Defect identification method and system for lithium iron phosphate mobile fast charging battery

Abstract

The invention discloses a defect identification method and system for a lithium iron phosphate mobile fast-charging battery, belongs to the technical field of battery defect identification, and is used for solving the problems that in the quality detection process of the lithium iron phosphate fast-charging battery in the prior art, defect features are difficult to effectively extract, multiple defects cannot be detected at the same time, staged treatment is needed, and the detection efficiency is high. And the detection efficiency is low. The method comprises the following steps: acquiring multivariate fusion image data of an original image of the lithium iron phosphate mobile fast charging battery; performing multi-modal data enhancement on the multivariate fusion image data to generate a synthetic defect sample data set; constructing a multi-scale feature fusion network, and performing network training by synthesizing a defect sample data set to obtain a battery defect identification model; identifying the lithium iron phosphate mobile fast charging battery to be detected through the battery defect identification model, and outputting a defect detection result; and according to a defect detection result in a preset time period, updating a confidence coefficient threshold value in the battery defect identification model so as to optimize the battery defect identification model.

Description

Technical Field

[0001] The present invention relates to the technical field of battery defect identification, and in particular to a defect identification method and system for a lithium iron phosphate mobile fast-charging battery. Background Art

[0002] In today's industrial testing landscape, quality inspection is crucial for ensuring product safety and reliability, particularly during the manufacturing process of lithium iron phosphate fast-charge batteries in the new energy industry. However, traditional quality inspection methods face numerous technical limitations that not only impact inspection efficiency but also restrict accuracy, making them unable to meet the stringent quality standards required of lithium iron phosphate fast-charge batteries.

[0003] While traditional manual visual inspection methods have been widely used for some time, their inherent flaws are becoming increasingly apparent. Manual inspection is not only inefficient and susceptible to factors such as fatigue and distraction, resulting in a high rate of missed detections, but also highly dependent on the experience of the inspector, making it difficult to guarantee the consistency and accuracy of test results.

[0004] With the development of computer vision technology, automated inspection solutions based on traditional image processing algorithms have emerged. However, these algorithms, such as edge detection and threshold segmentation, exhibit significant insensitivity when dealing with defects such as tiny cracks, bubble distribution, and impurity contamination found in complex structures such as the internal electrodes and electrolyte coatings of lithium iron phosphate fast-charge batteries. These subtle and complex defects are often obscured by noise or background information, making it difficult for traditional image processing algorithms to effectively extract features, thus affecting the accuracy and reliability of inspection.

[0005] In addition, in small sample defect scenarios, existing machine learning or deep learning models often face the challenge of poor generalization ability. Due to limited training data, it is difficult for the model to learn sufficient feature information, resulting in a high false detection rate in actual applications, affecting the stability and practicality of the detection system. To make matters more complicated, the types of defects that may occur in the production process of lithium iron phosphate fast-charging batteries are diverse, including but not limited to cracks, bubbles, impurity contamination, etc. Traditional detection schemes often require different types of defects to be processed in stages, which not only increases the complexity of the detection process, but also prolongs the detection time and reduces the overall detection efficiency.

[0006] In summary, the existing technology has many shortcomings in the quality detection of lithium iron phosphate fast-charging batteries. There is an urgent need for a high-precision and high-efficiency online detection solution to achieve rapid and accurate identification of defects in key parts such as internal battery electrodes and electrolyte coatings, thereby ensuring the safety and service life of the battery. Summary of the Invention

[0007] An embodiment of the present invention provides a defect identification method and system for a lithium iron phosphate mobile fast-charging battery, which is used to solve the following technical problems: In the existing technology, it is difficult to effectively extract defect features during the quality inspection process of lithium iron phosphate fast-charging batteries, and it is impossible to detect multiple defects at the same time. It requires staged processing and has low detection efficiency.

[0008] The embodiment of the present invention adopts the following technical solutions:

[0009] In one aspect, an embodiment of the present invention provides a method for identifying defects in a lithium iron phosphate mobile fast-charge battery, the method comprising: acquiring multivariate fusion image data of an original image of the lithium iron phosphate mobile fast-charge battery;

[0010] Performing multimodal data enhancement on the multivariate fused image data to generate a synthetic defect sample data set;

[0011] Constructing a multi-scale feature fusion network and performing network training using the synthetic defect sample data set to obtain a battery defect recognition model;

[0012] Identify the lithium iron phosphate mobile fast-charging battery to be tested using the battery defect recognition model and output a defect detection result;

[0013] According to the defect detection results within a preset time period, the confidence threshold in the battery defect recognition model is updated to optimize the battery defect recognition model.

[0014] In a feasible implementation, obtaining multivariate fusion image data of the original image of the lithium iron phosphate mobile fast-charging battery specifically includes:

[0015] The original images of the lithium iron phosphate mobile fast-charging battery sample are collected synchronously by a high-resolution industrial camera, an infrared thermal imager, and an X-ray imager respectively; wherein the original images include the surface image of the battery electrode, the infrared image of the battery electrode, and the X-ray image of the battery electrode;

[0016] Convert the collected original images into a unified image format and adjust them to the same size;

[0017] The adjusted original images are subjected to feature point registration to align the original images, and the aligned original images are subjected to image fusion to obtain the multivariate fused image data.

[0018] In a feasible implementation, feature point registration is performed on the adjusted original image to align the original image, specifically including:

[0019] For each original image, the pixel values of each pixel are compared with those of its adjacent pixels to determine the local extreme value points, and the local extreme value points are determined as the initial feature points;

[0020] Among the initial feature points, feature points with contrast lower than a preset threshold and feature points located on the edge are removed to obtain accurate feature points in each original image;

[0021] The neighborhood of each precise feature point is divided into multiple sub-regions, and the magnitude and direction of the gradient are calculated in each sub-region to form a local gradient direction histogram; the local gradient direction histograms of each sub-region are connected to obtain a feature point descriptor for each precise feature point; the descriptor contains local structural information around the precise feature point;

[0022] Based on the Euclidean distance between feature point descriptors, the matching feature point pairs between the two original images are determined;

[0023] Calculating an affine transformation matrix between each pair of original images based on the matching feature point pairs;

[0024] The original image is resampled based on the affine transformation matrix, so that the pixels in the image are moved according to the rules of the transformation matrix, and image registration is achieved, so that the corresponding feature points of the three original images are aligned in spatial position.

[0025] In a feasible implementation, performing image fusion on the aligned original images to obtain the multi-dimensional fused image data specifically includes:

[0026] Generate corresponding feature vectors for the precise feature points in each subsequent original image, and merge the feature vectors of the three original images into a feature matrix based on the preset weights;

[0027] Performing principal component analysis on the feature matrix using a principal component analysis algorithm to obtain a corresponding eigenvector matrix;

[0028] Multiplying the eigenvector matrix by the feature matrix to project the feature matrix into a principal component space to obtain a projected eigenvector;

[0029] The projected feature vectors are spliced into a one-dimensional vector to obtain the multivariate fusion image data.

[0030] In a feasible implementation, performing multimodal data enhancement on the multivariate fused image data to generate a synthetic defect sample dataset specifically includes:

[0031] Determining, based on the defect locations marked in the original images, the feature vector segments corresponding to each defect type in each original image in the multivariate fused image;

[0032] The feature vector segments corresponding to each defect type are copied and reorganized, and randomly inserted into the multivariate fusion image data corresponding to the defect-free battery electrode image to obtain a synthetic defect sample;

[0033] The synthetic defect samples and the multivariate fused image data corresponding to the original image are combined into the synthetic defect sample data set.

[0034] In a feasible implementation, performing multimodal data enhancement on the multivariate fused image data to generate a synthetic defect sample dataset specifically includes:

[0035] Construct a conditional generative adversarial network and train it using real defect images;

[0036] Inputting the defect-free battery pole piece image into the trained conditional generative adversarial network to obtain a defective composite image; wherein the defect-free battery pole piece image includes a battery pole piece surface image, a battery pole piece infrared image, and a battery pole piece X-ray image;

[0037] Fusing the defect composite images to obtain corresponding multi-element fused image data to obtain a composite defect sample;

[0038] The synthetic defect samples and the multivariate fused image data corresponding to the original image are combined into the synthetic defect sample data set.

[0039] In a feasible implementation, a multi-scale feature fusion network is constructed, and the network is trained using the synthetic defect sample dataset to obtain a battery defect recognition model, specifically including:

[0040] Build the basic network architecture based on the ResNet50 network;

[0041] In each residual block of the basic network architecture, an SE attention module is embedded to obtain an advanced network architecture; wherein the SE attention module includes a compression layer and an excitation layer; the compression layer is used to perform global average pooling on the input feature map to obtain a channel description vector; the excitation layer is used to learn the relationship between channels through a fully connected layer to generate a weight vector; in the SE attention module, the generated weight vector is multiplied by the input feature map channel by channel, and the output feature map is passed to the next residual block for calculation;

[0042] Based on the feature pyramid network, a multi-scale feature fusion layer is constructed; the input of the multi-scale feature fusion layer is the feature map output at the four scales of the advanced network architecture, and the output is the probability distribution of each defect type and the location box of each defect;

[0043] The advanced network architecture and the multi-scale feature fusion layer constitute the multi-scale feature fusion network, and the network is trained using the synthetic defect sample data set to obtain a battery defect recognition model.

[0044] In a feasible implementation, a multi-scale feature fusion layer is constructed based on a feature pyramid network, specifically including:

[0045] Extracting feature maps at four scales of the advanced network architecture to obtain a multi-scale feature map; starting from the high-level semantic feature map of the multi-scale feature map, unifying the number of channels of the multi-scale feature map through 1×1 convolution; and performing step-by-step upsampling and adding it to the previous layer feature to eliminate channel differences;

[0046] Each pyramid layer of the multi-scale feature fusion layer is independently connected to two parallel branches; the parallel branches include a defect type classification branch and a defect position regression branch;

[0047] The defect type classification branch is used to determine the probability distribution of each defect type based on the pyramid features output by the multi-scale feature fusion layer; the defect position regression branch is used to frame the defect position in the image using a bounding box based on the pyramid features.

[0048] In a feasible implementation, updating the dynamic threshold in the battery defect recognition model according to the defect detection results within a preset time period to optimize the battery defect recognition model specifically includes:

[0049] Obtaining defect detection results within a preset time period; wherein the defect detection results at least include a confidence level of the defect detection results;

[0050] Determining a distribution interval of the confidence level of the defect detection result within a preset time period;

[0051] If the distribution interval is a low confidence interval, the confidence threshold in the battery defect recognition model is lowered; if the distribution interval is a high confidence interval, the confidence threshold in the battery defect recognition model is increased to optimize the dynamic adaptability of the battery defect recognition model to the current batch of lithium iron phosphate mobile fast-charging batteries.

[0052] On the other hand, an embodiment of the present invention further provides a defect identification system for a lithium iron phosphate mobile fast-charging battery, the system comprising:

[0053] A data set construction module is used to obtain multi-dimensional fusion image data of the original image of the lithium iron phosphate mobile fast-charging battery; perform multi-modal data enhancement on the multi-dimensional fusion image data to generate a synthetic defect sample data set;

[0054] A model building module is used to build a multi-scale feature fusion network and perform network training using the synthetic defect sample data set to obtain a battery defect recognition model;

[0055] A defect detection module is used to identify the lithium iron phosphate mobile fast-charging battery to be detected through the battery defect recognition model and output the defect detection results; based on the defect detection results within a preset time period, the confidence threshold in the battery defect recognition model is updated to optimize the battery defect recognition model.

[0056] Compared with the prior art, the defect identification method and system for lithium iron phosphate mobile fast-charging batteries provided by the embodiments of the present invention have the following beneficial effects:

[0057] 1. Significantly improve detection accuracy: By using synchronous image acquisition and image fusion of industrial camera arrays, infrared thermal imagers, and X-ray imagers, combined with a deep convolutional neural network model, the defect features in the training data set are highlighted, improving the recognition accuracy of defects such as cracks, bubbles, and contamination in battery electrodes. This improvement greatly enhances the reliability of product quality control.

[0058] 2. Enhanced Small Sample Adaptability: Utilizing generative adversarial networks for multimodal data augmentation, we expand the number of training samples and improve the learning efficiency and recognition accuracy of defect detection models. This feature reduces reliance on large amounts of labeled data, shortens model training cycles, and increases the system's flexibility and applicability in practical applications.

[0059] 3. Improved adaptability: A dynamic threshold mechanism has been introduced to automatically adjust detection parameters based on different production environments and defect characteristics. This feature enhances the robustness and stability of the system and reduces production interruptions and cost waste caused by false alarms.

[0060] 4. Simultaneous detection of multiple defects: During a single inference process, the system can simultaneously output the labels and location information of multiple defect types, enabling simultaneous detection of multiple defects. This feature improves the comprehensiveness and accuracy of detection, providing more comprehensive and detailed data support for production line quality control.

[0061] In summary, the present invention has demonstrated significant beneficial effects in terms of detection accuracy, small sample adaptability, real-time optimization, self-adaptation, and simultaneous detection of multiple defects, providing an efficient and reliable solution for the fields of industrial automation and intelligent manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0063] Figure 1 A flow chart of a defect identification method for a lithium iron phosphate mobile fast-charging battery provided by an embodiment of the present invention;

[0064] Figure 2 A schematic structural diagram of a defect identification system for a lithium iron phosphate mobile fast-charging battery provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0066] The embodiment of the present invention provides a method for identifying defects of a lithium iron phosphate mobile fast-charging battery, such as Figure 1 As shown, the defect identification method of the lithium iron phosphate mobile fast charging battery specifically includes steps S101-S105:

[0067] S101. Acquire multi-element fusion image data of an original image of a lithium iron phosphate mobile fast-charging battery.

[0068] Specifically, the original images of lithium iron phosphate mobile fast-charging battery samples are synchronously collected by a high-resolution industrial camera, an infrared thermal imager, and an X-ray imager respectively; wherein the original images include the surface image of the battery pole piece, the infrared image of the battery pole piece, and the X-ray image of the battery pole piece.

[0069] Furthermore, the collected original images are converted into a unified image format and adjusted to the same size. Feature point registration is then performed on the adjusted original images to align the original images, and the aligned original images are fused to obtain multivariate fused image data.

[0070] As a feasible implementation method, feature point registration is performed on the adjusted original image to align the original image, specifically including:

[0071] For each original image, the pixel values of each pixel are compared with those of its neighbors to determine local extreme points, which are then used as initial feature points. Among these initial feature points, those with contrast below a preset threshold and those located on edges are removed to obtain the precise feature points for each original image.

[0072] The neighborhood of each precise feature point is divided into multiple sub-regions. The magnitude and direction of the gradient are calculated in each sub-region to form a local gradient direction histogram. The local gradient direction histograms of each sub-region are connected to obtain the feature point descriptor of each precise feature point. The descriptor contains the local structural information around the precise feature point.

[0073] Then, based on the Euclidean distance between the feature point descriptors, matching feature point pairs are determined between each pair of original images. Based on these matching feature point pairs, the affine transformation matrix between each pair of original images is calculated. Based on the affine transformation matrix, the original images are resampled, shifting the pixels according to the transformation matrix. This achieves image registration, aligning the corresponding feature points of the three original images in space.

[0074] In one embodiment, feature point extraction is performed on the synchronously collected battery pole piece surface image, battery pole piece infrared image, and battery pole piece X-ray image, respectively, and the feature point descriptors corresponding to the precise feature points are obtained based on the above-mentioned implementation method. Then, a target precise feature point is taken from the calculated battery pole piece surface image, and the Euclidean distance of the feature point descriptor is calculated with each precise feature point in the battery pole piece infrared image. The precise feature point with the smallest Euclidean distance to the target precise feature point is determined as the matching feature point of the target precise feature point, thereby obtaining a matching feature point pair. In the same way, the matching feature points of other precise feature points in the battery pole piece surface image are obtained, thereby obtaining all matching feature point pairs between the battery pole piece surface image and the battery pole piece infrared image. For other image combinations, the same method is used to obtain matching feature point pairs.

[0075] As a feasible implementation method, the aligned original images are subjected to image fusion to obtain multivariate fused image data, specifically including:

[0076] The corresponding feature vector will be generated for the precise feature points in each subsequent original image, and the feature vectors of the three original images will be merged into a feature matrix based on the preset weights.

[0077] Then, a principal component analysis (PCA) algorithm is used to perform principal component analysis on the feature matrix to obtain the corresponding eigenvector matrix. The eigenvector matrix is multiplied by the feature matrix to project the feature matrix into the principal component space, obtaining the projected eigenvectors. The projected eigenvectors are then concatenated into a one-dimensional vector to obtain the multivariate fused image data.

[0078] S102: Perform multimodal data enhancement on the multivariate fusion image data to generate a synthetic defect sample data set.

[0079] Specifically, the present invention provides two data enhancement methods to perform multimodal data enhancement on multivariate fusion image data.

[0080] The first data augmentation method involves determining the feature vector segments corresponding to each defect type in the multivariate fused image based on the defect locations marked in the original images. The feature vector segments corresponding to each defect type are then copied and reorganized, and randomly inserted into the multivariate fused image data corresponding to the defect-free battery electrode image to generate synthetic defect samples. These synthetic defect samples are then combined with the multivariate fused image data corresponding to the original images to form a synthetic defect sample dataset.

[0081] The second data augmentation method involves constructing a conditional generative adversarial network (CGN) and training it with real defect images. Images of defect-free battery electrodes are fed into the trained CGN to generate synthetic defect images. These images include surface images, infrared images, and X-ray images of the electrodes. The synthetic defect images are fused to obtain corresponding multivariate fused image data, generating synthetic defect samples. The synthetic defect samples are then combined with the multivariate fused image data corresponding to the original images to form a synthetic defect sample dataset.

[0082] The synthetic defect samples obtained by the two data enhancement methods and the multivariate fusion image data sets corresponding to the original images are combined to form a complete defect sample dataset.

[0083] S103. Construct a multi-scale feature fusion network, and perform network training using a synthetic defect sample data set to obtain a battery defect recognition model.

[0084] Specifically, we build a basic network architecture based on the ResNet50 network. We also embed the SE attention module into each residual block of the basic network architecture to obtain an advanced network architecture.

[0085] Among them, the SE attention module includes a compression layer and an excitation layer; the compression layer is used to perform global average pooling on the input feature map to obtain a channel description vector; the excitation layer is used to learn the relationship between channels through the fully connected layer and generate a weight vector; in the SE attention module, the generated weight vector is multiplied by the input feature map channel by channel, and the output feature map is passed to the next residual block for calculation.

[0086] Furthermore, based on the feature pyramid network, a multi-scale feature fusion layer is constructed; the input of the multi-scale feature fusion layer is the feature map output at the four scales of the advanced network architecture, and the output is the probability distribution of each defect type and the location box of each defect.

[0087] Furthermore, the advanced network architecture and the multi-scale feature fusion layer are combined to form a multi-scale feature fusion network, and the network is trained using a synthetic defect sample dataset to obtain a battery defect recognition model.

[0088] As a feasible implementation method, a multi-scale feature fusion layer is constructed based on the feature pyramid network, specifically including:

[0089] Feature maps are extracted at four scales of the advanced network architecture to obtain a multi-scale feature map. Starting from the high-level semantic feature map of the multi-scale feature map, the number of channels of the multi-scale feature map is unified through 1×1 convolution. The map is then upsampled step by step and added to the features of the previous layer to eliminate channel differences.

[0090] In one embodiment, the four scales are:

[0091] C2: Stage 2 output (256 channels, downsampled 4 times);

[0092] C3: Stage 3 output (512 channels, downsampled 8 times);

[0093] C4: Stage 4 output (1024 channels, downsampled 16 times);

[0094] C5: Stage 5 output (2048 channels, downsampled 32 times);

[0095] Starting from the high-level semantic feature C5, the number of channels at the four scales is unified to 256 channels through 1×1 convolution, and then upsampled step by step. The pyramid features are added to the features of the previous layer to obtain the following:

[0096] P4=Upsample(P5)+Conv1x1(C4);

[0097] P3=Upsample(P4)+Conv1x1(C3);

[0098] P2=Upsample(P3)+Conv1x1(C2).

[0099] Furthermore, each pyramid layer in the multi-scale feature fusion layer is independently connected to two parallel branches: a defect type classification branch and a defect location regression branch. The defect type classification branch determines the probability distribution of each defect type based on the pyramid features output by the multi-scale feature fusion layer. The defect location regression branch uses the pyramid features to create a bounding box to identify the defect location in the image.

[0100] S104: Identify the lithium iron phosphate mobile fast-charging battery to be tested using a battery defect recognition model, and output a defect detection result.

[0101] Specifically, a high-resolution industrial camera, infrared thermal imager, and X-ray imager are used to simultaneously capture raw images of the lithium iron phosphate mobile fast-charging battery to be inspected, generating multi-dimensional fused image data. This fused image data is then fed into a battery defect recognition model for defect detection. The model outputs the defect types contained in the image, annotates the locations of each defect type in the image with colored boxes, and provides confidence scores for each defect type and location.

[0102] S105 . Update the confidence threshold in the battery defect recognition model according to the defect detection results within a preset time period to optimize the battery defect recognition model.

[0103] Specifically, defect detection results within a preset time period are obtained; wherein the defect detection results at least include a confidence level of the defect detection results.

[0104] Furthermore, the confidence interval of the defect detection results within a preset time period is determined. If the confidence interval is low, the confidence threshold in the battery defect recognition model is lowered; if the confidence interval is high, the confidence threshold in the battery defect recognition model is increased to optimize the dynamic adaptability of the battery defect recognition model to the current batch of lithium iron phosphate mobile fast-charging batteries.

[0105] In addition, the embodiment of the present invention also provides a defect identification system for lithium iron phosphate mobile fast charging batteries, such as Figure 2 As shown, the defect identification system 200 for lithium iron phosphate mobile fast-charging batteries specifically includes:

[0106] The data set construction module 210 is used to obtain multi-element fusion image data of the original image of the lithium iron phosphate mobile fast-charging battery; perform multi-modal data enhancement on the multi-element fusion image data to generate a synthetic defect sample data set;

[0107] A model building module 220 is used to build a multi-scale feature fusion network and perform network training using the synthetic defect sample data set to obtain a battery defect recognition model;

[0108] The defect detection module 230 is used to identify the lithium iron phosphate mobile fast-charging battery to be detected through the battery defect recognition model and output the defect detection result; according to the defect detection result within a preset time period, the confidence threshold in the battery defect recognition model is updated to optimize the battery defect recognition model.

[0109] The various embodiments of the present invention are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the device, apparatus, and non-volatile computer storage medium embodiments are generally similar to the method embodiments, so their descriptions are simplified. For relevant details, refer to the descriptions of the method embodiments.

[0110] The above description of specific embodiments of the present invention is provided. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0111] The foregoing is merely an embodiment of the present invention and is not intended to limit the present invention. It will be apparent to those skilled in the art that various modifications and variations may be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A method for identifying defects in lithium iron phosphate mobile fast-charging batteries, characterized in that: The method comprises: Obtain multi-dimensional fusion image data of the original image of the lithium iron phosphate mobile fast-charging battery; Performing multimodal data enhancement on the multivariate fused image data to generate a synthetic defect sample data set; Constructing a multi-scale feature fusion network and performing network training using the synthetic defect sample data set to obtain a battery defect recognition model; Identify the lithium iron phosphate mobile fast-charging battery to be tested using the battery defect recognition model and output a defect detection result; According to the defect detection results within a preset time period, the confidence threshold in the battery defect recognition model is updated to optimize the battery defect recognition model.

2. The defect identification method for a lithium iron phosphate mobile fast-charging battery according to claim 1, characterized in that: Obtain multi-dimensional fusion image data of the original image of the lithium iron phosphate mobile fast-charging battery, including: The original images of the lithium iron phosphate mobile fast-charging battery sample are collected synchronously by a high-resolution industrial camera, an infrared thermal imager, and an X-ray imager respectively; wherein the original images include the surface image of the battery electrode, the infrared image of the battery electrode, and the X-ray image of the battery electrode; Convert the collected original images into a unified image format and adjust them to the same size; The adjusted original images are subjected to feature point registration to align the original images, and the aligned original images are subjected to image fusion to obtain the multivariate fused image data.

3. The defect identification method for a lithium iron phosphate mobile fast-charging battery according to claim 2, characterized in that: Perform feature point registration on the adjusted original image to align the original image, specifically including: For each original image, the pixel values of each pixel are compared with those of its adjacent pixels to determine the local extreme value points, and the local extreme value points are determined as the initial feature points; Among the initial feature points, feature points with contrast lower than a preset threshold and feature points located on the edge are removed to obtain accurate feature points in each original image; The neighborhood of each precise feature point is divided into multiple sub-regions, and the magnitude and direction of the gradient are calculated in each sub-region to form a local gradient direction histogram; the local gradient direction histograms of each sub-region are connected to obtain a feature point descriptor for each precise feature point; the descriptor contains local structural information around the precise feature point; Based on the Euclidean distance between feature point descriptors, the matching feature point pairs between the two original images are determined; Calculating an affine transformation matrix between each pair of original images based on the matching feature point pairs; The original image is resampled based on the affine transformation matrix, so that the pixels in the image are moved according to the rules of the transformation matrix, and image registration is achieved, so that the corresponding feature points of the three original images are aligned in spatial position.

4. The defect identification method for a lithium iron phosphate mobile fast-charging battery according to claim 3, characterized in that: Performing image fusion on the aligned original images to obtain the multi-dimensional fused image data specifically includes: Generate corresponding feature vectors for the precise feature points in each subsequent original image, and merge the feature vectors of the three original images into a feature matrix based on the preset weights; Performing principal component analysis on the feature matrix using a principal component analysis algorithm to obtain a corresponding eigenvector matrix; Multiplying the eigenvector matrix by the feature matrix to project the feature matrix into a principal component space to obtain a projected eigenvector; The projected feature vectors are spliced into a one-dimensional vector to obtain the multivariate fusion image data.

5. The defect identification method for a lithium iron phosphate mobile fast-charging battery according to claim 1, characterized in that: Performing multimodal data enhancement on the multivariate fusion image data to generate a synthetic defect sample dataset specifically includes: Determining, based on the defect locations marked in the original images, the feature vector segments corresponding to each defect type in each original image in the multivariate fused image; The feature vector segments corresponding to each defect type are copied and reorganized, and randomly inserted into the multivariate fusion image data corresponding to the defect-free battery electrode image to obtain a synthetic defect sample; The synthetic defect samples and the multivariate fused image data corresponding to the original image are combined into the synthetic defect sample data set.

6. The defect identification method for a lithium iron phosphate mobile fast-charging battery according to claim 1, characterized in that: Performing multimodal data enhancement on the multivariate fusion image data to generate a synthetic defect sample dataset, specifically including: Construct a conditional generative adversarial network and train it using real defect images; Inputting the defect-free battery pole piece image into the trained conditional generative adversarial network to obtain a defective composite image; wherein the defect-free battery pole piece image includes a battery pole piece surface image, a battery pole piece infrared image, and a battery pole piece X-ray image; Fusing the defect composite images to obtain corresponding multi-element fused image data to obtain a composite defect sample; The synthetic defect samples and the multivariate fused image data corresponding to the original image are combined into the synthetic defect sample data set.

7. The defect identification method for a lithium iron phosphate mobile fast-charging battery according to claim 1, characterized in that: Construct a multi-scale feature fusion network and perform network training using the synthetic defect sample dataset to obtain a battery defect recognition model, specifically including: Build the basic network architecture based on the ResNet50 network; In each residual block of the basic network architecture, an SE attention module is embedded to obtain an advanced network architecture; wherein the SE attention module includes a compression layer and an excitation layer; the compression layer is used to perform global average pooling on the input feature map to obtain a channel description vector; the excitation layer is used to learn the relationship between channels through a fully connected layer to generate a weight vector; in the SE attention module, the generated weight vector is multiplied by the input feature map channel by channel, and the output feature map is passed to the next residual block for calculation; Based on the feature pyramid network, a multi-scale feature fusion layer is constructed; the input of the multi-scale feature fusion layer is the feature map output at the four scales of the advanced network architecture, and the output is the probability distribution of each defect type and the location box of each defect; The advanced network architecture and the multi-scale feature fusion layer constitute the multi-scale feature fusion network, and the network is trained using the synthetic defect sample data set to obtain a battery defect recognition model.

8. The defect identification method for a lithium iron phosphate mobile fast-charging battery according to claim 7, characterized in that: Based on the feature pyramid network, a multi-scale feature fusion layer is constructed, which includes: Extracting feature maps at four scales of the advanced network architecture to obtain a multi-scale feature map; starting from the high-level semantic feature map of the multi-scale feature map, unifying the number of channels of the multi-scale feature map through 1×1 convolution; and performing step-by-step upsampling and adding it to the previous layer feature to eliminate channel differences; Each pyramid layer of the multi-scale feature fusion layer is independently connected to two parallel branches; the parallel branches include a defect type classification branch and a defect position regression branch; The defect type classification branch is used to determine the probability distribution of each defect type based on the pyramid features output by the multi-scale feature fusion layer; the defect position regression branch is used to frame the defect position in the image using a bounding box based on the pyramid features.

9. The defect identification method for a lithium iron phosphate mobile fast-charging battery according to claim 1, characterized in that: Updating the dynamic threshold in the battery defect recognition model according to the defect detection results within a preset time period to optimize the battery defect recognition model specifically includes: Obtaining defect detection results within a preset time period; wherein the defect detection results at least include a confidence level of the defect detection results; Determining a distribution interval of the confidence level of the defect detection result within a preset time period; If the distribution interval is a low confidence interval, the confidence threshold in the battery defect recognition model is lowered; if the distribution interval is a high confidence interval, the confidence threshold in the battery defect recognition model is increased to optimize the dynamic adaptability of the battery defect recognition model to the current batch of lithium iron phosphate mobile fast-charging batteries.

10. A defect identification system for lithium iron phosphate mobile fast-charging batteries, characterized in that: The system comprises: A data set construction module is used to obtain multi-dimensional fusion image data of the original image of the lithium iron phosphate mobile fast-charging battery; perform multi-modal data enhancement on the multi-dimensional fusion image data to generate a synthetic defect sample data set; A model building module is used to build a multi-scale feature fusion network and perform network training using the synthetic defect sample data set to obtain a battery defect recognition model; A defect detection module is used to identify the lithium iron phosphate mobile fast-charging battery to be detected through the battery defect recognition model and output the defect detection results; based on the defect detection results within a preset time period, the confidence threshold in the battery defect recognition model is updated to optimize the battery defect recognition model.

Images