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method in the aforementioned embodiment is implemented.

[0061] In some embodiments, the computer-readable storage medium may be a memory device such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface mount memory, optical disk, or CD-ROM; or various devices including any one or any combination of the above memories. The computer may be various computing devices including smart terminals and servers.

[0062] In some embodiments, executable instructions may be in 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 stand-alone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0063] As an example, executable instructions may, but need not, correspond to a file in a file system, may be stored as part of a file that stores other programs or data, such as in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple coordinating files (e.g., files storing one or more modules, subroutines, or code portions).

[0064] By way of example, executable instructions may be deployed to be executed on one computing device, or on multiple computing devices at one site, or on multiple computing devices distributed across multiple sites and interconnected by a communication network.

[0065] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or system comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or system. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or system comprising the element.

[0066] The serial numbers of the above embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0067] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform. Of course, they can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium (such as a read-only memory / random access memory, a magnetic disk, or an optical disk), and includes a number of instructions for enabling a multimedia terminal device (which can be a mobile phone, a computer, a television receiver, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0068] The above are only preferred embodiments of the present application and do not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A defect recognition method based on semi-supervised learning and change detection, characterized in that: The method includes the following steps: Acquire single-phase images at different times under the same shooting conditions, and preprocess the single-phase images to construct a sample data set; Constructing a defect recognition model, and performing semi-supervised learning and change detection on the defect recognition model based on a sample data set to obtain a final defect recognition model; wherein the defect recognition model includes an unsupervised model and a supervised model; The final defect recognition model is used to perform defect recognition on the single-phase image to obtain the defect recognition result.

2. The defect recognition method based on semi-supervised learning and change detection according to claim 1, characterized in that: The process of preprocessing single-phase images is as follows: Denoising, cropping and scaling single-phase images taken at different times under the same shooting conditions, and feature labeling of partially denoised, cropped and scaled single-phase images; A sample dataset is constructed based on unlabeled single-phase images and labeled single-phase images.

3. The defect recognition method based on semi-supervised learning and change detection according to claim 1, characterized in that: The process of semi-supervised learning and change detection of the defect recognition model based on the sample dataset is as follows: The unlabeled single-phase images are input into the unsupervised model for unsupervised training. During the unsupervised training process, data augmentation, data perturbation, dual-phase exchange, spatial and channel feature exchange, and global feature enhancement extraction are performed on the unlabeled single-phase images; The labeled single-phase images are input into the supervised model for supervised training. During the supervised training process, the labeled single-phase images are subjected to random occlusion, data augmentation and bi-phase exchange.

4. The defect recognition method based on semi-supervised learning and change detection according to claim 3 is characterized in that: The process of data augmentation, data perturbation, dual-temporal exchange, spatial and channel feature exchange, and global feature enhancement extraction for unlabeled single-phase images is as follows: Perform data augmentation on unlabeled single-phase images to obtain enhanced images; Performing data perturbation and bi-temporal exchange on the enhanced image to obtain a distorted image; Perform feature extraction on the enhanced image and the distorted image to obtain a dual-phase feature image; Spatial feature exchange and channel feature exchange are performed on the dual-phase feature image, and global feature enhancement extraction is performed on the dual-phase feature image with spatial feature exchange and channel feature exchange to obtain resampled image features of the dual-phase feature image.

5. The defect recognition method based on semi-supervised learning and change detection according to claim 3 is characterized in that: The process of random occlusion, data augmentation, and bi-temporal exchange for the annotated single-phase image is as follows: Randomly select single-phase images and corresponding labels to obtain pseudo-bi-phase images and corresponding labels; A bi-phase swap is performed on the pseudo bi-phase image to obtain a time-symmetric image, and change detection labels are generated on corresponding labels of the pseudo bi-phase image to obtain corresponding change detection labels.

6. The defect recognition method based on semi-supervised learning and change detection according to claim 4, characterized in that: The process of exchanging spatial features and channel features for dual-phase feature images is as follows: Feature extraction is performed on the bi-temporal feature image, and the image features are projected into a feature space. An attention mechanism is introduced to adaptively re-weight the spatial features related to the current position in the bi-temporal feature image using the distributed attention mechanism. Feature extraction is performed on the bi-phase feature image, the image features are projected into a feature channel, and an attention mechanism is introduced to adaptively reweight the features related to the current channel in the bi-phase feature image using the distributed attention mechanism.

7. The defect recognition method based on semi-supervised learning and change detection according to claim 4, characterized in that: The process of global feature enhancement extraction for the bi-temporal feature image with spatial feature exchange and channel feature exchange is as follows: performing element-wise addition on the bi-temporal feature image to obtain an element-wise added image, and performing global average pooling on the element-wise added image to obtain a global pooled feature image; The global pooled feature image is activated and fused with two streams to obtain the resampled image features of the two-phase feature image.

8. A defect recognition system based on semi-supervised learning and change detection, characterized in that: The system comprises: An image data acquisition unit, configured to acquire single-phase images taken at different times under the same shooting conditions, and preprocess the single-phase images to construct a sample data set; A defect recognition model unit, which is used to construct a defect recognition model and perform semi-supervised learning and change detection on the defect recognition model based on a sample data set to obtain a final defect recognition model; wherein the defect recognition model includes an unsupervised model and a supervised model; The defect recognition and detection unit is used to use the final defect recognition model to perform defect recognition on the single-phase image to obtain a defect recognition result.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the defect recognition method based on semi-supervised learning and change detection described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the defect recognition method based on semi-supervised learning and change detection described in any one of claims 1 to 7 is implemented.

Images