Battery pole piece edge burr detection method, device and equipment and storage medium
By integrating grayscale texture and physical height information into a target detection network, the automation problem of burr detection on the edge of battery electrodes is solved, achieving efficient and accurate burr detection, which is suitable for lithium battery production lines.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENZHEN BAK POWER BATTERY CO LTD
- Filing Date
- 2025-12-23
- Publication Date
- 2026-04-21
AI Technical Summary
Existing methods for detecting burrs on the edges of battery electrodes are cumbersome, time-consuming, prone to missed detections, and cannot be fully automated, resulting in inaccurate and unreliable test results.
An object detection network is used to fuse grayscale texture images and physical height and shape images. The object detection network is optimized through a training dataset to achieve online automated detection of burrs on the edge of battery electrodes. Features are extracted using a dual-stream backbone network and an adaptive feature fusion network, and detection accuracy is optimized by combining a high consistency loss value.
It enables automated online detection of burrs on the edges of battery electrodes, improving detection efficiency and accuracy, avoiding misjudgments caused by factors such as light and dirt, and is suitable for large-scale continuous production.
Smart Images
Figure CN121904084A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of target detection technology, and in particular to a method, apparatus, device, and storage medium for detecting burrs on the edges of battery electrodes. Background Technology
[0002] Currently, most companies use visual microscopes for offline inspection of electrode burrs. However, this method has many drawbacks. First, the operation is extremely cumbersome, requiring staff to manually and meticulously adjust microscope parameters for each inspection, including focusing and magnification, consuming a significant amount of time and effort. Second, even with a large depth-of-field microscope, it is difficult to overcome the problems of insufficient depth of field and magnification, leading to inaccurate and incomplete inspection results. Third, prolonged manual inspection through a microscope can easily result in missed inspections due to visual fatigue, compromising the stability and reliability of the inspection. Furthermore, traditional inspection methods cannot achieve fully automated inspection, failing to provide timely and effective support for optimizing production processes. Summary of the Invention
[0003] In view of this, the purpose of the present invention is to overcome the shortcomings of the prior art and provide a method, apparatus, device and storage medium for detecting edge burrs of battery electrode sheets, for realizing online automated detection of edge burrs of battery electrode sheets.
[0004] This invention provides the following technical solution: In a first aspect, the present invention proposes a method for detecting burrs on the edge of battery electrodes, comprising: Acquire battery electrode edge images during the battery production line process, the battery electrode edge images including grayscale texture images and physical height shape images; Edge type annotation is performed on the grayscale texture image to obtain a training dataset; wherein, the edge type includes burr edges and normal edges; The training dataset is input into the object detection network for training to obtain the trained object detection network. The training process includes: generating predicted values using the object detection network based on the training dataset; calculating classification loss, bounding box regression loss, and high consistency loss based on the predicted values and their corresponding ground truth labels; calculating a total loss based on the classification loss, bounding box regression loss, and high consistency loss; and optimizing the training direction of the object detection network based on the total loss. The trained object detection network is used to detect burrs on the new battery electrode edge images.
[0005] In one embodiment, calculating the high consistency loss value based on the predicted value and its corresponding true labeled value includes: Based on the predicted value, a positive sample is determined, and based on the anchor frame in the grayscale texture image corresponding to the positive sample, the anchor frame region and non-anchor frame region in the physical height shape image corresponding to the positive sample are determined; Calculate the difference between the average pixel height of the anchor frame region and the average pixel height of the non-anchor frame region to obtain the average pixel height difference; The height consistency loss value is determined based on the average difference in pixel height and a preset positive threshold.
[0006] In one embodiment, determining the height consistency loss value based on the average pixel height difference and a preset positive threshold includes: The target difference is obtained by calculating the difference between the average difference in pixel height and the preset positive threshold. The high consistency loss value is determined from the target difference and the preset value based on the maximum value function.
[0007] In one embodiment, the target detection network includes a two-stream backbone network, an adaptive feature fusion network, and a detection head. The step of generating predicted values using the target detection network based on the training dataset includes: The dual-stream backbone network extracts grayscale and shape features based on the training dataset. The grayscale features and the shape features are fused using the adaptive feature fusion network to obtain fused features; The predicted value is obtained by making a prediction based on the fusion features using a detection head.
[0008] In one embodiment, calculating the total loss value based on the classification loss value, the bounding box regression loss value, and the high consistency loss value includes: The high consistency loss value is weighted according to a preset weight, and the sum of the weighted high consistency loss value, the classification loss value, and the bounding box regression loss value is calculated as the total loss value.
[0009] In one embodiment, the dual-stream backbone network includes a grayscale feature branch and a shape feature branch, wherein the parameters of the grayscale feature branch and the shape feature branch are not shared. The step of extracting grayscale features and shape features from the training dataset using the dual-stream backbone network includes: The grayscale features are extracted from the training dataset using the grayscale feature branch. The shape features are extracted from the training dataset using the shape feature branch.
[0010] In one embodiment, the step of fusing the grayscale features and the shape features through the adaptive feature fusion network to obtain fused features includes: An attention mechanism is introduced through the adaptive feature fusion network to generate a spatial attention weight map based on the grayscale features and the shape features. The grayscale features and the shape features are fused according to the spatial attention weight map to obtain the fused features.
[0011] Secondly, the present invention provides a battery electrode edge burr detection device, comprising: The acquisition module is used to acquire battery electrode edge images, which include grayscale texture images and physical height shape images; An annotation module is used to annotate the edge types of the grayscale texture image to obtain a training dataset; wherein, the edge types include burr edges and normal edges; A training module is used to input the training dataset into the object detection network for training, resulting in a trained object detection network. The training process includes: generating predicted values using the object detection network based on the training dataset; calculating classification loss, bounding box regression loss, and high consistency loss based on the predicted values and their corresponding ground truth labels; calculating a total loss based on the classification loss, bounding box regression loss, and high consistency loss; and optimizing the training direction of the object detection network based on the total loss. The detection module is used to perform burr detection on the new battery electrode edge image through the trained target detection network.
[0012] Thirdly, the present invention proposes a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, it implements the battery electrode edge burr detection method as described in the first aspect.
[0013] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the battery electrode edge burr detection method as described in the first aspect.
[0014] This invention discloses a method, apparatus, device, and storage medium for detecting burrs on the edges of battery electrodes during the battery production line process. The method involves acquiring edge images of battery electrodes during production, where each edge image includes a grayscale texture image and a physical height and shape image. Edge types are labeled on the grayscale texture image to obtain a training dataset. The edge types include burr edges and normal edges. The training dataset is then input into a target detection network for training, resulting in a trained target detection network. The training process includes: generating predicted values using the target detection network based on the training dataset; calculating classification loss, bounding box regression loss, and height consistency loss based on the predicted values and their corresponding ground truth labels; calculating a total loss based on the classification loss, bounding box regression loss, and height consistency loss; optimizing the training direction of the target detection network based on the total loss; and performing burr detection on new battery electrode edge images using the trained target detection network. In this way, by integrating both texture and height information, misjudgment of a single mode is avoided; the loss of high consistency effectively eliminates false positives caused by factors such as lighting and dirt; the entire process requires no manual intervention and is suitable for large-scale continuous production; in addition, it can be quickly deployed on existing lithium battery production lines. Attached Figure Description
[0015] To more clearly illustrate the technical solution of the present invention, 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 the present invention and should not be regarded as a limitation on the scope of protection of the present invention. In the various drawings, similar components are numbered similarly.
[0016] Figure 1 A flowchart of the battery electrode edge burr detection method proposed in this embodiment is shown. Figure 2 This diagram illustrates another step in the process of the battery electrode edge burr detection method proposed in this embodiment. Figure 3 This illustrates another flowchart of the battery electrode edge burr detection method proposed in this embodiment; Figure 4 A schematic diagram of the battery electrode edge burr detection device proposed in this embodiment is shown.
[0017] Explanation of reference numerals in the attached diagram: 400 - Battery electrode edge burr detection device; 401 - Acquisition module; 402 - Labeling module; 403 - Training module; 404 - Detection module. Detailed Implementation
[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.
[0019] The components of the embodiments of the invention described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.
[0020] In the following, the terms “comprising,” “having,” and their cognates, which may be used in various embodiments of the invention, are intended only to indicate a particular feature, number, step, operation, element, component, or combination thereof, and should not be construed as excluding, firstly, the presence of one or more other features, numbers, steps, operations, elements, components, or combinations thereof, or adding the possibility of one or more features, numbers, steps, operations, elements, components, or combinations thereof.
[0021] Furthermore, the terms "first," "second," and "third" are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.
[0022] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of the invention pertain. Terms (such as those defined in commonly used dictionaries) shall be interpreted as having the same meaning as in their contextual meaning in the relevant technical field and shall not be interpreted as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of the invention.
[0023] Example 1 This disclosure provides a method for detecting edge burrs on battery electrodes, enabling online automated detection of edge burrs on battery electrodes.
[0024] Please see Figure 1 The method for detecting burrs on the edge of the battery electrode includes steps S101 to S104, and each step is described in detail below.
[0025] Step S101: Obtain battery electrode edge images during the battery production line process. The battery electrode edge images include grayscale texture images and physical height shape images.
[0026] In this embodiment, a high-resolution 2D camera device is used to acquire images of the battery electrode edges during the battery production line process. The battery electrode edge images include grayscale texture images and physical height shape images.
[0027] Among them, the grayscale texture image is obtained by a high-resolution industrial camera under specific lighting conditions, reflecting the microstructural details of the electrode edge surface (such as cracks, particle shedding, irregular protrusions, etc.), which is an important basis for judging the location of potential burrs; the physical height shape image is obtained by three-dimensional topography measurement equipment such as laser confocal scanning microscope or white light interferometer, recording the vertical height value corresponding to each pixel point to form a height map, which can intuitively show whether there are abnormal protrusions or depressions at the edge.
[0028] It should be noted that the battery electrode edge images, including grayscale texture images and physical height and shape images, need to be preprocessed to improve the accuracy of subsequent detection and counting. As an example, a sliding window cropping algorithm is used to crop the grayscale texture images and physical height and shape images to the same size, while simultaneously augmenting the dataset.
[0029] Furthermore, median filtering and Gaussian filtering algorithms are used to remove noise and interference in the image, such as granular noise and high-frequency interference.
[0030] Furthermore, by adjusting the brightness, contrast, and color balance of the image, the visualization effect and edge information of the image are enhanced, ensuring that the details of defects can be better captured.
[0031] Furthermore, the images are adjusted to a uniform size and orientation to ensure consistency of the data input to the neural network. Simultaneously, image mean normalization and standard deviation normalization are employed to eliminate differences in brightness and color between different images.
[0032] Step S102: Mark the edge types of the grayscale texture image to obtain a training dataset; wherein the edge types include burr edges and normal edges.
[0033] In this embodiment, edge type annotation is performed on grayscale texture images. Exemplarily, bounding boxes are annotated for polar edges in grayscale texture images, especially areas containing burrs, along with edge category labels, such as normal edges, burr edges, scratches, etc. Furthermore, it is ensured that the annotations on each grayscale texture image completely correspond to its corresponding physical height and shape image, thus constructing a paired training dataset.
[0034] Step S103: Input the training dataset into the object detection network for training to obtain the trained object detection network; wherein, the training process includes: generating predicted values through the object detection network based on the training dataset; calculating classification loss, bounding box regression loss, and high consistency loss based on the predicted values and their corresponding ground truth labels; calculating a total loss based on the classification loss, the bounding box regression loss, and the high consistency loss; and optimizing the training direction of the object detection network based on the total loss.
[0035] In this embodiment, the training dataset is input into the object detection network, and the object detection network learns and predicts the features in the training dataset to obtain the trained object detection network.
[0036] During training, the object detection network generates predicted values based on the training dataset. Based on the predicted values and their corresponding ground truth labels, it calculates classification loss to mitigate the imbalance between positive and negative samples, bounding box regression loss to improve localization accuracy, and high consistency loss to enhance the model's ability to recognize differences in real physical height. The total loss is then calculated based on the classification loss, bounding box regression loss, and high consistency loss. The training direction of the object detection network is then optimized based on the total loss.
[0037] The entire training process uses gradient descent to optimize network parameters until convergence, resulting in a trained object detection network with strong generalization ability.
[0038] Step S104: Perform burr detection on the new battery electrode edge image using the trained target detection network.
[0039] In this embodiment, new battery electrode edge images acquired on the battery production line are input into a trained target detection network for burr detection. The trained target detection network extracts features from the new battery electrode edge images, including edges, textures, and shapes. Further, it obtains edge type predictions (normal edges, burrs, and scratch interference terms) and position regression results through forward propagation. The model output is normalized using a softmax function to obtain the probability distribution of various predicted values. The edge type prediction with the highest probability is used as the battery electrode edge burr detection result, thereby achieving automated online detection of battery electrode edge burrs. Simultaneously, the use of target detection algorithms improves detection efficiency and accuracy.
[0040] Please see Figure 2 In one specific embodiment, the target detection network includes a dual-stream backbone network, an adaptive feature fusion network, and a detection head. Step S103 includes steps S201 to S203, and each step is described in detail below.
[0041] Step S201: Extract grayscale features and shape features from the training dataset using the dual-stream backbone network.
[0042] In this embodiment, grayscale features and shape features are extracted from the training dataset using a dual-stream backbone network.
[0043] In one specific embodiment, the dual-stream backbone network includes a grayscale feature branch and a shape feature branch, wherein the parameters of the grayscale feature branch and the shape feature branch are not shared. Step S201 includes: extracting the grayscale features through the grayscale feature branch based on the training dataset; and extracting the shape features through the shape feature branch based on the training dataset.
[0044] In this embodiment, the original single-input backbone network is split into two feature extraction branches with non-shared parameters: a grayscale feature branch and a shape feature branch. The shape feature branch can employ different convolutional kernel initialization strategies in the initial layers to better capture the edges and gradient information of the height field.
[0045] The grayscale feature branch extracts multi-level texture semantic features from the labeled grayscale texture image, while the shape feature branch extracts spatial topological features from the physical height shape image.
[0046] Step S202: The grayscale features and the shape features are fused using the adaptive feature fusion network to obtain fused features.
[0047] In this embodiment, the grayscale features and the shape features are fused using an Adaptive Feature Fusion Network (AFFN) to obtain fused features.
[0048] In one specific embodiment, step S202 includes: introducing an attention mechanism through the adaptive feature fusion network to generate a spatial attention weight map based on the grayscale features and the shape features; and fusing the grayscale features and the shape features based on the spatial attention weight map to obtain the fused features.
[0049] In this embodiment, an attention mechanism is introduced through an adaptive feature fusion network. The spliced features are then fed into a lightweight convolutional layer to generate a spatial attention weight map. The spatial attention weight map is used to perform weighted fusion of grayscale features and shape features to obtain fused features, thereby enhancing the response of key regions (such as suspected burrs) and suppressing irrelevant background noise.
[0050] Step S203: The detection head makes a prediction based on the fusion features to obtain the predicted value.
[0051] In this embodiment, the final prediction task is performed based on the fused features, and the edge class probability, bounding box offset, and confidence score of each anchor box are output as predicted values.
[0052] Please see Figure 3 In one specific embodiment, the high consistency loss value is calculated based on the predicted value and its corresponding true labeled value, including steps S301 to S303, each step of which is described in detail below.
[0053] Step S301: Determine positive samples based on the predicted values, and determine the anchor frame region and non-anchor frame region in the physical height shape image corresponding to the positive samples based on the anchor frames in the grayscale texture image corresponding to the positive samples.
[0054] In this embodiment, samples with predicted values of burr category are determined as positive samples, and the anchor frame region and non-anchor frame region in the physical height shape image corresponding to the positive sample are determined based on the anchor frame in the grayscale texture image corresponding to the positive sample.
[0055] Step S302: Calculate the difference between the average pixel height of the anchor frame region and the average pixel height of the non-anchor frame region to obtain the average pixel height difference.
[0056] In this embodiment, the average pixel height H_roi corresponding to the anchor frame region and the average pixel height H_bg corresponding to the non-anchor frame region are obtained, and the difference between the two values is used as the average difference in pixel height.
[0057] Step S303: Determine the height consistency loss value based on the average difference in pixel height and a preset positive threshold.
[0058] In this embodiment, the height consistency loss value is determined based on the average difference in pixel height and a preset positive threshold. This loss only applies to positive samples and guides the network to adjust the position or confidence level of the predicted bounding box during backpropagation, making it more consistent with physical reality.
[0059] In one specific embodiment, step S303 includes: calculating the difference between the average pixel height difference and the preset positive threshold to obtain a target difference; and determining the height consistency loss value from the target difference and the preset value based on the maximum value function.
[0060] In this embodiment, the formula for calculating the high consistency loss value L_height is: L_height = max(0, τ-(H_roi - H_bg)), where τ is a preset positive threshold and 0 is a preset value.
[0061] In one specific embodiment, step S103 includes: weighting the high consistency loss value according to a preset weight, and calculating the sum of the weighted high consistency loss value, the classification loss value, and the bounding box regression loss value as the total loss value.
[0062] In this embodiment, the formula for calculating the total loss value L_total is: L_total=L_cls+L_box+λ×L_height, where L_cls is the classification loss value, L_box is the bounding box regression loss value, and λ is a hyperparameter used to control the weight of Lheight in the total loss. Its optimal value is usually determined by the validation set before model training and can be trained during training.
[0063] This embodiment proposes a method for detecting burrs on battery electrode edges. The method acquires battery electrode edge images during the battery production line process. These images include grayscale texture images and physical height / shape images. Edge types are labeled on the grayscale texture images to obtain a training dataset. The edge types include burr edges and normal edges. The training dataset is then input into a target detection network for training, resulting in a trained target detection network. The training process includes: generating predicted values using the target detection network based on the training dataset; calculating classification loss, bounding box regression loss, and height consistency loss based on the predicted values and their corresponding ground truth labels; calculating a total loss based on the classification loss, bounding box regression loss, and height consistency loss; optimizing the training direction of the target detection network based on the total loss; and performing burr detection on new battery electrode edge images using the trained target detection network. This method integrates texture and height information, avoiding misjudgments based on a single modality. The height consistency loss effectively eliminates false positives caused by factors such as lighting and dirt. The entire process requires no manual intervention and is suitable for large-scale continuous production. Furthermore, it can be quickly deployed on existing lithium battery production lines.
[0064] Example 2 Furthermore, this disclosure provides a battery electrode edge burr detection device 400, please refer to [link to relevant documentation]. Figure 4 ,include: The acquisition module 401 is used to acquire a battery electrode edge image, wherein the battery electrode edge image includes a grayscale texture image and a physical height shape image; The annotation module 402 is used to annotate the grayscale texture image with edge types to obtain a training dataset; wherein the edge types include burr edges and normal edges; Training module 403 is used to input the training dataset into the object detection network for training, thereby obtaining the trained object detection network. The training process includes: generating predicted values using the object detection network based on the training dataset; calculating classification loss, bounding box regression loss, and high consistency loss based on the predicted values and their corresponding ground truth labels; calculating a total loss based on the classification loss, bounding box regression loss, and high consistency loss; and optimizing the training direction of the object detection network based on the total loss. The detection module 404 is used to perform burr detection on the new battery electrode edge image through the trained target detection network.
[0065] Optionally, the training module 403 is further configured to determine positive samples based on the predicted values, determine anchor box regions and non-anchor box regions in the physical height shape image corresponding to the positive samples based on the anchor boxes in the grayscale texture image corresponding to the positive samples, calculate the difference between the average pixel height of the anchor box regions and the average pixel height of the non-anchor box regions to obtain the average pixel height difference, and determine the height consistency loss value based on the average pixel height difference and a preset positive threshold.
[0066] Optionally, the training module 403 is further configured to calculate the difference between the average pixel height difference and the preset positive threshold to obtain a target difference; and to determine the height consistency loss value from the target difference and the preset value based on the maximum value function.
[0067] Optionally, the training module 403 is further configured to extract grayscale features and shape features from the training dataset through the dual-stream backbone network; fuse the grayscale features and shape features through the adaptive feature fusion network to obtain fused features; and make predictions based on the fused features using a detection head to obtain the predicted values.
[0068] Optionally, the training module 403 is further configured to weight the high consistency loss value according to a preset weight, and calculate the sum of the weighted high consistency loss value, the classification loss value, and the bounding box regression loss value as the total loss value.
[0069] Optionally, the dual-stream backbone network includes a grayscale feature branch and a shape feature branch, wherein the parameters of the grayscale feature branch and the shape feature branch are not shared. The training module 403 is further configured to extract the grayscale features based on the training dataset through the grayscale feature branch and extract the shape features based on the training dataset through the shape feature branch.
[0070] Optionally, the training module 403 is further configured to introduce an attention mechanism through the adaptive feature fusion network, generate a spatial attention weight map based on the grayscale features and the shape features, and fuse the grayscale features and the shape features based on the spatial attention weight map to obtain the fused features.
[0071] The apparatus provided in this embodiment can perform the steps of the battery electrode edge burr detection method provided in Embodiment 1. To avoid repetition, the steps will not be repeated.
[0072] This embodiment proposes a battery electrode edge burr detection device that acquires battery electrode edge images during the battery production line process. These images include grayscale texture images and physical height / shape images. Edge type annotations are performed on the grayscale texture images to obtain a training dataset. The edge types include burr edges and normal edges. The training dataset is then input into a target detection network for training, resulting in a trained target detection network. The training process includes: generating predicted values using the target detection network based on the training dataset; calculating classification loss, bounding box regression loss, and height consistency loss based on the predicted values and their corresponding ground truth labels; calculating a total loss based on the classification loss, bounding box regression loss, and height consistency loss; optimizing the training direction of the target detection network based on the total loss; and performing burr detection on new battery electrode edge images using the trained target detection network. This approach integrates texture and height information, avoiding misjudgments based on a single modality. The height consistency loss effectively eliminates false positives caused by factors such as lighting and dirt. The entire process requires no manual intervention and is suitable for large-scale continuous production. Furthermore, it can be quickly deployed on existing lithium battery production lines.
[0073] Example 3 Furthermore, this disclosure provides a computer device including a memory and a processor. The memory stores a computer program, which, when executed by the processor, implements the battery electrode edge burr detection method described in Embodiment 1.
[0074] The device provided in this embodiment can perform the steps of the battery electrode edge burr detection method provided in Embodiment 1. To avoid repetition, the steps will not be repeated.
[0075] Example 4 This disclosure provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the battery electrode edge burr detection method described in Embodiment 1.
[0076] In this embodiment, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.
[0077] The computer-readable storage medium provided in this embodiment can implement the battery electrode edge burr detection method provided in Embodiment 1. To avoid repetition, it will not be described again here.
[0078] In all examples shown and described herein, any specific values should be interpreted as merely exemplary and not as limitations; therefore, other examples of exemplary embodiments may have different values.
[0079] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.
[0080] The above-described embodiments are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.
Claims
1. A method for detecting burrs on the edge of battery electrodes, characterized in that, include: Acquire battery electrode edge images during the battery production line process, the battery electrode edge images including grayscale texture images and physical height shape images; Edge type annotation is performed on the grayscale texture image to obtain a training dataset; wherein, the edge type includes burr edges and normal edges; The training dataset is input into the object detection network for training to obtain the trained object detection network. The training process includes: generating predicted values using the object detection network based on the training dataset; calculating classification loss, bounding box regression loss, and high consistency loss based on the predicted values and their corresponding ground truth labels; calculating a total loss based on the classification loss, bounding box regression loss, and high consistency loss; and optimizing the training direction of the object detection network based on the total loss. The trained object detection network is used to detect burrs on the new battery electrode edge images.
2. The method for detecting burrs on the edge of battery electrode sheets according to claim 1, characterized in that, The high consistency loss value is calculated based on the predicted value and its corresponding true labeled value, including: Based on the predicted value, a positive sample is determined, and based on the anchor frame in the grayscale texture image corresponding to the positive sample, the anchor frame region and non-anchor frame region in the physical height shape image corresponding to the positive sample are determined; Calculate the difference between the average pixel height of the anchor frame region and the average pixel height of the non-anchor frame region to obtain the average pixel height difference; The height consistency loss value is determined based on the average difference in pixel height and a preset positive threshold.
3. The method for detecting burrs on the edge of battery electrode sheets according to claim 2, characterized in that, The step of determining the height consistency loss value based on the average difference in pixel height and a preset positive threshold includes: The target difference is obtained by calculating the difference between the average difference in pixel height and the preset positive threshold. The high consistency loss value is determined from the target difference and the preset value based on the maximum value function.
4. The method for detecting burrs on the edge of battery electrode sheets according to claim 1, characterized in that, The target detection network includes a dual-stream backbone network, an adaptive feature fusion network, and a detection head. Generating predicted values using the target detection network based on the training dataset includes: The dual-stream backbone network extracts grayscale and shape features based on the training dataset. The grayscale features and the shape features are fused using the adaptive feature fusion network to obtain fused features; The predicted value is obtained by making a prediction based on the fusion features using a detection head.
5. The method for detecting burrs on the edge of battery electrode sheets according to claim 1, characterized in that, The step of calculating the total loss value based on the classification loss value, the bounding box regression loss value, and the high consistency loss value includes: The high consistency loss value is weighted according to a preset weight, and the sum of the weighted high consistency loss value, the classification loss value, and the bounding box regression loss value is calculated as the total loss value.
6. The method for detecting burrs on the edge of battery electrode sheets according to claim 4, characterized in that, The dual-stream backbone network includes a grayscale feature branch and a shape feature branch. The parameters of the grayscale feature branch and the shape feature branch are not shared. The step of extracting grayscale features and shape features from the training dataset using the dual-stream backbone network includes: The grayscale features are extracted from the training dataset using the grayscale feature branch. The shape features are extracted from the training dataset using the shape feature branch.
7. The method for detecting burrs on the edge of battery electrode sheets according to claim 4, characterized in that, The step of fusing the grayscale features and the shape features through the adaptive feature fusion network to obtain fused features includes: An attention mechanism is introduced through the adaptive feature fusion network to generate a spatial attention weight map based on the grayscale features and the shape features. The grayscale features and the shape features are fused according to the spatial attention weight map to obtain the fused features.
8. A device for detecting burrs on the edge of battery electrodes, characterized in that, include: The acquisition module is used to acquire battery electrode edge images, which include grayscale texture images and physical height shape images; An annotation module is used to annotate the edge types of the grayscale texture image to obtain a training dataset; wherein, the edge types include burr edges and normal edges; A training module is used to input the training dataset into the object detection network for training, resulting in a trained object detection network. The training process includes: generating predicted values using the object detection network based on the training dataset; calculating classification loss, bounding box regression loss, and high consistency loss based on the predicted values and their corresponding ground truth labels; calculating a total loss based on the classification loss, bounding box regression loss, and high consistency loss; and optimizing the training direction of the object detection network based on the total loss. The detection module is used to perform burr detection on the new battery electrode edge image through the trained target detection network.
9. A computer device, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the battery electrode edge burr detection method as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the battery electrode edge burr detection method as described in any one of claims 1 to 7.