A method and apparatus for inspecting the appearance of battery negative electrodes based on deep learning

Abstract

This application provides a method and apparatus for inspecting the appearance of a battery negative electrode based on deep learning, belonging to the field of battery inspection technology. In this method, an image reflecting the appearance of the negative electrode post of the battery to be inspected can be generated based on the appearance images of the negative electrode post in at least four preset positions. Then, a battery inspection model is used to identify the image to be inspected, accurately identifying the target battery type of the battery to be inspected and determining the target defect detection model corresponding to the target battery type. Then, by comprehensively considering multi-dimensional physical inspection data and the image to be inspected, a comprehensive feature vector with stronger feature representation can be obtained, enabling the target defect detection model to quickly and accurately predict and judge the type and location of defects in the battery appearance based on the comprehensive feature vector. This not only improves the compatibility of appearance inspection for different battery types, but also effectively improves the accuracy and efficiency of defect detection.

Description

Technical Field

[0001] This application relates to the field of battery testing technology, and in particular to a method and apparatus for detecting the appearance of a battery negative electrode based on deep learning. Background Technology

[0002] With the continuous development of technology, more and more smart products are appearing in people's lives. However, the operation of smart products is inseparable from power sources, especially large smart devices, which require massive batteries to provide power output. Batteries typically encapsulate multiple cells together in a single housing frame and connect to the outside through a unified boundary, ultimately providing power to the smart device. During the battery manufacturing process, some batteries may not be properly sealed, resulting in electrolyte leakage and surface corrosion. Additionally, bumps and scratches during production can cause defects on the battery surface. These defects significantly affect the safety and quality of the battery, therefore, rigorous testing for battery surface defects is necessary.

[0003] Currently, battery appearance defect detection mostly relies on manual inspection by staff, which is not only inefficient but also lacks accuracy due to the human element. Some methods use traditional machine vision algorithms, but these are usually based on images obtained under specific imaging conditions, which are difficult to reproduce in actual application. Furthermore, they can typically only identify a limited number of defect types, making them less applicable and less effective in actual production scenarios. Summary of the Invention

[0004] This application provides a method and apparatus for detecting the appearance of a battery negative electrode based on deep learning. The technical solution adopted in this application is as follows:

[0005] Firstly, a deep learning-based method for detecting the appearance of a battery negative electrode is provided, the method comprising:

[0006] Acquire multidimensional physical detection data of the battery under test and appearance images of the negative terminal of the battery under test in at least four preset positions;

[0007] Based on the appearance images of the negative electrode post at least four preset locations, generate the image to be detected;

[0008] The image to be detected is input into the battery detection model to determine the target battery type of the battery to be detected.

[0009] Based on the target battery type, a target defect detection model corresponding to the target battery type is determined from a preset defect detection model library;

[0010] The multidimensional physical detection data and the image to be detected are fused to obtain a comprehensive feature vector;

[0011] The integrated feature vector is input into the target defect detection model, and the detection result for the negative terminal of the battery under test is output; the detection result includes the defect type of the negative terminal and the location of the defect type in the negative terminal of the battery under test.

[0012] In one embodiment of this application, the step of generating an image to be detected based on the appearance images of the negative electrode post at at least four preset orientations includes:

[0013] Based on the imaging distance of the negative electrode post appearance images at at least four preset orientations, the size of the negative electrode post appearance images at the at least four preset orientations is calibrated.

[0014] Based on a preset pixel mapping table, the negative pole appearance images at at least four preset orientations after size calibration are mapped onto a preset negative pole model to obtain a local image corresponding to each negative pole appearance image; wherein, the pixel mapping table is used to characterize the mapping relationship between the key points of the negative pole model and the pixels of the negative pole appearance images at different preset orientations.

[0015] The overlapping portions of the local images are stitched together to obtain the image to be detected.

[0016] In one embodiment of this application, the step of inputting the image to be detected into a battery detection model to determine the target battery type of the battery to be detected includes:

[0017] The image to be detected is input into the battery detection model, and the confidence scores between the image to be detected and various battery types are output. The battery detection model is trained based on negative electrode sample images of sample batteries of various battery types.

[0018] The maximum value among the confidence scores between the image to be detected and various battery types is determined as the target confidence score;

[0019] If the target confidence level is greater than the confidence level threshold, the battery type corresponding to the target confidence level is determined as the target battery type.

[0020] In one embodiment of this application, the step of determining the target defect detection model corresponding to the target battery type from a preset defect detection model library based on the target battery type includes:

[0021] Obtain a preset model lookup table; wherein the model lookup table represents different defect detection models corresponding to different battery types;

[0022] Based on the target battery type and the model comparison table, the target fault detection model is determined from among multiple defect detection models in the defect detection model library.

[0023] In one embodiment of this application, the step of fusing features of the multidimensional physical detection data and the image to be detected to obtain a comprehensive feature vector includes:

[0024] Feature extraction is performed on the multidimensional physical detection data and the image to be detected to obtain physical feature vectors and visual feature vectors;

[0025] The physical feature vector and the visual feature vector are fused to obtain the comprehensive feature vector.

[0026] In one embodiment of this application, the step of fusing the physical feature vector and the visual feature vector to obtain the comprehensive feature vector includes:

[0027] Obtain a first weight for the physical feature vector and a second weight for the visual feature vector;

[0028] Based on the physical feature vector and the first weight, a first weighted vector is obtained; and based on the visual feature vector and the second weight, a second weighted vector is obtained.

[0029] The comprehensive feature vector is obtained based on the first weighted vector and the second weighted vector.

[0030] In one embodiment of this application, the image to be detected includes a visible light image and an ultrasonic image; wherein, the visible light image is generated based on the appearance images of the negative electrode post at at least four preset positions acquired by a visible light sensor, and the ultrasonic image is generated based on the appearance images of the negative electrode post at at least four preset positions acquired by an ultrasonic sensor.

[0031] The step of extracting features from the image to be detected to obtain a visual feature vector includes:

[0032] Feature extraction is performed on the visible light image and the ultrasonic image respectively to obtain visible light feature vector and ultrasonic feature vector;

[0033] The step of obtaining the second weighted vector based on the visual feature vector and the second weight includes:

[0034] Based on the second weight and the preset weight ratio, a third weight for the visible light feature vector and a fourth weight for the ultrasonic feature vector are determined.

[0035] The second weighted vector is obtained based on the visible light feature vector and the third weight, and the ultrasonic feature vector and the fourth weight.

[0036] In one embodiment of this application, any defect detection model in the detection model library is obtained through the following training method:

[0037] Based on the defect detection model and the model lookup table, the battery type corresponding to the defect detection model is determined;

[0038] From a variety of battery sample data in a sample database, target battery training samples corresponding to the battery type are determined; wherein, the variety of battery sample data is used to train defect detection models corresponding to the various battery types;

[0039] Based on the target battery training samples, a preset initial model is trained until the initial model meets the training cutoff condition, thus obtaining the trained defect detection model.

[0040] In one embodiment of this application, after the step of inputting the comprehensive feature vector into the target defect detection model and outputting the detection result for the negative electrode post of the battery to be tested, the method further includes:

[0041] Generate defect identification information to indicate the location of the defect type at the negative terminal of the battery under test;

[0042] The control display device displays the defect identification information on the image to be inspected.

[0043] Secondly, based on the same inventive concept, a deep learning-based battery negative electrode appearance inspection device is provided, the device comprising:

[0044] The data acquisition module is used to acquire multidimensional physical detection data of the battery under test and the appearance images of the negative terminal of the battery under test in at least four preset positions.

[0045] The image generation module is used to generate an image to be detected based on the appearance images of the negative electrode posts at the at least four preset orientations;

[0046] The type determination module is used to input the image to be detected into the battery detection model to determine the target battery type of the battery to be detected;

[0047] The model determination module is used to determine the target defect detection model corresponding to the target battery type from a preset defect detection model library based on the target battery type.

[0048] The feature determination module is used to perform feature fusion on the multidimensional physical detection data and the image to be detected to obtain a comprehensive feature vector.

[0049] The defect detection module is used to input the comprehensive feature vector into the target defect detection model and output the detection result for the negative terminal of the battery to be tested; the detection result includes the defect type of the negative terminal and the location of the defect type in the negative terminal of the battery to be tested.

[0050] In one embodiment of this application, the image generation module includes:

[0051] The image calibration submodule is used to perform size calibration on the negative electrode appearance images at at least four preset orientations based on the imaging distance of the negative electrode appearance images at at least four preset orientations.

[0052] The pixel mapping submodule is used to map the negative pole appearance images at at least four preset orientations after size calibration onto a preset negative pole model based on a preset pixel mapping table, so as to obtain a local image corresponding to each negative pole appearance image; wherein, the pixel mapping table is used to characterize the mapping relationship between the key points of the negative pole model and the pixels of the negative pole appearance images at different preset orientations.

[0053] The image stitching submodule is used to stitch together the overlapping parts between the local images to obtain the image to be detected.

[0054] In one embodiment of this application, the type determination module includes:

[0055] The first confidence determination submodule is used to input the image to be detected into the battery detection model and output the confidence between the image to be detected and multiple battery types; the battery detection model is trained based on negative electrode sample images of sample batteries of multiple battery types.

[0056] The second confidence determination submodule is used to determine the maximum value among the confidence scores between the image to be detected and multiple battery types as the target confidence score;

[0057] The battery type determination submodule is used to determine the battery type corresponding to the target confidence level as the target battery type of the model determination module when the target confidence level is greater than the confidence threshold.

[0058] In one embodiment of this application, the model determination module includes:

[0059] The lookup table acquisition submodule is used to acquire a preset model lookup table; wherein, the model lookup table represents different defect detection models corresponding to different battery types;

[0060] The detection model determination submodule is used to determine the target fault detection model from multiple defect detection models in the defect detection model library based on the target battery type and the model lookup table.

[0061] In one embodiment of this application, the feature determination module includes:

[0062] The feature extraction submodule is used to extract features from the multidimensional physical detection data and the image to be detected, respectively, to obtain physical feature vectors and visual feature vectors.

[0063] The feature fusion submodule is used to fuse the physical feature vector and the visual feature vector to obtain the comprehensive feature vector.

[0064] In one embodiment of this application, the feature fusion submodule includes:

[0065] A weight acquisition unit is used to acquire a first weight for the physical feature vector and a second weight for the visual feature vector.

[0066] A vector weighting unit is used to obtain a first weighted vector based on the physical feature vector and the first weight; and to obtain a second weighted vector based on the visual feature vector and the second weight.

[0067] The feature fusion unit is used to obtain the comprehensive feature vector based on the first weighted vector and the second weighted vector.

[0068] In one embodiment of this application, the image to be detected includes a visible light image and an ultrasonic image; wherein, the visible light image is generated based on the appearance images of the negative electrode post at at least four preset positions acquired by a visible light sensor, and the ultrasonic image is generated based on the appearance images of the negative electrode post at at least four preset positions acquired by an ultrasonic sensor.

[0069] The feature extraction submodule includes:

[0070] The feature extraction unit is used to extract features from the visible light image and the ultrasonic image respectively, to obtain visible light feature vector and ultrasonic feature vector;

[0071] The vector weighting unit includes:

[0072] The weight determination subunit is used to determine the third weight for the visible light feature vector and the fourth weight for the ultrasonic feature vector based on the second weight and a preset weight ratio.

[0073] The vector weighting subunit is used to obtain the second weighting vector based on the visible light feature vector and the third weight, as well as the ultrasonic feature vector and the fourth weight.

[0074] In one embodiment of this application, the deep learning-based battery negative electrode appearance inspection device further includes:

[0075] The type determination module is used to determine the battery type corresponding to the defect detection model based on the defect detection model and the model lookup table.

[0076] The sample determination module is used to determine the target battery training sample corresponding to the battery type from multiple battery sample data in the sample database; wherein, the multiple battery sample data is used to train the defect detection model corresponding to the multiple battery types;

[0077] The model training module is used to train a preset initial model based on the target battery training samples until the initial model meets the training cutoff condition, thereby obtaining the trained defect detection model.

[0078] In one embodiment of this application, the deep learning-based battery negative electrode appearance inspection device further includes:

[0079] An information generation module is used to generate defect identification information indicating the location of the defect type on the negative terminal of the battery under test.

[0080] An information display module is used to control a display device to display the defect identification information on the image to be inspected.

[0081] In summary, the above-mentioned deep learning-based method and device for detecting the appearance of battery negative electrodes have the following technical advantages:

[0082] This application first acquires multidimensional physical inspection data of the battery under test and appearance images of the negative terminal of the battery under test at at least four preset positions. Based on the appearance images of the negative terminal in at least four preset positions, an image under test that comprehensively reflects the appearance of the negative terminal can be generated. Then, a battery detection model is used to identify the image under test, which can accurately identify the target battery type of the battery under test and determine the target defect detection model corresponding to the target battery type. Then, by comprehensively considering the multidimensional physical inspection data and the image under test and fusing the two types of data, a comprehensive feature vector with stronger feature representation can be obtained, enabling the target defect detection model to output more accurate and reliable detection results based on the comprehensive feature vector. This application, by matching a suitable target defect detection model to the battery under test according to the battery type, can quickly and accurately predict and judge the type and location of defects in the battery appearance. This not only improves the compatibility of appearance inspection for different battery types, but also effectively improves the accuracy and efficiency of defect detection. Attached Figure Description

[0083] Figure 1 A schematic diagram illustrating the steps of a deep learning-based battery negative electrode appearance detection method provided in this application embodiment;

[0084] Figure 2 This is a schematic diagram of the functional modules of a battery negative electrode appearance inspection device based on deep learning, provided in an embodiment of this application. Detailed Implementation

[0085] The technical solutions in this application will now be described with reference to the accompanying drawings.

[0086] It should be noted that during the battery production process, due to manufacturing process reasons, some batteries may not be properly sealed, resulting in electrolyte leakage and corrosion of the battery surface. Alternatively, bumps and scratches during the production process may cause defects such as bulges, dents, scratches, dirt, electrolyte residue, and damage to the explosion-proof valve membrane on the battery surface. These defects greatly affect the safety and quality of the battery, so it is necessary to conduct strict testing on battery surface defects.

[0087] With the rapid development of new materials and new energy vehicles, lithium battery technology is increasingly widely used in high-tech fields. It is a high-performance battery that uses lithium metal or lithium alloys as positive / negative electrode materials and a non-aqueous electrolyte solution. Due to its high energy density, low self-discharge rate, and long lifespan, it is widely used in new energy vehicles, aerospace, mobile devices, and other fields. For lithium batteries, the negative electrode is usually made of materials such as graphite or silicon. These materials are relatively soft and more sensitive to physical damage, making them more susceptible to damage during manufacturing or processing. Furthermore, the coating and compaction steps of the negative electrode material require precise control; any deviation can lead to quality problems, affecting the battery's charge / discharge performance and cycle life. Therefore, more stringent appearance inspection is necessary.

[0088] In related technologies, the basic process of machine vision-based methods includes image segmentation, morphological processing, and feature value calculation. However, most existing algorithms can typically only identify defects such as scratches and dents. Other problems exist in actual battery production, such as peeling at the battery opening, gaps, rolled edges, triangular rolled edges, tilted negative electrode caps, and leaking sealing rings. These defects are more likely to occur and have a greater impact on the production process. Furthermore, most algorithms are based on images acquired under specific imaging conditions, making them difficult to reproduce in real-world applications. Moreover, these algorithms lack robustness analysis for potential problems in actual production. In battery production, oil contamination on the production line can contaminate passing batteries, affecting the effectiveness of normal battery defect detection.

[0089] Based on this, the inventive concept of this application is proposed: by stitching together the appearance images of the negative electrode post at least four preset orientations, a test image that comprehensively reflects the appearance of the negative electrode post can be generated. Then, by using a battery detection model to identify the test image, the target battery type of the battery to be tested can be accurately identified, and the target defect detection model corresponding to the target battery type can be determined. Then, by comprehensively considering multi-dimensional physical detection data and the test image, and fusing the two types of data, a comprehensive feature vector with stronger feature representation can be obtained. This allows the target defect detection model to quickly and accurately predict and judge the type and location of defects in the battery appearance based on the comprehensive feature vector. This not only improves the compatibility of appearance inspection for different battery types, but also effectively improves the accuracy and efficiency of defect detection.

[0090] Reference Figure 1 The embodiments of this application provide a deep learning-based method for detecting the appearance of a battery negative electrode, which may include the following steps:

[0091] S101: Acquire multidimensional physical test data of the battery under test and negative terminal appearance images of the battery under test in at least four preset positions.

[0092] In this embodiment, a physical inspection device and an image inspection device can be configured on the production line of the battery to be tested. The physical inspection device is used to collect multi-dimensional physical inspection data of the battery to be tested, and the image inspection device (e.g., a camera) is used to collect images of the negative terminal of the battery to be tested from at least four preset positions.

[0093] In practical implementation, to reduce the number of image detection devices and the cost of data acquisition, a turntable can be used to transport the batteries to be tested. The image detection device is set on the side of the turntable and can rotate the turntable at preset time intervals. Each time the turntable completes a rotation, for example, every time the turntable rotates 90 degrees, the image detection device is controlled to acquire an image of the appearance of the negative terminal. In this way, the image of the appearance of the negative terminal at least four preset positions can be acquired by using only one image detection device.

[0094] In this embodiment, considering the correlation between various defects and the physical testing data of the battery under test, the multidimensional physical testing data may include, but is not limited to, the weight, negative electrode size, particle size distribution, and graphitization degree of the battery under test.

[0095] In practical implementation, high-precision pressure sensors installed on the production line can be used to collect the weight of the battery under test. Deviations in the weight indicate potential defects such as dirt, peeling, or breakage. Laser scanning technology can be used to scan the three-dimensional shape of the battery's negative electrode, thus obtaining its dimensions. Deviations in these dimensions may indicate defects such as peeling, creasing, tilted negative electrode caps, and leakage of the sealing ring. A laser particle size analyzer can be used to test the particle size distribution of the negative electrode material. Particle size distribution affects the compaction density and electrochemical performance of the battery electrode sheets; poor particle size distribution may lead to defects such as porosity, cracking, and detachment. X-ray diffraction (XRD) can be used to detect the graphitization degree of the negative electrode material. Poor graphitization reduces the mechanical strength of the battery's negative electrode, making it more prone to defects such as bulging, pitting, and cracking.

[0096] S102: Generate an image to be detected based on the appearance images of the negative electrode post at least four preset orientations.

[0097] In this embodiment, by stitching together the appearance images of the negative terminal of the battery under test from at least four preset positions, an image to be tested can be obtained. This image is a global image of the negative terminal, which can comprehensively reflect the appearance of the negative terminal and effectively avoid missing defects. At the same time, by acquiring appearance images of the negative terminal from multiple positions, the difficulty of image stitching is reduced, and image distortion is effectively reduced, avoiding image falsification.

[0098] S103: Input the image to be detected into the battery detection model to determine the target battery type of the battery to be detected.

[0099] In this embodiment, after generating the image to be detected, the battery detection model can be used to identify the image to obtain the target battery type of the battery to be detected.

[0100] In this embodiment, considering that the appearance and physical characteristics of batteries under test may differ depending on the battery type, a battery detection model can be used to automatically identify the battery type of the battery under test. This improves the compatibility of appearance inspection for different battery types. In other words, it can detect appearance defects in batteries of multiple types, thereby effectively improving the compatibility and accuracy of the solution.

[0101] S104: Based on the target battery type, determine the target defect detection model corresponding to the target battery type from the preset defect detection model library.

[0102] In this embodiment, to improve the accuracy of defect detection, a defect detection model library is pre-configured for storing multiple defect detection models. After the battery detection model identifies the target battery type of the battery to be detected, it can specifically match a suitable target defect detection model to perform defect detection on the battery to be detected.

[0103] For example, if manufacturing line A is used to produce battery A, then after the battery detection model identifies that the battery type of battery A is A, it can match the target defect detection model A in the defect detection model library to perform defect detection on the appearance of the negative electrode of battery A; and if manufacturing line B is used to produce battery B, then after the battery detection model identifies that the battery type of battery B is B, it can match the target defect detection model B in the defect detection model library to perform defect detection on the appearance of the negative electrode of battery B.

[0104] S105: Perform feature fusion on multidimensional physical detection data and the image to be detected to obtain a comprehensive feature vector.

[0105] In this embodiment, after extracting features from the multidimensional physical detection data and the image to be detected, feature information in two dimensions can be obtained. Then, by fusing the feature information in the two dimensions, a comprehensive feature vector with stronger feature representation can be obtained.

[0106] In this embodiment, the image to be inspected can provide image-level information, while multidimensional physical inspection data can provide more accurate quantitative indicators such as size and weight. By fusing these two types of data, their advantages can be fully utilized to improve the comprehensiveness and accuracy of negative electrode appearance defect detection.

[0107] S106: Input the comprehensive feature vector into the target defect detection model and output the detection results for the negative electrode post of the battery to be tested.

[0108] It should be noted that the test results include the type of defect in the negative terminal and the location of the defect in the negative terminal of the battery under test. The defect type can include at least one of the following: dent, bulge, film breakage, notch, curling, scratch, and dirt.

[0109] In this embodiment, since the comprehensive feature vector can reflect the feature information of the negative electrode post in multiple dimensions, the target defect detection model can identify the comprehensive feature vector to quickly and accurately predict and judge the type and location of defects in the battery appearance.

[0110] In this embodiment, after obtaining the detection results output by the target defect detection model, defect identification information indicating the location of the defect type on the negative terminal of the battery under test can be generated; and the display device can be controlled to display the defect identification information on the image under test. Thus, by timely outputting defect identification information, inspectors can intuitively understand the defect type and location of the negative terminal of the battery under test, and promptly repair the battery, preventing defective batteries from entering the market and causing quality problems.

[0111] In this embodiment, by stitching together the appearance images of the negative electrode post at at least four preset locations, a comprehensive image reflecting the appearance of the negative electrode post can be generated. Then, using a battery detection model to identify the image, the target battery type can be accurately identified, and the corresponding target defect detection model can be determined. Furthermore, by comprehensively considering multi-dimensional physical inspection data and the image to be detected, and fusing the two types of data, a stronger comprehensive feature vector can be obtained. This allows the target defect detection model to quickly and accurately predict and judge the type and location of defects in the battery appearance based on the comprehensive feature vector. This not only improves the compatibility of appearance inspection for different battery types but also effectively improves the accuracy and efficiency of defect detection.

[0112] In one feasible implementation, S102 may specifically include the following sub-steps:

[0113] S102-1: Based on the imaging distance of the negative electrode appearance images at at least four preset orientations, perform size calibration on the negative electrode appearance images at at least four preset orientations.

[0114] In this embodiment, considering that the actual installation position of the image detection device or the placement position of the battery to be detected may be deviated and affect the image stitching effect, the imaging distance of the negative electrode appearance image will be detected simultaneously each time a negative electrode appearance image is acquired. Then, after acquiring negative electrode appearance images in at least four preset positions, uniform size calibration will be performed to calibrate the negative electrode appearance images in at least four preset positions to the same preset size, thereby improving the imaging effect of the image to be detected.

[0115] S102-2: Based on a preset pixel mapping table, the appearance images of the negative poles at least four preset orientations after size calibration are mapped onto a preset negative pole model to obtain a local image corresponding to each negative pole appearance image.

[0116] It should be noted that the pixel mapping table is used to characterize the mapping relationship between the key points of the negative pole model and the pixels of the negative pole appearance image at different preset orientations.

[0117] In this embodiment, the negative electrode post model consists of multiple key points, each of which is considered a pixel of the negative electrode post model. By mapping the negative electrode post appearance image onto the preset negative electrode post model, pixel filling of the key points can be achieved, thereby obtaining a local image corresponding to each negative electrode post appearance image. The shape and size of the negative electrode post model can be determined based on the target battery type of the battery to be tested.

[0118] In this embodiment, the negative electrode model can be divided into multiple local regions that correspond one-to-one with preset directions. For example, when the number of preset directions is four, the negative electrode model can be divided into four local regions, and each local region corresponds to the appearance image of the negative electrode in the corresponding direction.

[0119] In this embodiment, since the pixel mapping table can reflect the mapping relationship between the key points of the negative pole model and the pixels of the negative pole appearance image at different preset orientations, after obtaining the negative pole appearance images at different preset orientations, the pixels in the negative pole appearance image can be directly mapped to the local area corresponding to the negative pole model according to the pixel mapping table, thereby obtaining the local image corresponding to each negative pole appearance image.

[0120] S102-2: Stitch together the overlapping parts between local images to obtain the image to be detected.

[0121] In this embodiment, by stitching together the local images corresponding to the appearance images of each negative electrode post, the image to be tested of the negative electrode post of the battery to be tested can be generated.

[0122] In this embodiment, considering that there may be overlapping parts between local images when there are many preset orientations, the final image to be detected can be obtained by stitching the overlapping parts together.

[0123] In a practical implementation, for any key point in the overlapping part, the two pixels of the two overlapping local images corresponding to the key point can be fused to obtain the pixel value corresponding to the key point. For example, the average value of the two pixels can be determined as the pixel value corresponding to the key point. Then, the pixel value of the overlapping part is smoothed. In this way, the stitching of the overlapping parts between any two local images can be achieved.

[0124] In this embodiment, based on a preset pixel mapping table, the texture of the negative electrode post appearance image is directly mapped onto the negative electrode post model, which can effectively improve the image stitching effect and stitching efficiency. At the same time, by calibrating the size of the negative electrode post appearance image, the influence of installation errors can be avoided.

[0125] In one feasible implementation, S103 may specifically include the following sub-steps:

[0126] S103-1: Input the image to be detected into the battery detection model and output the confidence level between the image to be detected and various battery types.

[0127] In this embodiment, the battery detection model is trained based on negative electrode sample images of sample batteries of various battery types. The negative electrode sample images are generated based on the appearance images of the negative electrode posts of the sample batteries at at least four preset orientations.

[0128] In this embodiment, the battery detection model can identify the confidence level between the image to be detected and each battery type by the image to be detected. This confidence level can be understood as the similarity between the battery to be detected and sample batteries of multiple battery types. The higher the confidence level, the higher the similarity between the battery to be detected and the sample batteries.

[0129] S103-2: Determine the maximum value among the confidence scores between the image to be detected and multiple battery types as the target confidence score.

[0130] In this embodiment, after obtaining the confidence levels corresponding to each of the various battery types, the maximum value among the multiple confidence levels is selected as the target confidence level.

[0131] S103-3: If the target confidence level is greater than the confidence level threshold, the battery type corresponding to the target confidence level shall be determined as the target battery type.

[0132] In this embodiment, to ensure detection accuracy and avoid false detections, the battery type corresponding to the target confidence level can be determined as the target battery type when the target confidence level is greater than the confidence level threshold.

[0133] It should be noted that if the target confidence level is greater than or equal to the confidence level threshold, a prompt message can be output. For example, the prompt message may include information such as "The battery under test cannot be tested, please check if the battery under test is placed correctly".

[0134] In this embodiment, the accurate identification of battery type through a battery detection model can ensure the accuracy of subsequent defect detection.

[0135] In one feasible implementation, S105 may specifically include the following sub-steps:

[0136] S105-1: Perform feature extraction on the multidimensional physical detection data and the image to be detected respectively to obtain physical feature vectors and visual feature vectors.

[0137] In this embodiment, the physical feature vector represents the feature representation of the battery to be detected in the physical feature dimension; the visual feature vector represents the feature representation of the battery to be detected in the visual feature dimension.

[0138] In practical implementation, since the physical parameters of multidimensional physical detection data are different, when extracting features from multidimensional physical detection data, filtering and normalization processes can be performed first to remove outliers and noise and to standardize the data. Then, for any physical detection data, the deviation value between the physical detection data and its corresponding standard data is calculated. Based on the deviation value corresponding to the multidimensional physical detection data, the physical feature vector corresponding to the multidimensional physical detection data is extracted.

[0139] S105-2: Perform feature fusion on the physical feature vector and the visual feature vector to obtain a comprehensive feature vector.

[0140] In this embodiment, by fusing physical feature vectors and visual feature vectors, a stronger feature representation can be generated, thereby making full use of the advantages of the two data sources and improving detection performance through feature-level interaction.

[0141] In specific implementation, to further improve the detection effect, S105-2 may include the following sub-steps:

[0142] S105-2-1: Obtain the first weight for the physical feature vector and the second weight for the visual feature vector.

[0143] In this embodiment, corresponding weight parameters can be assigned to the physical feature vector and the visual feature vector according to the detection requirements and the battery's manufacturing process requirements. The sum of the first weight and the second weight is 1; for example, the first weight can be set to 0.2, and the visual feature vector to 0.8. It should be noted that if the manufacturing process parameters for the appearance of the battery's negative electrode are critical, the first weight can be increased.

[0144] S105-2-2: Based on the physical feature vector and the first weight, a first weighted vector is obtained; and based on the visual feature vector and the second weight, a second weighted vector is obtained.

[0145] In this embodiment, by performing weighted operations on the physical feature vector and the visual feature vector respectively, the corresponding weighted vectors can be obtained.

[0146] In this embodiment, considering that a single type of image cannot accurately and comprehensively reflect the various types of defects present in the negative electrode post, the images to be detected include visible light images and ultrasonic images; wherein, the visible light images are generated based on the appearance images of the negative electrode post at least in four preset positions collected by a visible light sensor, and the ultrasonic images are generated based on the appearance images of the negative electrode post at least in four preset positions collected by an ultrasonic sensor.

[0147] In this embodiment, based on the visible light image and the ultrasonic image, the step of extracting features from the image to be detected in S105-1 to obtain a visual feature vector may specifically include the following sub-steps: extracting features from the visible light image and the ultrasonic image respectively to obtain a visible light feature vector and an ultrasonic feature vector.

[0148] In this embodiment, the step of obtaining the second weighted vector based on the visual feature vector and the second weight in S105-2-2 may specifically include the following sub-steps: determining a third weight for the visible light feature vector and a fourth weight for the ultrasonic feature vector based on the second weight and a preset weight ratio; obtaining the second weighted vector based on the visible light feature vector and the third weight, and the ultrasonic feature vector and the fourth weight. The sum of the third weight and the fourth weight constitutes the second weight. That is, according to the contribution levels of the visible light image and the ultrasonic image, the second weight can be further allocated to the visible light feature vector and the ultrasonic feature vector.

[0149] It should be noted that by using the visible light feature vector extracted from the visible light image, defects such as damage, cracks, uneven color, and foreign matter in the negative electrode post can be detected more accurately; by using the ultrasonic feature vector extracted from the ultrasonic image, defects such as dents, bulging, delamination, cracks, and electrode welding quality in the negative electrode post can be detected more accurately.

[0150] In this embodiment, compared with the traditional detection scheme that only considers visible light images, by comprehensively considering both visible light images and ultrasonic images, it is possible to accurately identify various defects, effectively improving the comprehensiveness and accuracy of appearance inspection.

[0151] S105-2-3: Based on the first weighted vector and the second weighted vector, the comprehensive feature vector is obtained.

[0152] In this embodiment, by fusing feature information from these two dimensions, a comprehensive feature vector with stronger feature representation can be obtained, thereby leveraging their advantages to improve the comprehensiveness and accuracy of negative electrode appearance defect detection.

[0153] In a feasible implementation, any defect detection model in the detection model library can be obtained through the following training method:

[0154] S201: Based on the defect detection model and the model comparison table, determine the battery type corresponding to the defect detection model.

[0155] In this embodiment, since different defect detection models are used to detect the appearance of batteries of different battery types, during the model training phase, the battery type corresponding to the defect detection model to be trained is determined according to a preset model comparison table, and then the sample data of the battery type is used for targeted training.

[0156] S202: Determine the target battery training sample corresponding to the battery type from various battery sample data in the sample database.

[0157] In this embodiment, multiple battery sample data are used to train defect detection models corresponding to multiple battery types. That is, different battery sample data are used to train different defect detection models. For example, if the sample database contains three types of battery sample data, A, B, and C, then the three types of battery sample data can be used to train defect detection model A for detecting battery type A, defect detection model B for detecting battery type B, and defect detection model C for detecting battery type C, respectively.

[0158] In the specific implementation, each battery training sample in the sample database has a corresponding battery type identifier. After determining the battery type corresponding to the defect detection model, the target battery training sample belonging to that battery type can be indexed in the sample database according to the battery type identifier.

[0159] In this embodiment, battery training samples can be obtained by combining negative electrode sample images and multidimensional physical detection sample data. If there are N negative electrode sample images and M multidimensional physical detection sample data, then after combination, a maximum of N×M battery training samples can be obtained. The negative electrode sample images are generated based on the appearance images of the negative electrode posts of the sample battery at at least four preset orientations.

[0160] S203: Based on the target battery training samples, train the preset initial model until the initial model meets the training cutoff condition, and obtain the trained defect detection model.

[0161] In this embodiment, the initial model is iteratively trained using target battery training samples, which continuously optimizes the network parameters of the initial model. In specific implementation, after each training round, a loss value is calculated using a preset loss function, and then backpropagation is performed based on this loss value to update the model parameters until the model to be trained meets the training cutoff condition. The training cutoff condition can be either reaching a threshold number of training rounds or the loss function reaching a stable state.

[0162] In this embodiment, for any battery type, by repeatedly using the above training method, multiple defect detection models corresponding to each battery type can be obtained. Then, the multiple defect detection models obtained from the training are deployed to the defect detection model library. When performing actual defect detection, the target defect detection model corresponding to the target battery type can be matched in the defect detection model library to perform appearance defect detection based on the target battery type of the battery to be detected.

[0163] Secondly, based on the same inventive concept, and referring to... Figure 2 This application illustrates a deep learning-based battery negative electrode appearance inspection device 200 according to an embodiment of the present application. The device includes:

[0164] The data acquisition module 201 is used to acquire multidimensional physical detection data of the battery under test and negative electrode appearance images of the negative electrode in at least four preset positions.

[0165] Image generation module 202 is used to generate an image to be detected based on the appearance images of the negative pole at at least four preset orientations;

[0166] The type determination module 203 is used to input the image to be detected into the battery detection model and determine the target battery type of the battery to be detected.

[0167] The model determination module 204 is used to determine the target defect detection model corresponding to the target battery type from a preset defect detection model library based on the target battery type.

[0168] The feature determination module 205 is used to perform feature fusion on multi-dimensional physical detection data and the image to be detected to obtain a comprehensive feature vector.

[0169] The defect detection module 206 is used to input the comprehensive feature vector into the target defect detection model and output the detection results for the negative terminal of the battery to be tested; the detection results include the defect type of the negative terminal and the location of the defect type in the negative terminal of the battery to be tested.

[0170] In one embodiment of this application, the image generation module 202 includes:

[0171] The image calibration submodule is used to perform size calibration on the negative electrode appearance images at least four preset orientations based on the imaging distance of the negative electrode appearance images at at least four preset orientations.

[0172] The pixel mapping submodule is used to map the appearance images of the negative pole at least four preset orientations after size calibration onto the preset negative pole model based on a preset pixel mapping table, so as to obtain the local image corresponding to each negative pole appearance image; wherein, the pixel mapping table is used to characterize the mapping relationship between the key points of the negative pole model and the pixels of the negative pole appearance images at different preset orientations.

[0173] The image stitching submodule is used to stitch together overlapping parts of local images to obtain the image to be detected.

[0174] In one embodiment of this application, the type determination module 203 includes:

[0175] The first confidence determination submodule is used to input the image to be detected into the battery detection model and output the confidence between the image to be detected and multiple battery types; the battery detection model is trained based on negative electrode sample images of sample batteries of multiple battery types;

[0176] The second confidence determination submodule is used to determine the maximum value among the confidence scores between the image to be detected and multiple battery types as the target confidence score;

[0177] The battery type determination submodule is used to determine the battery type corresponding to the target confidence level as the target battery type of the model determination module when the target confidence level is greater than the confidence level threshold.

[0178] In one embodiment of this application, the model determination module 204 includes:

[0179] The lookup table acquisition submodule is used to acquire a preset model lookup table; wherein, the model lookup table represents different defect detection models corresponding to different battery types;

[0180] The detection model determination submodule is used to determine the target fault detection model from multiple defect detection models in the defect detection model library based on the target battery type and model lookup table.

[0181] In one embodiment of this application, the feature determination module 205 includes:

[0182] The feature extraction submodule is used to extract features from the multidimensional physical detection data and the image to be detected, respectively, to obtain physical feature vectors and visual feature vectors.

[0183] The feature fusion submodule is used to fuse physical feature vectors and visual feature vectors to obtain a comprehensive feature vector.

[0184] In one embodiment of this application, the feature fusion submodule includes:

[0185] The weight acquisition unit is used to acquire the first weight for the physical feature vector and the second weight for the visual feature vector.

[0186] The vector weighting unit is used to obtain a first weighted vector based on the physical feature vector and the first weight; and to obtain a second weighted vector based on the visual feature vector and the second weight.

[0187] The feature fusion unit is used to obtain a comprehensive feature vector based on the first weighted vector and the second weighted vector.

[0188] In one embodiment of this application, the image to be detected includes a visible light image and an ultrasonic image; wherein, the visible light image is generated based on the appearance images of the negative electrode post at at least four preset positions collected by a visible light sensor, and the ultrasonic image is generated based on the appearance images of the negative electrode post at at least four preset positions collected by an ultrasonic sensor.

[0189] The feature extraction submodule includes:

[0190] The feature extraction unit is used to extract features from the visible light image and the ultrasonic image respectively, to obtain the visible light feature vector and the ultrasonic feature vector;

[0191] The vector weighting unit includes:

[0192] The weight determination subunit is used to determine the third weight for the visible light feature vector and the fourth weight for the ultrasonic feature vector based on the second weight and a preset weight ratio.

[0193] The vector weighting subunit is used to obtain the second weighting vector based on the visible light feature vector and the third weight, as well as the ultrasonic feature vector and the fourth weight.

[0194] In one embodiment of this application, the deep learning-based battery negative electrode appearance inspection device 200 further includes:

[0195] The type determination module is used to determine the battery type corresponding to the defect detection model based on the defect detection model and the model lookup table;

[0196] The sample determination module is used to determine the target battery training sample corresponding to the battery type from the various battery sample data in the sample database; wherein, the various battery sample data are used to train the defect detection model corresponding to the various battery types;

[0197] The model training module is used to train a preset initial model based on the target battery training samples until the initial model meets the training cutoff condition, thus obtaining a trained defect detection model.

[0198] In one embodiment of this application, the deep learning-based battery negative electrode appearance inspection device 200 further includes:

[0199] The information generation module is used to generate defect identification information that indicates the location of the defect type at the negative terminal of the battery under test.

[0200] The information display module is used to control the display device to display defect identification information on the image to be inspected.

[0201] It should be noted that the specific implementation of the deep learning-based battery negative electrode appearance detection device 200 in this application embodiment refers to the specific implementation of the deep learning-based battery negative electrode appearance detection method proposed in the first aspect of the above-mentioned embodiment, and will not be repeated here.

[0202] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable system. The computer program or instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, a computer program or instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0203] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0204] In this application, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0205] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0206] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0207] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, units, and processes described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0208] In the embodiments provided in this application, it should be understood that the disclosed systems, methods, and approaches can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between systems or units may be electrical, mechanical, or other forms.

[0209] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0210] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0211] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0212] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for detecting the appearance of a battery negative electrode based on deep learning, characterized in that, The method includes: Acquire multidimensional physical detection data of the battery under test and appearance images of the negative terminal of the battery under test in at least four preset positions; Based on the appearance images of the negative electrode post at least four preset locations, generate the image to be detected; The image to be detected is input into the battery detection model to determine the target battery type of the battery to be detected. Based on the target battery type, a target defect detection model corresponding to the target battery type is determined from a preset defect detection model library; The multidimensional physical detection data and the image to be detected are fused to obtain a comprehensive feature vector; The comprehensive feature vector is input into the target defect detection model, and the detection result for the negative terminal of the battery under test is output; the detection result includes the defect type of the negative terminal and the location of the defect type in the negative terminal of the battery under test; The step of fusing features between the multidimensional physical detection data and the image to be detected to obtain a comprehensive feature vector includes: Feature extraction is performed on the multidimensional physical detection data and the image to be detected to obtain physical feature vectors and visual feature vectors; The physical feature vector and the visual feature vector are fused to obtain the comprehensive feature vector. The step of generating an image to be detected based on the appearance images of the negative electrode post at at least four preset orientations includes: Based on the imaging distance of the negative electrode post appearance images at at least four preset orientations, the size of the negative electrode post appearance images at the at least four preset orientations is calibrated. Based on a preset pixel mapping table, the negative pole appearance images at at least four preset orientations after size calibration are mapped onto a preset negative pole model to obtain a local image corresponding to each negative pole appearance image; wherein, the pixel mapping table is used to characterize the mapping relationship between the key points of the negative pole model and the pixels of the negative pole appearance images at different preset orientations. The overlapping portions of the local images are stitched together to obtain the image to be detected; The step of inputting the image to be detected into the battery detection model to determine the target battery type includes: The image to be detected is input into the battery detection model, and the output is the confidence score between the image to be detected and multiple battery types. The confidence score is used to characterize the similarity between the battery to be detected and the sample battery. The battery detection model is trained based on the negative electrode sample images of sample batteries of multiple battery types. The maximum value among the confidence scores between the image to be detected and various battery types is determined as the target confidence score; If the target confidence level is greater than the confidence level threshold, the battery type corresponding to the target confidence level is determined as the target battery type.

2. The method for detecting the appearance of a battery negative electrode based on deep learning according to claim 1, characterized in that, Based on the target battery type, the step of determining the target defect detection model corresponding to the target battery type from a preset defect detection model library includes: Obtain a preset model lookup table; wherein the model lookup table represents different defect detection models corresponding to different battery types; Based on the target battery type and the model comparison table, the target defect detection model is determined from among multiple defect detection models in the defect detection model library.

3. The method for detecting the appearance of a battery negative electrode based on deep learning according to claim 1, characterized in that, The step of fusing the physical feature vector and the visual feature vector to obtain the comprehensive feature vector includes: Obtain a first weight for the physical feature vector and a second weight for the visual feature vector; Based on the physical feature vector and the first weight, a first weighted vector is obtained; and based on the visual feature vector and the second weight, a second weighted vector is obtained. The comprehensive feature vector is obtained based on the first weighted vector and the second weighted vector.

4. The method for detecting the appearance of a battery negative electrode based on deep learning according to claim 3, characterized in that, The image to be detected includes a visible light image and an ultrasonic image; wherein, the visible light image is generated based on the appearance images of the negative electrode posts at at least four preset positions acquired by the visible light sensor, and the ultrasonic image is generated based on the appearance images of the negative electrode posts at at least four preset positions acquired by the ultrasonic sensor. The step of extracting features from the image to be detected to obtain a visual feature vector includes: Feature extraction is performed on the visible light image and the ultrasonic image respectively to obtain visible light feature vector and ultrasonic feature vector; The step of obtaining the second weighted vector based on the visual feature vector and the second weight includes: Based on the second weight and the preset weight ratio, a third weight for the visible light feature vector and a fourth weight for the ultrasonic feature vector are determined. The second weighted vector is obtained based on the visible light feature vector and the third weight, and the ultrasonic feature vector and the fourth weight.

5. The method for detecting the appearance of a battery negative electrode based on deep learning according to claim 2, characterized in that, Any defect detection model in the detection model library is obtained through the following training method: Based on the defect detection model and the model lookup table, the battery type corresponding to the defect detection model is determined; From a variety of battery sample data in a sample database, target battery training samples corresponding to the battery type are determined; wherein, the variety of battery sample data is used to train defect detection models corresponding to the various battery types; Based on the target battery training samples, a preset initial model is trained until the initial model meets the training cutoff condition, thus obtaining the trained defect detection model.

6. The method for detecting the appearance of a battery negative electrode based on deep learning according to claim 1, characterized in that, After inputting the comprehensive feature vector into the target defect detection model and outputting the detection result for the negative electrode post of the battery to be tested, the method further includes: Generate defect identification information to indicate the location of the defect type at the negative terminal of the battery under test; The control display device displays the defect identification information on the image to be inspected.

7. A battery negative electrode appearance inspection device based on deep learning, characterized in that, The apparatus for implementing the method according to any one of claims 1-6, the apparatus comprising: The data acquisition module is used to acquire multidimensional physical detection data of the battery under test and the appearance images of the negative terminal of the battery under test in at least four preset positions. The image generation module is used to generate an image to be detected based on the appearance images of the negative electrode posts at the at least four preset orientations; The type determination module is used to input the image to be detected into the battery detection model to determine the target battery type of the battery to be detected; The model determination module is used to determine the target defect detection model corresponding to the target battery type from a preset defect detection model library based on the target battery type. The feature determination module is used to perform feature fusion on the multidimensional physical detection data and the image to be detected to obtain a comprehensive feature vector. The defect detection module is used to input the comprehensive feature vector into the target defect detection model and output the detection result for the negative terminal of the battery to be tested; the detection result includes the defect type of the negative terminal and the location of the defect type in the negative terminal of the battery to be tested.

Images