A method for detecting surface defects in products
By combining anomaly detection networks and target detection networks, surface defects of lithium battery electrodes can be detected quickly and accurately, solving the problems of generalization and cost of traditional methods, achieving efficient defect detection, and making it suitable for lithium battery electrode production.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- TSINGHUA SHENZHEN INTERNATIONAL GRADUATE SCHOOL
- Filing Date
- 2023-03-27
- Publication Date
- 2026-05-26
AI Technical Summary
Existing methods for detecting defects in lithium battery electrodes lack generalization, making it difficult to identify new defect types. Furthermore, traditional methods are sensitive to changes in the production environment, are costly, and require manual definition of defect characteristics.
An anomaly detection network and a target detection network based on residual blocks are used to detect large-size through-hole defects and small-size particle defects, respectively. Images are acquired by a linear scanning camera, and features are extracted and distance scores are calculated using a deep residual network. False detections are filtered out by combining nonmaximum suppression methods.
It enables rapid and accurate detection of surface defects in lithium battery electrodes, reduces the false negative rate, covers a variety of defect types, does not rely on manual definition, is applicable to existing production equipment, and requires no modification.
Smart Images
Figure CN116385396B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of computer vision technology, and in particular to a method for detecting surface defects in products. Background Technology
[0002] Lithium-ion batteries have advantages such as high energy density, long lifespan, and low pollution, and are therefore widely used in new energy vehicles, power supplies, and energy storage. The lithium-ion battery electrode is a crucial component of the battery; defects in the electrode can severely affect the quality of the lithium-ion battery and even pose safety hazards.
[0003] As a key material in lithium battery manufacturing, electrodes are susceptible to various quality defects during actual production, including indentations, metal leakage, and air bubbles, due to factors such as raw materials, production equipment, and manual operation. Therefore, electrode defect detection is a crucial step in ensuring the performance and safety of lithium batteries.
[0004] Traditional machine vision-based surface defect detection methods often employ conventional image processing algorithms or manually designed features combined with classifiers. They typically utilize the different properties of the surface or defect being inspected to design corresponding imaging schemes, thereby obtaining uniformly illuminated images that clearly reveal surface defects. This carefully constructed imaging scheme can significantly reduce the difficulty of designing traditional detection algorithms, but it also increases the application cost of the detection system. Furthermore, in many open industrial environments, expecting the designed imaging system to completely eliminate the influence of changes in the scene or the inspected material on the detection system is often unrealistic. In recent years, with the successful application of deep learning models, represented by convolutional neural networks, in many computer vision fields, numerous deep learning-based defect detection methods have been widely applied in various industrial scenarios. Simultaneously, in the process of applying deep learning to the industrial field, the requirements for defect detection have gradually become more comprehensive.
[0005] Currently, most of the algorithms used for inspection during the production of lithium battery electrode sheets are traditional machine vision solutions. These solutions require manual definition of defect features and the use of traditional image algorithms to extract defect feature edges and perform classification. These solutions lack generalization ability and are not easy to identify new defect types.
[0006] CN 112858334 A discloses a lithium battery electrode detection method. Its algorithm uses camera positioning measurement to detect defects, which has strict requirements on camera parameters in the production environment and can only identify a limited number of defect types.
[0007] It should be noted that the information disclosed in the background section above is only for understanding the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention
[0008] The main objective of this invention is to overcome the deficiencies of the above-mentioned background technology and provide a method for detecting surface defects in products.
[0009] To achieve the above objectives, the present invention adopts the following technical solution:
[0010] A method for detecting surface defects in a product includes the following steps:
[0011] S1. Acquire an image of the product surface; preferably, on the product image acquisition platform, a linear scanning camera is used to acquire a grayscale image of the product transmitted along the horizontal direction.
[0012] S2. Use a residual block-based anomaly detection network to extract and save the feature matrix of manually selected qualified images, and at the same time calculate the threshold of feature distance score using qualified images.
[0013] S3. For the collected images of the products to be inspected, features are extracted from the anomaly detection network and the distance score with the feature matrix is calculated. Products with a distance score exceeding the threshold are marked as unqualified.
[0014] S4. For products that pass the inspection in step S3, use a pre-trained target detection network to detect small-sized particle defects. Products with detected defects are marked as unqualified.
[0015] Furthermore:
[0016] The product in question is a lithium battery electrode sheet.
[0017] In step S1, linear scanning cameras are used above and below the product being transported in the horizontal direction to capture images of the front and back of the product.
[0018] In step S2, an unsupervised anomaly detection network feature extractor is established using a deep residual network. The qualified image is downsampled four times and fed into the anomaly detection network for feature extraction. The extracted features are the features of the second and third layers of the deep residual network. The features of the second and third layers are sparsified and then saved.
[0019] The feature sparsity processing includes principal component analysis feature dimensionality reduction, wherein the features of the second and third layers of the network are subjected to principal component analysis, and 10% of the principal component features are selected as core features to reduce the dimensionality of the feature matrix.
[0020] The threshold is set by selecting 30 or more qualified images from the training process and feeding them into the network for inference to calculate the maximum distance score, and using this distance score as the threshold.
[0021] In step S3, the image of the product to be inspected is downsampled four times and sent to the anomaly detection network to extract image features, and the distance score between it and the qualified image features of the feature matrix saved in step S2 is calculated. Images that exceed the set threshold are marked as unqualified.
[0022] In step S3, the distance score between the features of the image to be detected and the features of the qualified image is calculated using the nearest neighbor retrieval method.
[0023] The anomaly detection network detects large-sized through-hole defects.
[0024] In step S4, when using the target detection network to detect small-sized particulate defects, a non-maximum suppression method is used to filter out false detection defects. For defects with a confidence level greater than 0.5, the corresponding electrode is detected and determined to be unqualified.
[0025] The present invention has the following beneficial effects:
[0026] This invention provides a method for detecting surface defects in products, particularly suitable for the rapid detection of surface defects in lithium battery electrodes. It employs an anomaly detection network and a target detection network in a stepwise manner to detect large-size through-hole defects and small-size particle defects in the acquired lithium battery electrode images. Electrodes with detected defects are marked as unqualified and a rejection signal is output. This invention achieves rapid and accurate detection of large-size defects through anomaly detection and small-size defects through target detection, enabling rapid and accurate detection of unqualified lithium battery electrodes with various types of surface defects. This method can detect defects in lithium battery electrodes of different sizes, effectively reducing the false negative rate. This method can cover multiple defect types without relying on manual definition. It does not require modification of existing production equipment and has high portability. This invention has broad application prospects and significant market value for improving the pass rate and battery safety in the lithium battery production process. Attached Figure Description
[0027] Figure 1 This is a flowchart of a method for detecting surface defects of lithium battery electrodes provided in an embodiment of the present invention. Detailed Implementation
[0028] The embodiments of the present invention will be described in detail below. It should be emphasized that the following description is merely exemplary and not intended to limit the scope and application of the present invention.
[0029] See Figure 1 This invention provides a method for detecting surface defects in a product, comprising the following steps:
[0030] S1. Acquire an image of the product surface;
[0031] Preferably, on the product image acquisition platform, a linear scanning camera is used to acquire grayscale images of the product transmitted along the horizontal direction;
[0032] S2. Use a residual block-based anomaly detection network to extract and save the feature matrix of manually selected qualified images, and at the same time calculate the threshold of feature distance score using qualified images.
[0033] S3. For the collected images of the products to be inspected, features are extracted from the anomaly detection network and the distance score with the feature matrix is calculated. Products with a distance score exceeding the threshold are marked as unqualified.
[0034] S4. For products that pass the inspection in step S3, use a pre-trained target detection network to detect small-sized particle defects. Products with detected defects are marked as unqualified.
[0035] This invention is particularly suitable for rapidly detecting lithium battery electrodes with surface defects that are substandard. However, the surface defect detection method of this invention is not limited to lithium battery electrodes.
[0036] The following further describes specific embodiments of the present invention for detecting lithium battery electrodes.
[0037] In some embodiments, the lithium battery electrode surface defect detection method of the present invention includes the following steps: On a lithium battery electrode image acquisition platform, a linear scanning camera is used to acquire grayscale images of lithium battery electrodes transmitted along the horizontal direction; a residual block-based anomaly detection network is used to extract and save the matrix of qualified images, and a feature distance score threshold is calculated using the qualified images; for other acquired images, features are extracted in the aforementioned network and the distance score between the extracted images and the saved feature matrix is calculated; electrodes with a distance score exceeding the threshold are marked as unqualified; a pre-trained target detection network is used to detect small-size particle defects in the aforementioned qualified electrodes, and electrodes with detected defects are marked as unqualified; the information of the unqualified electrodes in the aforementioned two networks is fed back to the system for detecting defective electrodes.
[0038] In a further embodiment, a method for detecting surface defects of lithium battery electrodes includes the following steps:
[0039] S1, lithium battery electrode image acquisition platform, uses a linear scanning camera to acquire grayscale images of lithium battery electrodes transported along the horizontal direction;
[0040] S2, establish an unsupervised anomaly detection network to detect large-size through-type defects. Manually select 50 qualified images and downsample the qualified images 4 times and send them into the network for feature extraction. The features of the second and third layers of the network feature extractor are sparsified and then saved.
[0041] S3: Acquire images of the lithium battery electrode surface, downsample the images four times and send them to the anomaly detection network described in S2 to extract image features, calculate the distance score between the image features and the qualified image features saved in S2, and mark images that exceed the set threshold as unqualified;
[0042] S4: Establish a target detection network to detect small-sized granular defects. Cut the original image that is qualified in S3 into blocks and send them into the detection network. Mark the image with detected defects as unqualified.
[0043] In some embodiments, in step S1, the lithium battery electrode acquisition platform uses a linear scanning camera to acquire images of the lithium battery electrodes being transported horizontally. The platform includes an upper linear scanning camera and a lower linear scanning camera. The upper linear scanning camera acquires images of the front side of the lithium battery electrodes on the transport platform, and the lower linear scanning camera acquires images of the back side of the lithium battery electrodes on the transport platform.
[0044] In some embodiments, in step S2, for anomaly detection networks used for large-size through-hole defects, an unsupervised anomaly detection network feature extractor is established using a deep residual network to extract features from 50 qualified images. The extracted features are the second and third layer features of the deep residual network.
[0045] In some embodiments, in step S3, the features extracted by the aforementioned network are sparsified. The feature sparsification method is principal component analysis feature dimensionality reduction. Principal component analysis is performed on the extracted second and third layer network features, and 10% of the principal component features are selected as core features to reduce the dimensionality of the feature matrix.
[0046] In some embodiments, in step S3, the method for calculating the distance score between the image feature to be detected and the qualified image feature is the nearest neighbor retrieval method.
[0047] In some embodiments, in step S3, the threshold setting criterion is to remove the images used in the training process, select 30 qualified images again and send them to the network inference calculation to obtain the maximum distance score, and use this distance score as the threshold.
[0048] In some embodiments, in step S4, a target detection network is used to detect small-sized particulate defects, a non-maximum suppression method is used to filter false detection defects, and defects with a confidence level greater than 0.5 are detected and the corresponding electrode is determined to be unqualified.
[0049] like Figure 1 As shown, the specific example of the lithium battery electrode surface defect detection method includes the following steps:
[0050] This lithium-ion battery electrode image acquisition platform uses linear scanning cameras to acquire grayscale images of lithium-ion battery electrodes transported horizontally. The platform comprises two linear scanning cameras, located above and below the platform respectively. The upper linear scanning camera primarily acquires images of the front side of the lithium-ion battery electrodes on the transport platform, while the lower linear scanning camera acquires images of the back side of the lithium-ion battery electrodes. During image acquisition, the cameras scan the surface of the lithium-ion battery electrodes line by line with a width of 4096 pixels, converting the images into digital signals, and finally outputting a grayscale image.
[0051] Eighty qualified images of defect-free pole pieces were manually selected, and 50 of them were used as training images to be input into an anomaly detection network for feature extraction. The anomaly detection network used residual blocks as a feature extraction module. The selected 50 qualified images were downsampled four times to a size of 128×256, and then convolutional operations were performed on the residual block structure to extract the feature vectors of the second and third layers of the residual block.
[0052] In the two layers of feature vectors extracted above, the features extracted in the third layer are matched with the dimension of the feature vector in the second layer using bilinear interpolation. The two layers of feature vectors are then aggregated using adaptive average pooling to obtain the feature matrix of the qualified image. Principal component analysis is used to reduce the dimension of the feature matrix to 10% of the original dimension. The feature matrix after dimension reduction is denoted as M and saved.
[0053] Of the 80 qualified poleme images selected above, the remaining 30 qualified images that were not used during the training phase were selected. Feature extraction was performed on these 30 images using the residual feature extractor of the aforementioned anomaly detection network. The distance score between each feature block and the previously saved training feature matrix M was calculated using the nearest neighbor search method. The maximum distance score was denoted as s. * s * This is the distance score threshold for anomaly detection.
[0054] The lithium battery electrode images collected by the platform are downsampled using the aforementioned steps, then fed into an anomaly detection network to extract the feature matrix and calculate the distance score s. For s > s * The electrode was marked as unqualified.
[0055] For the electrodes marked as qualified in the aforementioned anomaly detection network, a pre-trained target detection network is used to detect small-sized particulate defects. Each detection result includes the position and confidence level of multiple particulate defect candidate boxes.
[0056] The detection confidence level for particle defects is set to 0.5. Non-maximum suppression is applied to multiple candidate boxes with confidence levels higher than the confidence threshold in each detection result to obtain the final detection result for each defect.
[0057] Electrodes with detected particle defects are marked as unqualified.
[0058] Electrode images are processed by an anomaly detection network and a target detection network to detect large-size through-hole defects and small-size particle defects, respectively. Electrodes marked as unqualified are output to the control center for rejection, while electrode images not marked as unqualified enter the subsequent workstations.
[0059] The lithium battery electrode surface defect detection method proposed in this invention includes an image acquisition method, an anomaly detection network, and a target detection network. It detects large-size through-hole defects and small-size particle defects in the acquired lithium battery electrode images, marks the electrode with detected defects as unqualified, and outputs a rejection signal. This achieves the effect of rapid and accurate detection of unqualified lithium battery electrodes with various types of surface defects.
[0060] This invention can be used by battery manufacturers for surface defect detection of electrodes during lithium battery production, enabling automated defect product detection. This method covers multiple defect types without relying on manual definitions. It requires no modification to existing production equipment, exhibiting high portability. The detection method of this invention can be introduced without altering battery manufacturing lines, providing faster and more effective defect detection.
[0061] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0062] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0063] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0064] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0065] The background section of this invention may include background information about the problems or environment in which the invention is being developed, and is not necessarily a description of prior art. Therefore, the content included in the background section does not constitute an admission of prior art by the applicant.
[0066] The above description provides a further detailed explanation of the present invention in conjunction with specific / preferred embodiments, and it should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various substitutions or modifications can be made to these described embodiments without departing from the concept of the present invention, and all such substitutions or modifications should be considered within the scope of protection of the present invention. In the description of this specification, the reference to terms such as "an embodiment," "some embodiments," "preferred embodiment," "example," "specific example," or "some examples," etc., indicates that the specific features, structures, materials, or characteristics described in connection with that embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any suitable manner in one or more embodiments or examples. Without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification and the features of different embodiments or examples. Although the embodiments of the present invention and their advantages have been described in detail, it should be understood that various changes, substitutions, and modifications can be made herein without departing from the scope of protection of the patent application.
Claims
1. A method for detecting surface defects in lithium battery electrode products, characterized in that, Includes the following steps: S1. Acquire images of the product surface; wherein, on the product image acquisition platform, a linear scanning camera is used to acquire grayscale images of the lithium battery electrode products transported along the horizontal direction. S2. Use a residual block-based anomaly detection network to extract and save the feature matrix of manually selected qualified images, and calculate the threshold of feature distance score using qualified images. The anomaly detection network detects large-size through-type defects. In step S2, an unsupervised anomaly detection network feature extractor is established using a deep residual network. Qualified images are downsampled four times and fed into the anomaly detection network for feature extraction. The extracted features are the features of the second and third layers of the deep residual network. The features of the second and third layers are sparsified and then saved. S3. For the collected images of the products to be detected, features are extracted in the anomaly detection network and the distance score with the feature matrix is calculated. Products with a distance score exceeding the threshold are marked as unqualified. The threshold is set by selecting 30 or more qualified images again after removing the images used in the training process and feeding them into the network for inference to calculate the maximum distance score, and using this distance score as the threshold. S4. For products that pass the inspection in step S3, use a pre-trained target detection network to detect small-sized particle defects. Products with detected defects are marked as unqualified.
2. The method for detecting surface defects in lithium battery electrode products as described in claim 1, characterized in that, The feature sparsity processing includes principal component analysis feature dimensionality reduction, wherein the features of the second and third layers of the network are subjected to principal component analysis, and 10% of the principal component features are selected as core features to reduce the dimensionality of the feature matrix.
3. The method for detecting surface defects in lithium battery electrode products as described in any one of claims 1 to 2, characterized in that, In step S3, the image of the product to be inspected is downsampled four times and sent to the anomaly detection network to extract image features, and the distance score between it and the qualified image features of the feature matrix saved in step S2 is calculated. Images that exceed the set threshold are marked as unqualified.
4. The method for detecting surface defects in lithium battery electrode products as described in any one of claims 1 to 2, characterized in that, In step S3, the distance score between the features of the image to be detected and the features of the qualified image is calculated using the nearest neighbor retrieval method.
5. The method for detecting surface defects in lithium battery electrode products as described in any one of claims 1 to 2, characterized in that, In step S1, linear scanning cameras are used above and below the product being transported in the horizontal direction to capture images of the front and back of the product.
6. The method for detecting surface defects in lithium battery electrode products as described in any one of claims 1 to 2, characterized in that, In step S4, when using the target detection network to detect small-sized particulate defects, a non-maximum suppression method is used to filter out false detection defects. For defects with a confidence level greater than 0.5, the corresponding electrode is detected and determined to be unqualified.