Appearance inspection method for battery cells, storage medium, and electronic device

Abstract

The present application discloses an appearance inspection method and apparatus for battery cells, a storage medium, and an electronic device. The present application relates to the technical fields of data processing and artificial intelligence, and other related technical fields. The method comprises: acquiring a cylindrical developed view corresponding to a battery cell to be inspected; performing feature extraction on the cylindrical developed view by means of a target inspection neural network to obtain target feature information corresponding to said battery cell; and inspecting the appearance of said battery cell by means of the target inspection neural network on the basis of the target feature information to obtain a target inspection result corresponding to said battery cell.

Description

Battery cell appearance inspection methods, storage media and electronic devices

[0001] This application claims priority to Chinese Patent Application No. 2024116210802, filed with the Chinese Patent Office on November 13, 2024, the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the fields of data processing technology, artificial intelligence, and other related technologies. Specifically, it relates to a method and apparatus for inspecting the appearance of battery cells, a storage medium, and an electronic device. Background Technology

[0003] Lithium-ion batteries are widely used in daily life due to their advantages such as high energy density, long lifespan, and environmental friendliness. Before entering the market, finished batteries must undergo rigorous appearance inspection. Due to the diversity, complexity, and unpredictable nature of product appearance, appearance defect detection has always been a challenge in visual inspection technology for lithium-ion batteries. To prevent defective products from reaching end customers, the lithium-ion battery industry currently relies heavily on manual, multi-stage appearance inspections. Technical issues

[0004] However, manual inspection is time-consuming and labor-intensive, with low efficiency and productivity, which greatly increases the labor intensity of workers. It is also prone to false detections and missed detections, making it difficult to meet the requirements and standards of inspection and unable to meet market demands.

[0005] Currently, there is no effective solution to the problem that the inspection of the appearance of coated battery cells is relatively inefficient due to the manual inspection method used in related technologies. Technical solutions

[0006] In a first aspect, this application provides a method for inspecting the appearance of a battery cell. The method includes: acquiring a cylindrical unfolded image corresponding to a battery cell to be inspected; extracting features from the cylindrical unfolded image using a target detection neural network to obtain target feature information corresponding to the battery cell to be inspected; and detecting the appearance of the battery cell to be inspected based on the target feature information using the target detection neural network to obtain a target detection result corresponding to the battery cell to be inspected, wherein the target detection result is set to characterize whether there is an abnormality in the appearance of the battery cell to be inspected.

[0007] Secondly, this application provides a computer-readable storage medium storing a program, wherein the program, when running, controls the device where the storage medium is located to execute the battery cell appearance inspection method described in any one of the above claims.

[0008] Thirdly, this application provides an electronic device, which includes one or more processors and a memory, wherein the memory is used to store the one or more processors implementing the cell appearance inspection method described in any one of the above. Beneficial effects

[0009] The beneficial effects provided by this application are as follows: This application employs the following steps: obtaining the cylindrical unfolded image corresponding to the battery cell to be tested; extracting features from the cylindrical unfolded image using a target detection neural network to obtain target feature information corresponding to the battery cell to be tested; and using the target detection neural network to detect the appearance of the battery cell to be tested based on the target feature information to obtain the target detection result corresponding to the battery cell to be tested. The target detection result is used to characterize whether the appearance of the battery cell to be tested is abnormal. This application solves the problem in related technologies where the appearance of coated battery cells is detected manually, resulting in low detection efficiency. In this solution, firstly, features are extracted from the cylindrical unfolded image corresponding to the battery cell to be tested using a target detection neural network to obtain target feature information; then, based on the target feature information, the appearance of the battery cell to be tested is detected using a target detection neural network to obtain the corresponding target detection result, and the presence of any abnormality in the appearance of the battery cell to be tested is determined based on the target detection result. Detecting the appearance of the battery cell to be tested using a target detection neural network achieves automated detection of abnormal battery cell appearances, reduces manual inspection costs, improves detection efficiency and accuracy, and has strong scalability and versatility. Attached Figure Description

[0010] Figure 1 is a flowchart of a battery cell appearance inspection method provided according to an embodiment of this application;

[0011] Figure 2 is a schematic diagram of a feature extraction module provided according to an embodiment of this application;

[0012] Figure 3 is a schematic diagram of a multi-dimensional information aggregation module provided according to an embodiment of this application;

[0013] Figure 4 is a schematic diagram of the deep feature alignment and conversion module provided according to an embodiment of this application;

[0014] Figure 5 is a schematic diagram of a battery cell appearance inspection method provided according to an embodiment of this application;

[0015] Figure 6 is a schematic diagram of a battery cell appearance inspection device provided according to an embodiment of this application;

[0016] Figure 7 is a schematic diagram of an electronic device provided according to an embodiment of this application.

[0017] Implementation methods of this application

[0018] The present application will now be described in conjunction with the implementation steps. Figure 1 is a flowchart of a cell appearance inspection method provided according to an embodiment of the present application. As shown in Figure 1, the method includes the following steps:

[0019] Step S101: Obtain the cylindrical unfolded diagram corresponding to the cell to be tested.

[0020] Optionally, after the battery cell to be tested is wrapped in the coating machine, the transport station uses appearance inspection equipment (e.g., a 2.5D camera and a bar line scanning light source) to obtain a cylindrical unfolded image based on the rotation of the battery cell.

[0021] It should be noted that the 2.5D camera can combine diffuse reflection, specular reflection and low-angle reflection to extract the cylindrical features of the cylindrical cell (i.e. the cell to be tested) to obtain the cylindrical unfolded image.

[0022] Step S102: Extract features from the cylindrical unfolded diagram using a target detection neural network to obtain the target feature information corresponding to the cell to be detected.

[0023] Optionally, feature extraction of the cylindrical unfolded map can be performed using the reparameter alignment fusion (RAF), multi-scale information aggregation (MIA), deep feature alignment conversion (DFAC), and local neighborhood information fusion (LNIF) modules in the YOLOV8 neural network model or other neural network models (i.e., target detection neural networks) to obtain the target feature information corresponding to the battery cell to be detected.

[0024] Step S103: The appearance of the battery cell to be tested is detected by the target detection neural network based on the target feature information to obtain the target detection result corresponding to the battery cell to be tested. The target detection result is set to characterize whether there is an abnormality in the appearance of the battery cell to be tested.

[0025] Optionally, the appearance of the battery cell to be tested can be detected by a target detection neural network based on the target feature information of the battery cell to be tested, so as to obtain the target detection result (i.e. whether there are defects in the appearance of the battery cell to be tested).

[0026] It should be noted that the appearance defects of the coated battery cell cylindrical surface (i.e., the cell to be tested) include, but are not limited to, pits, wrinkles, foreign objects, dirt, and damage. Because the outer shell of the cylindrical battery cell is made of aluminum, which is relatively thin and light, it is easily dented by external forces. The characteristic of its appearance is that the area of ​​the defect is relatively large and the distribution is small on the surface of the cylindrical battery cell. Because the incoming PET blue film of the battery cell has creases, it forms wrinkles when it comes into contact with the cylindrical battery cell during heat shrinking. The characteristic of its appearance is that the wrinkles mainly occupy the entire shoulder height on the surface of the cylindrical battery cell, presenting an irregular strip shape. Due to the influence of the production environment during the production process, foreign objects such as blue film fibers or Teflon adhesive may adhere to the cylindrical battery cell during coating; or welding slag may be present on the surface during welding. The characteristic of its appearance is that the area of ​​the defect is relatively small on the surface of the cylindrical battery cell, and the appearance is an approximately circular or filamentous foreign object. Since the dirt on the PET blue film of the battery cell is mainly due to oven grease adhering to the surface, the defects are characterized by: the dirt covering a large area on the cylindrical battery cell surface, irregular distribution, uneven shape, and obvious color difference. Due to external forces during transportation, the PET blue film of the battery cell may be damaged. The defects are characterized by: the damaged blue film exposing the aluminum shell, obvious color difference, appearing bright white, and irregular shape; and numerous full-weld joints on the cylindrical battery cell.

[0027] It's important to note that to improve detection accuracy, you can use a larger dataset when training the YOLOv8 model and fine-tune and optimize the model. Additionally, combining other techniques and algorithms, including but not limited to object tracking and image enhancement, can further improve detection performance.

[0028] In summary, this method first utilizes a target detection neural network to extract features from the cylindrical unfolded image of the battery cell under test, thereby obtaining the target feature information of the battery cell. Then, based on the target feature information, the target detection neural network is used to detect the appearance of the battery cell under test, obtaining the corresponding target detection results. Based on the target detection results, it is determined whether there are any abnormalities in the appearance of the battery cell under test. By using a target detection neural network to detect the appearance of the battery cell under test, automated detection of battery cell appearance anomalies is achieved, reducing manual inspection costs, improving detection efficiency and accuracy, and demonstrating strong scalability and versatility.

[0029] Optionally, in the battery cell appearance inspection method provided in this application embodiment, the method of extracting features from the cylindrical unfolded image using a target detection neural network to obtain target feature information corresponding to the battery cell to be inspected includes: extracting features from the cylindrical unfolded image using the feature extraction module of the target detection neural network to obtain first feature information; aggregating the first feature information using the multi-dimensional information aggregation module of the target detection neural network to obtain second feature information; aligning and transforming the second feature information using the deep feature alignment and transformation module of the target detection neural network to obtain third feature information; and fusing the third feature information using the local neighborhood information fusion module of the target detection neural network to obtain target feature information.

[0030] In an optional embodiment, the cylindrical unfolded image is input into a YOLOV8 neural network or other neural network model. A reparameter alignment and fusion module (i.e., a feature extraction module) extracts features from the cylindrical unfolded image, converting the feature information into a form understandable by the neural network to obtain first feature information. A multi-dimensional information aggregation module aggregates the first feature information, combining feature information from different dimensions to extract more comprehensive and complete feature information, resulting in second feature information. A deep feature alignment and transformation module aligns and transforms the second feature information, aligning feature information at different levels to improve the expressive power and discriminative power of the features, resulting in third feature information. A local neighborhood information fusion module fuses the third feature information, integrating local and global information to extract more discriminative features, thereby better describing the appearance characteristics of the battery cell under test, to obtain the final target feature information.

[0031] It should be noted that the local neighborhood information fusion module removes the process of generating feature components from the multi-dimensional information aggregation module and adopts a simpler local neighborhood fusion strategy. It processes large-scale features through adaptive average pooling and then fuses them with similar local neighborhood features before performing information fusion, so that localized feature information can be fused with multi-layer neighborhood features.

[0032] By extracting features from the cylindrical unfolded diagram through a target detection neural network and processing them to obtain target feature information, the detection accuracy can be improved, the ability to detect anomalies can be enhanced, and the model performance can be optimized, thereby better describing and judging the appearance features of the battery cell to be tested.

[0033] Optionally, in the battery cell appearance inspection method provided in this application embodiment, the feature extraction module of the target detection neural network extracts features from the cylindrical unfolded image to obtain the first feature information, including: performing a convolution operation on the cylindrical unfolded image at a first scale to obtain a first feature map corresponding to the cylindrical unfolded image; performing a convolution operation on the cylindrical unfolded image at a second scale to obtain a second feature map corresponding to the cylindrical unfolded image, wherein the first scale is larger than the second scale; performing feature calculation and feature fusion on the first feature map and the second feature map to obtain a third feature map; and performing feature extraction on the third feature map to obtain the first feature information.

[0034] In an optional embodiment, Figure 2 is a schematic diagram of the feature extraction module provided according to an embodiment of this application. As shown in Figure 2, the reparameter alignment fusion module (i.e., the feature extraction module) acquires defect features of different scales from the second to fourth layers and the last layer of the backbone. After receiving the cylindrical unfolded image of the cell to be tested, 3×3 (i.e., the first scale) convolution operation and 1×1 (i.e., the second scale) convolution operation are performed on it respectively to obtain a large-scale feature map (i.e., the first feature map) and a small-scale feature map (i.e., the second feature map) of the cylindrical unfolded image. Feature calculation and feature fusion are performed on the large-scale feature map and the small-scale feature map to obtain a third feature map. Feature extraction is performed on the third feature map through 1×1 convolution operation, normalization operation and SiLu activation function operation in the reparameter VGG structure to obtain the global appearance feature information of the coated cylindrical cell (i.e., the cell to be tested). (i.e., the first feature information).

[0035] By extracting and fusing features at multiple scales, feature information at different scales can be fully utilized to improve the expressive power and richness of features, thereby better describing the appearance features of the battery cell under test. This provides a more reliable basis for subsequent anomaly detection and processing, and also improves the accuracy and robustness of target detection.

[0036] Optionally, in the battery cell appearance inspection method provided in this application embodiment, performing feature calculation and feature fusion on the first feature map and the second feature map to obtain the third feature map includes: performing an average pooling operation on the first feature map to obtain a processed first feature map; performing linear interpolation calculation on the second feature map to obtain a processed second feature map; and performing feature fusion on the processed first feature map and the processed second feature map to obtain the third feature map.

[0037] In an optional embodiment, the large-scale feature map (i.e., the first feature map) is subjected to average pooling to obtain a processed first feature map; the small-scale feature map (i.e., the second feature map) is aligned to a (H / 4, W / 4) size using linear interpolation in bilinear mode to obtain a processed second feature map. The processed first and second feature maps are then concatenated and fused to obtain a third feature map.

[0038] By performing feature calculations, adjustments, and fusion on the first and second feature maps, a third feature map is obtained. This allows for full utilization of multi-scale information, improving the expressive power and discriminative power of the features, thereby better describing the appearance characteristics of the battery cell under test and providing a more reliable basis for subsequent processing and judgment.

[0039] Optionally, in the battery cell appearance inspection method provided in this application embodiment, the process of aggregating the first feature information through the multi-dimensional information aggregation module of the target detection neural network to obtain the second feature information includes: performing convolution operation and linear interpolation calculation on the first feature information to obtain multiple feature components corresponding to the first feature information; aggregating the multiple feature components to obtain initial second feature information; performing convolution operation on the initial second feature information to obtain convolved initial second feature information; and processing the convolved initial second feature information through a target activation function to obtain the second feature information.

[0040] In an optional embodiment, FIG3 is a schematic diagram of a multi-dimensional information aggregation module provided according to an embodiment of the present application. As shown in FIG3, the multi-dimensional information aggregation module receives the first feature information output by the reparameter alignment and fusion module. Then, adaptive average pooling and 1×1 convolution operations are performed on the first feature information to generate feature components. and characteristic components The first feature information is then subjected to bilinear interpolation and a 1×1 convolution operation to generate feature components. Multiple feature components The components are spliced ​​together to unify different dimensions and fuse the local information contained in each component, in order to obtain the initial second feature information. .Will Perform a 1×1 convolution operation to obtain the initial second feature information after convolution. Then, use the ReLU activation function (i.e., the target activation function) to obtain the global feature components corresponding to the initial second feature information after convolution. and local feature components Based on global feature components and local feature components By performing calculations, the second feature information can be obtained. The calculation formula is as follows:

[0041]

[0042] in, This is the second feature information; The corresponding weight value; The corresponding weight value; The corresponding weight value.

[0043] By aggregating the first feature information through a multi-dimensional information aggregation module, the second feature information is obtained, which helps to improve the diversity and expressive power of the features, increase the richness of the features, improve the feature description of the battery cell to be detected, and thus improve the accuracy of target detection.

[0044] Optionally, in the battery cell appearance inspection method provided in this application embodiment, the third feature information is obtained by aligning and transforming the second feature information through the deep feature alignment and transformation module of the target detection neural network. This includes: performing an average pooling operation on the second feature information to obtain processed second feature information; performing feature fusion on the processed second feature information to obtain fused second feature information; and performing alignment and transformation on the fused second feature information to obtain the third feature information.

[0045] In an optional embodiment, FIG4 is a schematic diagram of a deep feature alignment and transformation module provided according to an embodiment of the present application. As shown in FIG4, the deep feature alignment and transformation module receives the second feature information output by the multi-dimensional information aggregation module. Then, adaptive average pooling is applied to align the feature map scale to (H / 16, W / 16), resulting in the processed second feature information. The second feature information is mined and processed through the Transformer structure. The deep defect features are obtained by fusing the query vector Q, key vector K, and value vector V corresponding to the processed second feature information. The query vector Q and key vector K are then fused, and the result is smoothed using Softmax before being fused with the value vector V to obtain the fused second feature information. Finally, deep feature mining and alignment transformation are performed on the fused second feature information to output the high-level semantic information of the deep feature alignment transformation process. (i.e., the third characteristic information).

[0046] By performing average pooling, feature fusion, and alignment transformation on the second feature information, the third feature information can be obtained, which can improve the integration and consistency of features, increase the accuracy of features, and provide more reliable feature information for subsequent target detection and anomaly judgment.

[0047] Optionally, in the battery cell appearance inspection method provided in this application embodiment, after detecting the appearance of the battery cell to be inspected based on target feature information through a target detection neural network and obtaining the target detection result corresponding to the battery cell to be inspected, the method further includes: if the target detection result indicates that the appearance of the battery cell to be inspected is abnormal, then feeding back the target detection result to the target object; obtaining the target processing strategy input by the target object; and processing the battery cell to be inspected according to the target processing strategy.

[0048] In an optional embodiment, based on the target detection results, it is determined whether the appearance of the battery cell under test is abnormal. If an abnormality is found, the target detection results are fed back to the target object for subsequent processing. The target object provides a target processing strategy for the battery cell under test based on the received target detection results, including but not limited to: repair, replacement, and scrapping. The battery cell under test is processed accordingly based on the target processing strategy provided by the target object. For example, if the target processing strategy is repair, a repair operation can be performed; if it is replacement, the battery cell can be replaced; if it is scrapping, the battery cell can be disposed of.

[0049] This method enables the detection and subsequent processing of anomalies in battery cells under test, effectively addressing cells with defects and ensuring product quality and safety. Furthermore, by utilizing the processing strategies provided by the target, targeted processing can be implemented, improving efficiency and accuracy.

[0050] Optionally, in the battery cell appearance inspection method provided in this application embodiment, before extracting features from the cylindrical unfolded diagram using a target detection neural network to obtain the target feature information corresponding to the battery cell to be inspected, the method further includes: acquiring a training dataset, wherein the training dataset includes at least: multiple battery cell cylindrical unfolded diagrams and the actual appearance inspection results corresponding to each battery cell cylindrical unfolded diagram; configuring hyperparameters for the initial detection neural network to obtain the configured initial detection neural network; processing and predicting multiple battery cell cylindrical unfolded diagrams using the configured initial detection neural network to obtain the predicted appearance inspection results corresponding to multiple battery cell cylindrical unfolded diagrams; calculating the target loss value based on the actual appearance inspection results and the predicted appearance inspection results; and updating the hyperparameters based on the target loss value to obtain the target detection neural network.

[0051] In an optional embodiment, a suitable initial detection neural network is selected, and the network's hyperparameters are set, including but not limited to: learning rate, batch size, number of iterations, optimizer type, etc. The configured neural network is trained using a training dataset. During training, the network learns to extract features from the unfolded image of the battery cell cylinder and predict the detection result for each image. A loss function (including but not limited to: cross-entropy loss function or mean squared error loss function, etc.) is used to measure the difference between the predicted result (i.e., the predicted appearance detection result) and the ground truth label (i.e., the real appearance detection result). The loss value for each sample is calculated, and the overall average loss value is taken as the target loss value. Based on the target loss value, an optimization algorithm (e.g., gradient descent) is used to update the network's hyperparameters to improve model performance. The training and validation process is repeated until satisfactory performance metrics are achieved or a predetermined number of iterations is reached, and the current neural network is determined as the object detection neural network.

[0052] Through these steps, a target detection neural network can be constructed that can effectively extract features from the unfolded image of the battery cell cylinder, thereby providing support for the appearance inspection of the battery cell.

[0053] It should be noted that Figure 5 is a schematic diagram of the battery cell appearance inspection method provided according to an embodiment of this application. As shown in Figure 5, firstly, a cylindrical unfolded diagram corresponding to the rotation of the battery cell to be inspected is obtained through an inspection device. Then, the feature extraction module of the target detection neural network extracts features from the cylindrical unfolded diagram to obtain first feature information; the first feature information is aggregated through a multi-dimensional information aggregation module to obtain second feature information; the second feature information is aligned and transformed through a deep feature alignment and transformation module to obtain third feature information; and the third feature information is fused through a local neighborhood information fusion module to obtain target feature information. Finally, the appearance of the battery cell to be inspected is detected based on the target feature information by the target detection neural network to obtain the target detection result corresponding to the battery cell to be inspected.

[0054] The battery cell appearance inspection method provided in this application embodiment obtains the cylindrical unfolded image corresponding to the battery cell to be inspected; extracts features from the cylindrical unfolded image using a target detection neural network to obtain target feature information corresponding to the battery cell to be inspected; and then uses the target detection neural network to inspect the appearance of the battery cell to be inspected based on the target feature information to obtain the target detection result corresponding to the battery cell to be inspected. The target detection result is used to characterize whether there are any abnormalities in the appearance of the battery cell to be inspected. This solves the problem of low inspection efficiency in related technologies where the appearance of coated battery cells is inspected manually. In summary, in this solution, firstly, the target detection neural network is used to extract features from the cylindrical unfolded image corresponding to the battery cell to be inspected to obtain the target feature information of the battery cell to be inspected; then, based on the target feature information, the target detection neural network is used to inspect the appearance of the battery cell to be inspected to obtain the corresponding target detection result, and the presence of any abnormalities in the appearance of the battery cell to be inspected is determined based on the target detection result. By using the target detection neural network to inspect the appearance of the battery cell to be inspected, automated detection of battery cell appearance abnormalities is achieved, reducing manual inspection costs, improving inspection efficiency and accuracy, and exhibiting strong scalability and versatility.

[0055] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0056] This application also provides a battery cell appearance inspection device. It should be noted that the battery cell appearance inspection device of this application can be configured to perform the battery cell appearance inspection method provided in this application. The following describes the battery cell appearance inspection device provided in this application.

[0057] Figure 6 is a schematic diagram of a battery cell appearance inspection device according to an embodiment of this application. As shown in Figure 6, the device includes: a first acquisition unit 601, a feature extraction unit 602, and a detection unit 603.

[0058] The first acquisition unit 601 is configured to acquire the cylindrical unfolded diagram corresponding to the battery cell to be tested.

[0059] The feature extraction unit 602 is configured to extract features from the cylindrical unfolded diagram through a target detection neural network to obtain the target feature information corresponding to the cell to be detected.

[0060] The detection unit 603 is configured to detect the appearance of the battery cell to be detected based on the target feature information through a target detection neural network, and obtain the target detection result corresponding to the battery cell to be detected. The target detection result is configured to characterize whether there is any abnormality in the appearance of the battery cell to be detected.

[0061] The battery cell appearance inspection device provided in this application embodiment includes a first acquisition unit 601 acquiring a cylindrical unfolded image corresponding to the battery cell to be inspected; a feature extraction unit 602 extracting features from the cylindrical unfolded image using a target detection neural network to obtain target feature information corresponding to the battery cell to be inspected; and a detection unit 603 detecting the appearance of the battery cell to be inspected based on the target feature information using a target detection neural network to obtain a target detection result corresponding to the battery cell to be inspected. The target detection result is used to characterize whether there are any abnormalities in the appearance of the battery cell to be inspected, thus solving the problem of low efficiency in detecting the appearance of coated battery cells due to manual inspection in related technologies. In this solution, firstly, features are extracted from the cylindrical unfolded image corresponding to the battery cell to be inspected using a target detection neural network to obtain target feature information of the battery cell to be inspected; then, based on the target feature information, the appearance of the battery cell to be inspected is detected using a target detection neural network to obtain a corresponding target detection result, and the appearance of the battery cell to be inspected is determined to be abnormal based on the target detection result. By using a target detection neural network to detect the appearance of the battery cell under test, the detection of abnormal appearance of the battery cell is automated, which also reduces the cost of manual inspection, improves the efficiency and accuracy of inspection, and has strong scalability and versatility.

[0062] Optionally, in the battery cell appearance inspection device provided in this application embodiment, the feature extraction unit includes: a feature extraction module, configured to extract features from the cylindrical unfolded image through the feature extraction module of the target detection neural network to obtain first feature information; an aggregation module, configured to aggregate the first feature information through the multi-dimensional information aggregation module of the target detection neural network to obtain second feature information; a conversion module, configured to align and convert the second feature information through the deep feature alignment conversion module of the target detection neural network to obtain third feature information; and a fusion module, configured to fuse the third feature information through the local neighborhood information fusion module of the target detection neural network to obtain target feature information.

[0063] Optionally, in the battery cell appearance inspection device provided in this application embodiment, the feature extraction module includes: a first convolution submodule, configured to perform a first-scale convolution operation on the cylindrical unfolded image to obtain a first feature map corresponding to the cylindrical unfolded image; a second convolution submodule, configured to perform a second-scale convolution operation on the cylindrical unfolded image to obtain a second feature map corresponding to the cylindrical unfolded image, wherein the first scale is larger than the second scale; a first processing submodule, configured to perform feature calculation and feature fusion on the first feature map and the second feature map to obtain a third feature map; and a feature extraction submodule, configured to extract features from the third feature map to obtain first feature information.

[0064] Optionally, in the battery cell appearance inspection device provided in this application embodiment, the first processing submodule includes: a pooling submodule, configured to perform an average pooling operation on the first feature map to obtain a processed first feature map; a calculation submodule, configured to perform linear interpolation calculation on the second feature map to obtain a processed second feature map; and a fusion submodule, configured to perform feature fusion on the processed first feature map and the processed second feature map to obtain a third feature map.

[0065] Optionally, in the battery cell appearance inspection device provided in this application embodiment, the aggregation module includes: a second processing submodule, configured to perform convolution operation and linear interpolation calculation on the first feature information to obtain multiple feature components corresponding to the first feature information; an aggregation submodule, configured to aggregate information on the multiple feature components to obtain initial second feature information; a third convolution submodule, configured to perform convolution operation on the initial second feature information to obtain convolved initial second feature information; and a third processing submodule, configured to process the convolved initial second feature information through a target activation function to obtain second feature information.

[0066] Optionally, in the battery cell appearance inspection device provided in this application embodiment, the conversion module includes: a pooling submodule, configured to perform an average pooling operation on the second feature information to obtain processed second feature information; a fusion submodule, configured to perform feature fusion on the processed second feature information to obtain fused second feature information; and a conversion submodule, configured to perform alignment conversion on the fused second feature information to obtain third feature information.

[0067] Optionally, in the battery cell appearance inspection device provided in this application embodiment, the device further includes: a feedback unit, configured to, after detecting the appearance of the battery cell to be inspected based on target feature information through a target detection neural network and obtaining the target detection result corresponding to the battery cell to be inspected, if the target detection result indicates that the appearance of the battery cell to be inspected is abnormal, then feed back the target detection result to the target object; a second acquisition unit, configured to acquire the target processing strategy input by the target object; and a first processing unit, configured to process the battery cell to be inspected according to the target processing strategy.

[0068] Optionally, in the battery cell appearance inspection device provided in this application embodiment, the device further includes: a second acquisition unit, configured to acquire a training dataset before extracting features from the cylindrical unfolded diagrams through a target detection neural network to obtain target feature information corresponding to the battery cell to be inspected, wherein the training dataset includes at least: multiple battery cell cylindrical unfolded diagrams and the actual appearance inspection results corresponding to each battery cell cylindrical unfolded diagram; a configuration unit, configured to obtain a configured initial detection neural network by configuring hyperparameters for the initial detection neural network; a second processing unit, configured to process and predict multiple battery cell cylindrical unfolded diagrams through the configured initial detection neural network to obtain predicted appearance inspection results corresponding to multiple battery cell cylindrical unfolded diagrams; a calculation unit, configured to obtain a target loss value by calculating based on the actual appearance inspection results and the predicted appearance inspection results; and an update unit, configured to obtain a target detection neural network by updating the hyperparameters based on the target loss value.

[0069] The battery cell appearance inspection device includes a processor and a memory. The first determining unit 601, the second determining unit 602, the third determining unit 603, the first changing unit 304, etc., mentioned above are all stored in the memory as program units. The processor executes the program units stored in the memory to achieve accurate inspection of the battery cell appearance.

[0070] The processor contains a core, which retrieves the corresponding program unit from memory. One or more cores can be configured, and accurate detection of the battery cell's appearance can be achieved by adjusting the core parameters.

[0071] The memory may include non-permanent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip.

[0072] This application provides a computer-readable storage medium storing a program that, when executed by a processor, implements a method for inspecting the appearance of battery cells. The computer-readable storage medium can be non-volatile or volatile.

[0073] This application provides a processor configured to run a program, wherein the program executes a battery cell appearance inspection method during runtime.

[0074] As shown in Figure 7, this application embodiment provides an electronic device, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it performs the following steps: obtaining a cylindrical unfolded diagram corresponding to the battery cell to be tested; extracting features from the cylindrical unfolded diagram using a target detection neural network to obtain target feature information corresponding to the battery cell to be tested; and detecting the appearance of the battery cell to be tested based on the target feature information using a target detection neural network to obtain a target detection result corresponding to the battery cell to be tested. The target detection result is set to characterize whether there is an abnormality in the appearance of the battery cell to be tested.

[0075] Optionally, the target feature information corresponding to the battery cell to be tested is obtained by extracting features from the cylindrical unfolded image through the target detection neural network, including: extracting features from the cylindrical unfolded image through the feature extraction module of the target detection neural network to obtain first feature information; aggregating the first feature information through the multi-dimensional information aggregation module of the target detection neural network to obtain second feature information; aligning and transforming the second feature information through the deep feature alignment and transformation module of the target detection neural network to obtain third feature information; and fusing the third feature information through the local neighborhood information fusion module of the target detection neural network to obtain target feature information.

[0076] Optionally, the feature extraction module of the object detection neural network extracts features from the cylindrical unfolded image to obtain the first feature information, including: performing a convolution operation on the cylindrical unfolded image at a first scale to obtain a first feature map corresponding to the cylindrical unfolded image; performing a convolution operation on the cylindrical unfolded image at a second scale to obtain a second feature map corresponding to the cylindrical unfolded image, wherein the first scale is larger than the second scale; performing feature calculation and feature fusion on the first feature map and the second feature map to obtain a third feature map; and extracting features from the third feature map to obtain the first feature information.

[0077] Optionally, performing feature calculation and feature fusion on the first feature map and the second feature map to obtain the third feature map includes: performing an average pooling operation on the first feature map to obtain a processed first feature map; performing linear interpolation calculation on the second feature map to obtain a processed second feature map; and performing feature fusion on the processed first feature map and the processed second feature map to obtain the third feature map.

[0078] Optionally, the process of aggregating the first feature information using the multi-dimensional information aggregation module of the object detection neural network to obtain the second feature information includes: performing convolution and linear interpolation on the first feature information to obtain multiple feature components corresponding to the first feature information; aggregating the multiple feature components to obtain initial second feature information; performing convolution on the initial second feature information to obtain convolved initial second feature information; and processing the convolved initial second feature information using a target activation function to obtain the second feature information.

[0079] Optionally, the alignment transformation of the second feature information to obtain the third feature information by the deep feature alignment transformation module of the target detection neural network includes: performing average pooling on the second feature information to obtain the processed second feature information; performing feature fusion on the processed second feature information to obtain the fused second feature information; and performing alignment transformation on the fused second feature information to obtain the third feature information.

[0080] Optionally, after detecting the appearance of the battery cell to be detected based on target feature information using a target detection neural network and obtaining the target detection result corresponding to the battery cell to be detected, the method further includes: if the target detection result indicates that the appearance of the battery cell to be detected is abnormal, then feeding back the target detection result to the target object; obtaining the target processing strategy input by the target object; and processing the battery cell to be detected according to the target processing strategy.

[0081] Optionally, before extracting features from the cylindrical unfolded images using the target detection neural network to obtain the target feature information corresponding to the cell to be detected, the method further includes: acquiring a training dataset, wherein the training dataset includes at least: multiple cylindrical unfolded images of cells and the actual appearance detection results corresponding to each cylindrical unfolded image of cells; configuring hyperparameters for the initial detection neural network to obtain the configured initial detection neural network; processing and predicting multiple cylindrical unfolded images of cells using the configured initial detection neural network to obtain the predicted appearance detection results corresponding to multiple cylindrical unfolded images of cells; calculating the target loss value based on the actual appearance detection results and the predicted appearance detection results; and updating the hyperparameters based on the target loss value to obtain the target detection neural network.

[0082] The devices mentioned in this article can be servers, PCs, tablets, mobile phones, etc.

[0083] This application also provides a computer program product, which, when executed on a data processing device, is suitable for executing an initialization program having the following method steps: obtaining a cylindrical unfolded diagram corresponding to a battery cell to be tested; extracting features from the cylindrical unfolded diagram using a target detection neural network to obtain target feature information corresponding to the battery cell to be tested; and detecting the appearance of the battery cell to be tested based on the target feature information using a target detection neural network to obtain a target detection result corresponding to the battery cell to be tested, wherein the target detection result is set to characterize whether there is an abnormality in the appearance of the battery cell to be tested.

[0084] Optionally, the target feature information corresponding to the battery cell to be tested is obtained by extracting features from the cylindrical unfolded image through the target detection neural network, including: extracting features from the cylindrical unfolded image through the feature extraction module of the target detection neural network to obtain first feature information; aggregating the first feature information through the multi-dimensional information aggregation module of the target detection neural network to obtain second feature information; aligning and transforming the second feature information through the deep feature alignment and transformation module of the target detection neural network to obtain third feature information; and fusing the third feature information through the local neighborhood information fusion module of the target detection neural network to obtain target feature information.

[0085] Optionally, the feature extraction module of the object detection neural network extracts features from the cylindrical unfolded image to obtain the first feature information, including: performing a convolution operation on the cylindrical unfolded image at a first scale to obtain a first feature map corresponding to the cylindrical unfolded image; performing a convolution operation on the cylindrical unfolded image at a second scale to obtain a second feature map corresponding to the cylindrical unfolded image, wherein the first scale is larger than the second scale; performing feature calculation and feature fusion on the first feature map and the second feature map to obtain a third feature map; and extracting features from the third feature map to obtain the first feature information.

[0086] Optionally, performing feature calculation and feature fusion on the first feature map and the second feature map to obtain the third feature map includes: performing an average pooling operation on the first feature map to obtain a processed first feature map; performing linear interpolation calculation on the second feature map to obtain a processed second feature map; and performing feature fusion on the processed first feature map and the processed second feature map to obtain the third feature map.

[0087] Optionally, the process of aggregating the first feature information using the multi-dimensional information aggregation module of the object detection neural network to obtain the second feature information includes: performing convolution and linear interpolation on the first feature information to obtain multiple feature components corresponding to the first feature information; aggregating the multiple feature components to obtain initial second feature information; performing convolution on the initial second feature information to obtain convolved initial second feature information; and processing the convolved initial second feature information using a target activation function to obtain the second feature information.

[0088] Optionally, the alignment transformation of the second feature information to obtain the third feature information by the deep feature alignment transformation module of the target detection neural network includes: performing average pooling on the second feature information to obtain the processed second feature information; performing feature fusion on the processed second feature information to obtain the fused second feature information; and performing alignment transformation on the fused second feature information to obtain the third feature information.

[0089] Optionally, after detecting the appearance of the battery cell to be detected based on target feature information using a target detection neural network and obtaining the target detection result corresponding to the battery cell to be detected, the method further includes: if the target detection result indicates that the appearance of the battery cell to be detected is abnormal, then feeding back the target detection result to the target object; obtaining the target processing strategy input by the target object; and processing the battery cell to be detected according to the target processing strategy.

[0090] Optionally, before extracting features from the cylindrical unfolded images using the target detection neural network to obtain the target feature information corresponding to the cell to be detected, the method further includes: acquiring a training dataset, wherein the training dataset includes at least: multiple cylindrical unfolded images of cells and the actual appearance detection results corresponding to each cylindrical unfolded image of cells; configuring hyperparameters for the initial detection neural network to obtain the configured initial detection neural network; processing and predicting multiple cylindrical unfolded images of cells using the configured initial detection neural network to obtain the predicted appearance detection results corresponding to multiple cylindrical unfolded images of cells; calculating the target loss value based on the actual appearance detection results and the predicted appearance detection results; and updating the hyperparameters based on the target loss value to obtain the target detection neural network.

[0091] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application 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.

[0092] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. 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, create means configured to implement the functions specified in one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams.

[0093] 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 that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0094] These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process, such that the instructions, which execute on the computer or other programmable apparatus, provide steps configured to implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0095] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0096] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0097] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium configured to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

Claims

1. A method for inspecting the appearance of a battery cell, comprising: Obtain the cylindrical unfolded diagram corresponding to the battery cell to be tested; The target feature information corresponding to the battery cell to be tested is obtained by extracting features from the cylindrical unfolded diagram using a target detection neural network. The target detection neural network detects the appearance of the battery cell under test based on the target feature information to obtain the target detection result corresponding to the battery cell under test. The target detection result is set to characterize whether there is an abnormality in the appearance of the battery cell under test.

2. The cell appearance inspection method according to claim 1, wherein By extracting features from the cylindrical unfolded diagram using a target detection neural network, the target feature information corresponding to the battery cell to be tested is obtained, including: The feature extraction module of the target detection neural network extracts features from the cylindrical unfolded image to obtain the first feature information. The first feature information is aggregated by the multi-dimensional information aggregation module of the target detection neural network to obtain the second feature information; The second feature information is aligned and transformed by the deep feature alignment and transformation module of the target detection neural network to obtain the third feature information; The target feature information is obtained by fusing the third feature information through the local neighborhood information fusion module of the target detection neural network.

3. The cell appearance inspection method according to claim 2, wherein The feature extraction module of the target detection neural network extracts features from the cylindrical unfolded image to obtain the first feature information, including: Perform a first-scale convolution operation on the cylindrical unfolded image to obtain a first feature map corresponding to the cylindrical unfolded image; A second-scale convolution operation is performed on the cylindrical unfolded image to obtain a second feature map corresponding to the cylindrical unfolded image, wherein the first scale is larger than the second scale; The first feature map and the second feature map are subjected to feature calculation and feature fusion to obtain the third feature map; The first feature information is obtained by extracting features from the third feature map.

4. The cell appearance inspection method according to claim 3, wherein The third feature map is obtained by performing feature calculation and feature fusion on the first feature map and the second feature map, including: The first feature map is subjected to average pooling to obtain the processed first feature map; Linear interpolation is performed on the second feature map to obtain the processed second feature map; The processed first feature map and the processed second feature map are fused to obtain the third feature map.

5. The cell appearance inspection method according to claim 2, wherein The first feature information is aggregated by the multi-dimensional information aggregation module of the target detection neural network to obtain the second feature information, which includes: Perform convolution and linear interpolation on the first feature information to obtain multiple feature components corresponding to the first feature information; Information aggregation is performed on the multiple feature components to obtain initial second feature information; Perform a convolution operation on the initial second feature information to obtain the convolved initial second feature information; The initial second feature information after convolution is processed by the target activation function to obtain the second feature information.

6. The cell appearance inspection method according to claim 2, wherein The second feature information is aligned and transformed by the deep feature alignment and transformation module of the target detection neural network to obtain the third feature information, including: The second feature information is subjected to average pooling to obtain the processed second feature information; The processed second feature information is fused to obtain the fused second feature information; The fused second feature information is aligned and transformed to obtain the third feature information.

7. The cell appearance inspection method according to claim 1, wherein After detecting the appearance of the battery cell to be tested based on the target feature information using the target detection neural network and obtaining the target detection result corresponding to the battery cell to be tested, the method further includes: If the target detection result indicates that the appearance of the battery cell under test is abnormal, the target detection result is fed back to the target object; Obtain the target processing strategy input by the target object; The battery cell to be tested is processed according to the target processing strategy.

8. The cell appearance inspection method according to claim 1, wherein Before extracting features from the cylindrical unfolded image using a target detection neural network to obtain the target feature information corresponding to the battery cell to be detected, the method further includes: Obtain a training dataset, wherein the training dataset includes at least: multiple cell cylindrical unfolded diagrams and the actual appearance detection results corresponding to each cell cylindrical unfolded diagram; Configure hyperparameters for the initial detection neural network to obtain the configured initial detection neural network; The configured initial detection neural network is used to process and predict the multiple battery cell cylindrical unfolded diagrams to obtain the predicted appearance detection results corresponding to the multiple battery cell cylindrical unfolded diagrams. The target loss value is calculated based on the actual appearance detection results and the predicted appearance detection results. The hyperparameters are updated based on the target loss value to obtain the target detection neural network.

9. A computer-readable storage medium including a stored program, wherein, When the program is running, it controls the storage medium to perform the cell appearance inspection method according to any one of claims 1 to 8 on the device.

10. An electronic device, comprising one or more processors and memory arranged to store one or more programs, wherein, When the one or more programs are executed by the one or more processors, the one or more processors implement the cell appearance inspection method according to any one of claims 1 to 8.

Images