A method, system and storage medium for detecting battery surface defects
Through segmentation of image data on the battery surface, imaging angle correction and feature extraction, combined with defect hash value processing, efficient identification of small defects on the battery electrode surface is achieved, and the problems of low detection efficiency and high cost in the prior art are solved, and the problems of high detection efficiency and high cost are improved. Battery manufacturing quality and efficiency.
Patent Information
- Application Number
- CN202410970104.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-19
- Publication Date
- 2024-11-05
- Estimated Expiration
- 2044-07-19
AI Technical Summary
The prior art is difficult to efficiently identify defects such as tiny scratches and stains on the surface of the battery electrode, resulting in a degradation of battery performance and low efficiency and high cost of detection methods.
By collecting battery surface image data, battery position segmentation and imaging angle correction are performed, feature sequences are extracted and defect pixel classification is performed, defect hash value is calculated to eliminate foreign object interference, defect position distribution heat map is constructed, and risk assessment report is output.
It improves the accuracy and efficiency of battery surface defect detection, reduces visual deviation caused by different shooting angles, enhances the recognition accuracy and detection reliability of small defects, and improves the quality and efficiency of battery manufacturing.
Smart Images

Figure CN118518677B_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of image detection technology, and in particular, relates to a method, system and storage medium for realizing battery surface defect detection. Background Art
[0002] The performance of batteries is often affected by various defects generated during the manufacturing process, which may cause battery performance degradation or even failure. Common battery defects include scratches, stains, bubbles, cracks, etc. on the electrode surface. These defects may cause local overheating, overcurrent or uneven chemical reactions during the battery's charge and discharge process, thereby reducing the efficiency and reliability of the battery. For example, scratches on the electrode surface may penetrate the active material layer and cause internal short circuits; stains may contain impurities and affect the electrochemical performance of the battery; bubbles and cracks may cause structural failure during the battery's charge and discharge process. At present, the methods for detecting these defects mainly include manual visual inspection, X-ray imaging, ultrasonic testing, etc. However, manual inspection is inefficient and prone to errors; X-ray imaging and ultrasonic testing equipment are expensive and have limited ability to identify tiny surface defects. Summary of the invention
[0003] In response to the above-mentioned defects of the prior art, the present application provides a method, system and storage medium for battery surface defect detection, which improves the quality and efficiency of battery manufacturing by efficiently and effectively identifying defects such as tiny scratches and stains on the surface of battery electrodes.
[0004] To achieve the above objectives, the present application provides a method for detecting battery surface defects, the method comprising:
[0005] S100: starting a process to be detected, and collecting battery surface image data according to the process to be detected.
[0006] S200: performing battery position segmentation on the battery surface image data to extract the battery surface area; performing imaging angle correction on the battery surface area image to obtain a battery position feature image.
[0007] S300: extracting features from the battery position feature image to construct a feature sequence, classifying defective pixels in the feature sequence according to a preset defect probability scoring rule, and outputting a pixel distribution map that meets a preset probability threshold.
[0008] S400: After calculating the defect hash value according to the pixel distribution map, eliminate foreign matter interference pixels according to the defect hash value to obtain a pixel distribution map of the defect area.
[0009] Wherein, in the step S100, the process to be inspected is generated by entering the battery information of the current batch; wherein, the process to be inspected includes a plurality of inspection items, and the inspection items are configured by the battery information of the current batch; the battery surface image data includes a plurality of frames of upper surface data and lower surface data of any battery.
[0010] In the step S200, the battery position segmentation is performed on the battery surface image data, specifically:
[0011] The battery edge contour parameters are collected, and the battery surface image data is divided into a target area and a background area according to the battery edge contour parameters; the edge contour of the target area is optimized by using pixel features of a preset adjacent range of the battery edge to obtain the battery surface area.
[0012] The imaging angle correction of the battery surface area map is specifically performed as follows:
[0013] The preset standard template area is compared with the battery surface area to obtain an imaging angle difference; the battery surface area is spatially projected and transformed according to the imaging angle difference to correct the imaging angle and obtain the battery position characteristic image.
[0014] In the step S300, the battery position feature image includes a first image output and displayed in RGB format; the pixel distribution map includes a first pixel distribution map; all pixels in the first image are extracted and converted into a first preset dimension feature map, and a first feature extraction is performed through a first preset defect label; the output display form of the extracted pixel feature information is converted from the first preset dimension to the original dimension, and flattened into a first feature sequence.
[0015] The preset defect probability scoring rules are specifically:
[0016] A defect probability score for each pixel in a first feature sequence is calculated according to an activation function and the first preset defect label, and a first probability score map is generated according to the defect probability score; first defective pixels are classified according to the first probability score map and a preset probability threshold; the preset probability threshold includes several first preset probability thresholds corresponding to different defect types.
[0017] When the score of any pixel in the first probability score map exceeds any of the first preset probability thresholds, the pixel output is defined as a first target defect type corresponding to the first preset probability threshold; wherein the first target defect type includes a first target feature code; and the first pixel distribution map is output according to the first target feature code of each pixel.
[0018] In the step S300, the battery position characteristic image also includes a second image outputted and displayed in grayscale form; the pixel distribution map also includes a second pixel distribution map; all pixels in the second image are extracted and converted into a second preset dimension feature map, and a second feature extraction is performed through a second preset defect label; the output display form of the extracted pixel feature information is converted from the second preset dimension to the original dimension, and flattened into a second feature sequence.
[0019] The preset defect probability scoring rules are specifically:
[0020] A defect probability score for each pixel in a second feature sequence is calculated based on the activation function and the second preset defect label, and a second probability score map is generated based on the defect probability score; second defective pixels are classified based on the second probability score map and a preset probability threshold; the preset probability threshold also includes several second preset probability thresholds corresponding to different defect types.
[0021] When the score of any pixel in the second probability score map exceeds any of the second preset probability thresholds, the pixel output is defined as a second target defect type corresponding to the second preset probability threshold; wherein the second target defect type includes a second target feature code; and the second pixel distribution map is output according to the second target feature code of each pixel.
[0022] The step S400 is specifically as follows:
[0023] The defect hash value includes a first defect hash value and a second defect hash value; a first defect hash value is calculated by performing a first hash coding value on the first target feature code in the first pixel distribution map; a second defect hash value is calculated by performing a second hash coding value on the second target feature code in the second pixel distribution map.
[0024] The first defect hash value is added to the second defect hash value to obtain a third defect hash value, and a third pixel distribution map is generated; when it is detected that the third defect hash value meets the preset hash threshold, the foreign matter interference pixels are eliminated from the third pixel distribution map to obtain a defect area pixel distribution map.
[0025] In the present application, the method further comprises:
[0026] S500: constructing a defect position distribution heat map according to the defect area pixel distribution map, and outputting a risk assessment report according to the defect position distribution heat map.
[0027] Among them, according to the pixel distribution map of the defect area, the preset areas and defect types are divided into groups; the number of defect occurrences and the defect rate of any of the groups are calculated, the calculation results are output, and a defect position distribution heat map is constructed based on the calculation results; the defect position distribution heat map is compared with a preset standard battery process data map, and the risk assessment report is output through a preset defect degree threshold.
[0028] To achieve the above objectives, the present application also provides a battery surface defect detection system, the system comprising:
[0029] Image detection starting unit, image preprocessing unit, image defect detection processing unit and image defect optimization processing unit.
[0030] The image detection starting unit is used to start the process to be detected and collect battery surface image data according to the process to be detected.
[0031] The image preprocessing unit is used to perform battery position segmentation on the battery surface image data to extract the battery surface area; and obtain a battery position feature image after performing imaging angle correction on the battery surface area image.
[0032] The image defect detection processing unit is used to extract features from the battery position feature image and construct it into a feature sequence, and classify defective pixels in the feature sequence according to a preset defect probability scoring rule, and output a pixel distribution map that meets a preset probability threshold.
[0033] The image defect optimization processing unit is used to calculate the defect hash value according to the pixel distribution map, and then eliminate the foreign matter interference pixels according to the defect hash value to obtain the defect area pixel distribution map.
[0034] Furthermore, the image defect position prediction and analysis unit is used to construct a defect position distribution heat map according to the defect area pixel distribution map, and output a risk assessment report according to the defect position distribution heat map.
[0035] In the present application, the system further includes: an image defect position prediction and analysis unit and an abnormality alarm unit.
[0036] The image defect position prediction and analysis unit is used to construct a defect position distribution heat map according to the defect area pixel distribution map, and output a risk assessment report according to the defect position distribution heat map.
[0037] The abnormal alarm unit is used to output the current battery surface defect position coordinates and damage degree information to the management personnel when the risk assessment report triggers the abnormal alarm condition.
[0038] To achieve the above objectives, the present application also provides a storage medium, which is a type of computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements any of the battery surface defect detection methods described above.
[0039] Compared with the prior art, the present invention has the following beneficial effects:
[0040] The present application proposes a method, system and storage medium for realizing battery surface defect detection, which improves the accuracy of battery surface defect detection by segmenting the battery position and correcting the imaging angle of the battery surface image data, ensures that image processing and analysis are performed under a unified standard, and reduces the visual deviation caused by different shooting angles; the present application improves the recognition accuracy of battery surface defect detection by extracting features from the battery position feature image and constructing it into a feature sequence, and then classifying defective pixels according to preset defect probability scoring rules; the present application also eliminates foreign body interference pixels by calculating defect hash values, and can obtain a more accurate defect area pixel distribution map from the defect area pixel distribution map, thereby improving the reliability of defect detection; at the same time, the present application also improves the quality and efficiency of subsequent battery manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 This is a flow chart of a method for implementing battery surface defect detection in one embodiment of the present application.
[0042] Figure 2 It is a schematic diagram of segmentation in an embodiment of the present application.
[0043] Figure 3 Schematic diagram of angle correction in one embodiment of the present application.
[0044] Figure 4 FIG. 1 is a schematic diagram of defective pixel distribution of a visible light image in an embodiment of the present application.
[0045] Figure 5 Schematic diagram of defective pixel distribution of a grayscale image in one embodiment of the present application.
[0046] Figure 6 This is a partial schematic diagram of the thermal map of the defect location distribution on the front side of a battery in one embodiment of the present application.
[0047] Figure 7 Schematic diagram of a battery surface defect detection system in one embodiment of the present application.
[0048] Figure 8 This is a schematic diagram of a system interface displaying an analysis risk assessment report in one embodiment of the present application. DETAILED DESCRIPTION
[0049] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme will be clearly and completely described below in conjunction with the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application. Embodiment 1:
[0050] As attached Figure 1 As shown, in order to solve the above technical problems, the present application provides a method for realizing battery surface defect detection. It should be noted that the embodiment of the present application can collect image data of the battery surface in real time during the battery production process, and accurately analyze the tiny defects that may exist on the battery pole pieces, such as scratches, stains, electrode misalignment, etc.
[0051] In this embodiment, the method includes:
[0052] S100: starting a process to be detected, and collecting battery surface image data according to the process to be detected.
[0053] It should be noted that in step S100, the process to be inspected is generated by inputting the battery information of the current batch, such as the battery model, batch number, production date, production line information, production personnel, and the area of the upper and lower surfaces of the battery, etc., but not limited thereto. This information will be used to start the corresponding inspection process, that is, to determine which specific inspections need to be performed on the battery surface image data based on this information.
[0054] Among them, the process to be inspected includes several inspection items, and the inspection items are configured according to the battery information of the current batch; the battery surface image data includes several frames of upper surface data and lower surface data of any battery.
[0055] It should be noted that the inspection items may include: defect type, defect length, width, area, coordinate location, distribution area and other information. At the same time, users can fill in different search items in the data center page, enter the product's SN code, select date batch number, production line name, model name, severity, front and back, defect results, rule name, defect type, inspection area and other related search boxes to enter information to realize the function of quickly querying the inspection results.
[0056] In this embodiment, the upper surface data and lower surface data of the battery refer to a pair of upper and lower surfaces with the largest surface area of the battery pack. For example, assuming that the acquisition frequency of the upper and lower surface images of the battery pack by the external AOI acquisition device is set to 1000 Hz respectively, the sensor will collect 1000 frames of image data per second respectively, then the battery surface image data includes 1000 frames of image data of the upper and lower surfaces respectively, and is not limited thereto.
[0057] S200: performing battery position segmentation on the battery surface image data to extract the battery surface area; performing imaging angle correction on the battery surface area image to obtain a battery position feature image.
[0058] It should be noted that after obtaining the battery surface image data, because the original image contains complex factors such as uneven lighting, messy colors and disordered textures, it often causes the detection model to be distracted, thereby reducing the recognition accuracy of battery surface defects. When processing these images later, the deep learning network training model needs to distinguish which features are battery defects and which are background interference, which increases the difficulty of model training and may also lead to misjudgment or missed detection during the detection process.
[0059] Therefore, the embodiment of the present application accurately segments the battery image before model training and detection, thereby separating the battery surface from the background, ensuring that the model focuses on the characteristics of the battery itself rather than background information, thereby improving the accuracy of subsequent defect detection.
[0060] Wherein, in the step S200, the battery position segmentation of the battery surface image data is specifically performed as follows:
[0061] Collect the battery edge contour parameters, and divide the battery surface image data into a target area and a background area according to the battery edge contour parameters; optimize the edge contour of the target area by using the pixel features of the preset adjacent range of the battery edge to obtain the battery surface area. The segmentation diagram is shown in the attached figure. Figure 2 shown.
[0062] Preferably, in this embodiment, the edge contour in the battery image can be identified by using image processing technology, such as edge detection algorithm, or Canny edge detection and Sobel edge detection, etc. These algorithms can detect points with significant brightness changes in the image, thereby finding the edge of the battery.
[0063] According to the collected battery edge contour parameters, the image is divided into two parts: an area containing the battery, i.e., the target area, and an area without the battery, i.e., the background area. Among them, a threshold can be set to determine which pixels belong to the battery area and which belong to the background area, but it is not limited to this.
[0064] In this embodiment, in order to improve the accuracy of segmentation, the edge contour of the target area can be optimized. This can be achieved by analyzing the pixel features near the edge contour, for example, checking the features such as pixel color, brightness or texture around the edge contour to determine the exact position of the edge.
[0065] Through the above steps, an accurate battery surface area can be obtained, which contains complete information about the battery, while the background area is excluded. By optimizing the edge contour of the battery, this embodiment can more accurately separate the battery area from the background and reduce the occurrence of mis-segmentation. Excluding the background area can reduce the impact of background noise on subsequent image processing and defect detection, and improve the accuracy of detection. Accurate battery area segmentation helps in subsequent image processing and analysis, reduces unnecessary consumption of computing resources, and improves detection efficiency.
[0066] In this embodiment, the imaging angle correction of the battery surface area image is specifically performed as follows:
[0067] The preset standard template area is compared with the battery surface area to obtain an imaging angle difference; the battery surface area is spatially projected and transformed according to the imaging angle difference to correct the imaging angle and obtain the battery position characteristic image.
[0068] It should be noted that the battery imaging angle correction can uniformly eliminate the visual deviation caused by different shooting angles and ensure that all battery images are evaluated under the same standards, which is crucial for subsequent image processing and analysis. Secondly, angle correction helps to reduce deformation and perspective errors, so that features on the battery surface such as defects, scratches or bumps can be more accurately identified and located. In addition, a unified imaging angle can simplify the subsequent training process of the deep learning model, and the model no longer needs to adapt to multiple angle changes, so it converges faster and improves generalization capabilities. Finally, standardized viewing angles can also improve the consistency and repeatability of the entire automatic inspection system, reduce the variability of inspection results, and thus improve the overall efficiency of battery quality control on the production line. In this embodiment, the angle correction schematic is shown in the attached figure. Figure 3 shown.
[0069] Preferably, in this embodiment, a standardized battery image template may be defined first, which represents the imaging angle and position of the battery in an ideal state. The template area may be a standard view of the battery image, such as a front view or a side view of the battery.
[0070] The actual image of the battery is compared to a preset standard template area. This can be achieved through image registration techniques such as feature-based registration or template-based registration. The registration process aims to find the best alignment between the two images.
[0071] Through the comparison process, the angle difference between the actual battery image and the standard template can be calculated. This difference represents the rotation angle and position offset of the actual image relative to the standard template.
[0072] Next, based on the calculated imaging angle difference, the actual battery surface area image is spatially transformed in coordinate projection. This can be achieved through image processing techniques, such as rotation matrix transformation, perspective transformation or affine transformation. The transformed image will have an imaging angle consistent with the standard template.
[0073] Finally, the image after coordinate projection transformation is the corrected battery position feature image. This image eliminates the visual deviation caused by different shooting angles and ensures that all battery images are evaluated under the same standard.
[0074] S300: extracting features from the battery position feature image to construct a feature sequence, classifying defective pixels in the feature sequence according to a preset defect probability scoring rule, and outputting a pixel distribution map that meets a preset probability threshold.
[0075] It should be noted that the battery position feature images that have been preprocessed in the above steps, including the battery visible light image and the battery grayscale image, are grouped in two sides, together with the image defect labels, and the visible light defect segmentation operation and grayscale image segmentation operation are respectively performed, and finally the whole process of modeling a set of battery defect segmentation models is completed, which is used for defect reasoning of the battery visible light image and the corresponding battery grayscale image. The output results of the two models are merged and processed, and finally the pixel distribution map with battery defects is obtained.
[0076] Preferably, during the algorithm model training process, the algorithm implementation logic for defect segmentation of the battery visible light image and the corresponding battery grayscale image can be the same.
[0077] Wherein, in the step S300, the battery position characteristic image includes a first image output and displayed in RGB format; the pixel distribution map includes a first pixel distribution map; all pixels in the first image are extracted and converted into a first preset dimension feature map, and a first feature extraction is performed through a first preset defect label; the output display form of the extracted pixel feature information is converted from the first preset dimension to the original dimension, and flattened into a first feature sequence.
[0078] It should be noted that, in this embodiment, the preprocessed visible light battery image can be input into the defect segmentation algorithm. First, all pixels of the first image in RGB form are extracted into a smaller-sized high-dimensional feature map through the feature extraction module. The parameters of the feature extraction module are gradually optimized with the number of training times. The labeled data is used to supervise the learning of visible light battery defects: dust, dirt, scratches, L glue offset, VHB glue folding and other defects, so that the feature module can correctly learn the feature extraction capability of defects. Subsequently, the feature decoder upsamples and reconstructs the small-sized high-dimensional feature map, and the feature information of the high-dimensional feature map will be gradually released as a low-dimensional large-sized feature map, and finally regress to the same size as the original image. Preferably, the channel dimension can be finally defined as 56 dimensions, but it is not limited to this.
[0079] Finally, the reconstructed feature map is flattened into a feature sequence, such as a one-dimensional vector, and sent to a fully connected neural network for defective pixel classification, and finally a defect probability score for each pixel in the image is obtained. Preferably, the pixel distributions of the five defects of dust, dirt, scratches, L glue offset, and VHB glue folding are set to [1,1,1], [2,2,2], [3,3,3], [4,4,4], and [5,5,5]. When the model outputs a mask map, it will classify the defects at the pixel level through the probability score map, and assign the defect area that reaches the probability threshold with a classified pixel of a certain defect type from [1,1,1] to [5,5,5] to form a pixel distribution of the defect area. As shown in the attached figure Figure 4 , which is a schematic diagram of defective pixel distribution of a visible light image in this embodiment, wherein yellow represents L-glue offset, green represents stains, and red represents dust, for eliminating dust interference factors.
[0080] The preset defect probability scoring rules are specifically as follows:
[0081] A defect probability score for each pixel in a first feature sequence is calculated according to an activation function and the first preset defect label, and a first probability score map is generated according to the defect probability score; first defective pixels are classified according to the first probability score map and a preset probability threshold; the preset probability threshold includes several first preset probability thresholds corresponding to different defect types.
[0082] It should be noted that in this embodiment, the above feature sequence can be input into a fully connected neural network to classify defective pixels. The activation function in the network can be ReLU, Sigmoid or Softmax, etc., which is used to calculate the defect probability score of each pixel.
[0083] Then, based on the activation function and the preset defect label, the defect probability score of each pixel in the feature sequence is calculated. These scores are plotted as a probability score graph, showing the defect probability of each pixel in the image.
[0084] Defective pixels are classified based on the probability score map and the preset probability threshold. If the pixel score exceeds a certain threshold, it is classified as the corresponding defect type. Finally, based on the classification results, a pixel distribution map with battery defects is output. This map shows the location and type of all pixels in the image that are identified as defects.
[0085] When the score of any pixel in the first probability score map exceeds any of the first preset probability thresholds, the pixel output is defined as a first target defect type corresponding to the first preset probability threshold; wherein the first target defect type includes a first target feature code; and the first pixel distribution map is output according to the first target feature code of each pixel.
[0086] Preferably, for example, the preset probability thresholds correspond to defect types such as dust, dirt, scratches, L glue offset, VHB glue folding, etc., then when the pixel score exceeds the corresponding threshold, the pixel is classified as the corresponding defect type. The defect type code of each pixel (such as [1,1,1] corresponds to dust, [2,2,2] corresponds to dirt, etc.) will be used to output the pixel distribution map.
[0087] Preferably, the first target characteristic code may be a different color code, but is not limited thereto.
[0088] In practical applications, this embodiment involves training a deep learning model to identify and classify battery defects. The model is trained on a large amount of annotated image data to learn how to extract features from the image and predict the defect type. The trained model can be used for real-time battery defect detection, improving the automation level of the production line and the quality of battery products.
[0089] In the step S300, the battery position characteristic image also includes a second image outputted and displayed in grayscale form; the pixel distribution map also includes a second pixel distribution map; all pixels in the second image are extracted and converted into a second preset dimension feature map, and a second feature extraction is performed through a second preset defect label; the output display form of the extracted pixel feature information is converted from the second preset dimension to the original dimension, and flattened into a second feature sequence.
[0090] Preferably, in this embodiment, the second pixel distribution map can be firstly put into the feature extraction module, and all pixels of the grayscale image can be extracted into a smaller-sized high-dimensional feature map. The parameters of the feature extraction module are gradually optimized with the number of training times, and the labeled data is used to supervise the learning of grayscale battery defects: scratches, dents, crushing, corner drops, white glue bubbles, trademark wrinkles and other defects. By comparing with known defect labels, the parameters of the feature extraction module are adjusted so that it can more accurately extract features related to defects. In this process, optimization algorithms such as gradient descent are applied to adjust network parameters to minimize the difference between the model prediction results and the actual labels, so that the model learns better feature representations, so that the feature module can correctly learn the feature extraction capabilities of defects, and then the feature decoder reconstructs the small-sized high-dimensional feature map by upsampling, and the feature information of the high-dimensional feature map will be gradually released as a low-dimensional large-sized feature map, and finally regress to the same size as the original image, and finally the reconstructed feature map is flattened into a feature sequence.
[0091] It should be noted that the present embodiment first extracts the pixels of the grayscale battery image as a high-dimensional feature map. This process can be the same as the processing process of the visible light image, but it targets different defect types such as scratches, dents, etc. The parameters of the feature extraction module are also continuously optimized through training to improve the learning ability of grayscale defect features. The grayscale battery defects are supervised and learned using labeled data. By comparing with known defect labels, the parameters of the feature extraction module are adjusted so that it can more accurately extract features related to the defects. Among them, the high-dimensional feature map can be upsampled by using a feature decoder, such as a transposed convolutional layer, and reconstructed into a large-size feature map of the same size as the original image. This step allows the feature information in the feature map to be gradually released to low latitudes, which facilitates subsequent classification and recognition.
[0092] The preset defect probability scoring rules are specifically as follows:
[0093] A defect probability score for each pixel in a second feature sequence is calculated based on the activation function and the second preset defect label, and a second probability score map is generated based on the defect probability score; second defective pixels are classified based on the second probability score map and a preset probability threshold; the preset probability threshold also includes several second preset probability thresholds corresponding to different defect types.
[0094] When the score of any pixel in the second probability score map exceeds any of the second preset probability thresholds, the pixel output is defined as a second target defect type corresponding to the second preset probability threshold; wherein the second target defect type includes a second target feature code; and the second pixel distribution map is output according to the second target feature code of each pixel.
[0095] It should be noted that in this embodiment, after the above feature sequence is sent to the fully connected neural network for defective pixel classification, when the feature sequence is passed to the neural network, the neural network learns to associate the features with different types of defective pixels, propagates the feature information through the network layers, and performs nonlinear mapping through mechanisms such as activation functions, and finally obtains the classification results of the defective pixels. The algorithm output process is consistent with the visible light defect segmentation algorithm, and the classification pixel colors of the above six defect types are defined as [1,1,1]-[6,6,6]. As shown in the attached figure Figure 5 FIG. 1 is a schematic diagram of defect pixel distribution of a grayscale image in this embodiment, wherein pink represents VHB white glue bubbles, green represents depressions, and cyan represents corner drops.
[0096] Preferably, if the preset probability thresholds correspond to defect types such as scratches, dents, bruises, corner drops, white glue bubbles, trademark wrinkles, etc., then each threshold corresponds to a specific defect type. When the pixel score exceeds the corresponding threshold, the pixel is classified as the defect type corresponding to the threshold.
[0097] In summary, the RGB pixel distribution map helps to identify defects with significant color changes, such as dust, dirt, etc., which are significantly different from the surrounding normal areas in color and brightness; the grayscale pixel distribution map shows the brightness information of each pixel in the image without considering the color. This distribution map helps to identify defects that are not related to color, such as scratches, dents, bruises, etc., which show different brightness characteristics in grayscale images. Defect detection through grayscale images can reduce the interference of color on defect identification and improve the detection accuracy of certain defect types. By combining these two distribution maps, more comprehensive and accurate detection of battery surface defects can be achieved, thereby improving the quality control level of batteries.
[0098] S400: After calculating the defect hash value according to the pixel distribution map, eliminate foreign matter interference pixels according to the defect hash value to obtain a pixel distribution map of the defect area.
[0099] It should be noted that step S400 is mainly to merge the two pixel distribution maps output by the above two defect segmentation algorithms, aiming to eliminate the influence of false defect detection caused by dust environment, which is mainly manifested as dust attached to the battery, which will appear concave in the grayscale image. Based on this situation, the post-processing algorithm performs independent hash coding on the color values of the grayscale image concave and visible light dust. The hash coding can prevent interference with other defect category pixels. After the coding is completed, the respective defect areas are added. When the two defect hash values are added, a third independent hash value will appear. When the system detects a new hash value, the two defect types that intersect will be eliminated, which is manifested as setting the pixels of the two defect areas to [0,0,0], thereby eliminating the interference of dust on the grayscale image.
[0100] It should be noted that the main purpose of step S400 is to merge the pixel distribution maps output by two different defect segmentation algorithms to eliminate the interference of foreign matter such as dust on battery image analysis, which is specifically:
[0101] The defect hash value includes a first defect hash value and a second defect hash value; a first defect hash value is calculated by performing a first hash coding value on the first target feature code in the first pixel distribution map; a second defect hash value is calculated by performing a second hash coding value on the second target feature code in the second pixel distribution map.
[0102] Preferably, in this embodiment, a suitable hash function is first selected to convert the feature code into a hash value, for example, MD5, SHA-1, SHA-256, etc. The hash function should have good hashing properties, that is, different inputs should produce different outputs, and the outputs should be evenly distributed.
[0103] The first target feature code in the first pixel distribution map is processed, which may include extracting specific feature values or feature combinations, such as mean, variance, edge information, etc. The extracted feature code is converted into a digital format, such as an integer or a floating point number.
[0104] The converted feature code is used as input and the selected hash function is applied. The hash function converts the input into a fixed-length hash value, which is the first defect hash value.
[0105] Repeat the above steps, perform the same processing on the second target feature code in the second pixel distribution map, extract the features and apply the hash function to obtain the second defect hash value.
[0106] It should be noted that the obtained hash value is usually a binary string or a hexadecimal number. For ease of processing, the hash value can be converted to a numerical type, such as by converting from binary to decimal.
[0107] In subsequent processing, the hash value can be stored in a database or hash table for fast retrieval and calculation.
[0108] In this way, pixel feature encoding can be effectively converted into hash values for defect recognition and classification. This method is particularly suitable for scenarios that require efficient processing of large amounts of image data, as it can reduce calculation and matching time while maintaining high recognition accuracy.
[0109] Furthermore, the first defect hash value is added to the second defect hash value to obtain a third defect hash value, and a third pixel distribution map is generated; when it is detected that the third defect hash value meets the preset hash threshold, the third pixel distribution map is eliminated of foreign matter interference pixels to obtain a defect area pixel distribution map.
[0110] It should be noted that by merging two hash values, the features of two different defect types can be integrated, which means that the new hash value can reflect the common features of the two defect types, thereby improving the ability to identify complex defects.
[0111] When a new hash value is detected, the system can recognize that this is a combination of two defect types. This helps the system distinguish between a single defect type and a combination of multiple defect types, allowing it to more accurately classify and handle defects.
[0112] When the system detects a new hash value, it can eliminate pixels of the two defect types that intersect, which means that pixels that belong to both defect types can be removed from the final defect area pixel distribution map.
[0113] Especially for false detections caused by foreign objects such as dust, the new hash value can help the system eliminate these interferences. By setting pixels that belong to both dust and real defects to zero, the accuracy and reliability of defect detection can be improved.
[0114] In summary, this embodiment simplifies the subsequent processing steps through the merged hash value. It is no longer necessary to process each defect type separately. Instead, the merged hash value can be directly processed, thereby reducing the processing complexity and computing resource requirements, effectively combining two different defect detection methods, and improving the accuracy and robustness of the battery defect detection system.
[0115] In the present application, the method further comprises:
[0116] S500: constructing a defect position distribution heat map according to the defect area pixel distribution map, and outputting a risk assessment report according to the defect position distribution heat map.
[0117] It should be noted that this embodiment establishes a battery detection information database based on the defect area pixel distribution map output by the battery product inspection results in the above steps, further extracts production line defect data, performs data calculation and statistics on the number of defects and defect rates in specific areas of the battery product, and finally generates a defect location distribution heat map.
[0118] Among them, according to the pixel distribution map of the defect area, the preset areas and defect types are divided into groups; the number of defect occurrences and the defect rate of any of the groups are calculated, the calculation results are output, and a defect position distribution heat map is constructed based on the calculation results; the defect position distribution heat map is compared with a preset standard battery process data map, and the risk assessment report is output through a preset defect degree threshold.
[0119] Preferably, in this embodiment, the process of generating the defect location distribution heat map can be designed as follows:
[0120] Extract production line defect data from the database through a network application. The extracted data should include information such as production line name, defect occurrence time, defect type, area identification, and defect severity.
[0121] According to the location of the defect, the data is divided into positive and negative categories for separate processing. In order to avoid repeated calculations, the data needs to be deduplicated when counting defect data. Deduplication is performed through specific fields (such as product ID, defect type, and distribution area).
[0122] This embodiment can group the data by region and defect type.
[0123] Next, form a set containing each group of data, for example: {'center_area1': {'type1':count1, 'type2': count2, ...}, ...}.
[0124] Count each group and calculate the total number of defects of a certain type in each area.
[0125] It can be expressed by the number of elements in the collection, for example: {'center_area1': {'type1': count1, 'type2': count2, ...}, ...}.
[0126] For a particular region c and defect type t, the counts of all data points are summed.
[0127] The formula is: Σ indicator(x, c, t), where indicator(x, c, t) is an indicator function that returns 1 when the data point x belongs to region c and the defect type is t, otherwise it returns 0.
[0128] In this embodiment, the defect rate of each area can also be calculated, and the calculation of the defect rate is as follows:
[0129] This formula counts the number of data points in the statistical data set that match a specific region and defect type, that is, the number of a certain type of defect in the region. The defect rate of each region can be expressed as the number of a certain type of defect in the region divided by the total number of batteries, and then multiplied by 1000 to get a unit of one thousandth. The calculation formula is:
[0130] ;
[0131] This formula expresses the defect rate of defect type t in a specific area c. Due to the multiplication by 1000, the rate is expressed in parts per thousand.
[0132] In addition, the comprehensive defect rate of all defect types in a certain area can be calculated by summing the defect rates of all defect types:
[0133] ;
[0134] This represents the comprehensive defect rate of all defect types in a given area c. Based on the calculation results, the data structure required for the thermal map is constructed and finally transmitted to the system to draw the defect location distribution thermal map.
[0135] Preferably, assuming that you want to count the number of "scratch" type defects in the area "X1Y2", the formula will be: Σindicator(x, 'X1Y2', 'scratch').
[0136] Through the above steps, a detailed heat map can be generated to show the number of defects and defect rates in a specific area of the battery product, helping production line managers understand the quality status of the product and take corresponding improvement measures. Figure 6 As shown in FIG. 1 , a partial schematic diagram of the heat map of the battery front defect position distribution completed in this embodiment is shown. The battery surface detection area is divided into several small areas according to the predicted ratio, and then the quantity and type analysis is performed according to the defect characteristics in each area. For example, Figure 6 There are 3 dents in the image, accounting for 2.23%; there are 5 stains, accounting for 3.71%, and they are not limited to these.
[0137] In this embodiment, a risk assessment report is output according to the defect location distribution heat map, and the specific implementation process is as follows:
[0138] The acquisition of preset standard battery process data diagram can be: combining the contact points of equipment / carriers in the battery production process with the identified risk points in the battery production process experience library to draw a product process mapping diagram.
[0139] Compare the defect mapping diagram drawn by the software detection results with the process mapping diagram to quickly identify and lock the cause of the abnormality so as to intervene in time and avoid the occurrence of abnormalities and the spread of process risks.
[0140] Next, mark the point ranges of the equipment and carriers that come into contact with the battery during the entire battery production process on the battery mapping diagram. Based on the long-term accumulated experience library on battery production processes, mark the abnormal point ranges caused by raw materials, non-equipment, non-carrier and other reasons on the mapping diagram; combine the two markings to form a battery process mapping diagram.
[0141] Compare the battery process mapping diagram with the defect heat map, and characterize the defects by the defect type, defect area, defect frequency, and severity of the defects. Use the process knowledge experience library to identify risk points for rapid troubleshooting.
[0142] Finally, we conduct investigations on the locked risk points, and use defect detection software to discover abnormal time, scope, and defect information to assess risks, reduce the impact of abnormalities to the minimum scope in order to implement precise interception management, prevent the outflow of defective products, and save time for defective processing.
[0143] In summary, the embodiments of the present application can more accurately identify and distinguish different types of defects, such as dust, scratches, etc., by combining the defect pixel distribution map of visible light images and grayscale images; through hash value merging and deduplication processing, false detections caused by foreign matter (such as dust) can be reduced, while ensuring that real defects are not missed; through deduplication and grouping by area and defect type, data processing efficiency can be improved and unnecessary computing and storage requirements can be reduced; through the generated defect location distribution heat map, an intuitive quality assessment tool is provided, which helps production line managers to quickly identify and locate defective areas, so as to take corresponding quality control measures; and through the calculation of the comprehensive defective rate and the generation of heat maps, production line managers are supported to make data-based quality improvement decisions to improve the overall quality and market competitiveness of products. Embodiment 2:
[0144] As attached Figure 7 As shown, in order to solve the above technical problems, the present application also proposes a battery surface defect detection system, which at least includes: an image detection starting unit, an image preprocessing unit, an image defect detection processing unit and an image defect optimization processing unit.
[0145] It should be noted that Example 1 realizes real-time acquisition of image data on the battery surface during the battery production process, and analyzes possible minor defects on the battery pole pieces, such as scratches, stains, electrode misalignment, etc., through the defect detection system. If it is confirmed to be a defect, it will continue to be classified, and the type and location of the defect will be recorded. All detection information will be displayed on the interface of the battery surface defect detection system in Example 2. The operator can observe the quality status of the battery in real time and configure or intervene in the system when necessary. In addition, the system can also record detailed detection data online for subsequent quality analysis and improvement.
[0146] In this embodiment, the image detection starting unit is used to start the process to be detected and collect battery surface image data according to the process to be detected.
[0147] It should be noted that the unit provides a simple interface that allows the operator to select the battery surface image to be inspected and start the inspection process. The operator can easily start the inspection process without complicated settings.
[0148] The image preprocessing unit is used to perform battery position segmentation on the battery surface image data to extract the battery surface area; and obtain a battery position feature image after performing imaging angle correction on the battery surface area image.
[0149] It should be noted that the unit provides an image processing interface, including image segmentation, imaging angle correction and other functions, so that the operator can visually adjust the image. The operator can quickly extract the battery surface area from the original image and correct the imaging angle to improve the accuracy of subsequent defect detection.
[0150] The image defect detection processing unit is used to extract features from the battery position feature image and construct it into a feature sequence, and classify defective pixels in the feature sequence according to a preset defect probability scoring rule, and output a pixel distribution map that meets a preset probability threshold.
[0151] It should be noted that the unit provides an interface for feature extraction and defect pixel classification, including tools for setting defect probability scoring rules. The operator can view the construction process of the feature sequence and classify pixels according to the preset rules to output the pixel distribution map that meets the probability threshold.
[0152] The image defect optimization processing unit is used to calculate the defect hash value according to the pixel distribution map, and then eliminate the foreign matter interference pixels according to the defect hash value to obtain the defect area pixel distribution map.
[0153] It should be noted that the unit provides an interface for defect hash value calculation and foreign matter interference pixel elimination, including tools for setting hash thresholds and processing parameters. The operator can observe the defect hash value calculation process and eliminate foreign matter interference pixels based on the hash value to obtain a more accurate pixel distribution map of the defect area.
[0154] Furthermore, the image defect position prediction and analysis unit is used to construct a defect position distribution heat map according to the defect area pixel distribution map, and output a risk assessment report according to the defect position distribution heat map.
[0155] It should be noted that the unit provides an interface for building defect location distribution heat maps and outputting risk assessment reports, including tools for configuring heat map colors and thresholds. Operators can intuitively view defect location distribution heat maps and output risk assessment reports based on the heat maps so that they can take appropriate quality control measures.
[0156] In the implementation of the present application, the system also includes: an image defect position prediction and analysis unit and an abnormality alarm unit.
[0157] The image defect position prediction and analysis unit is used to construct a defect position distribution heat map according to the defect area pixel distribution map, and output a risk assessment report according to the defect position distribution heat map.
[0158] It should be noted that the unit provides a system interface for displaying a heat map of defect location distribution and allows operators to view and analyze risk assessment reports. The interface may include map views, heat map displays, charts, and statistical data. Operators can quickly identify the distribution of defects on the battery surface and understand the severity of defects based on the color depth of the heat map or data indicators. The risk assessment report provides a detailed analysis of potential defects to help operators make decisions. Figure 8 As shown, this is a schematic diagram of the system interface display and analysis risk assessment report of this embodiment, which mainly displays the quality inspection change trend, production line inspection results and important indicator analysis; in addition, the overall interface display also includes: alarm service, AI algorithm management, authority management, log management and debugging function and other functional interfaces, but are not limited to these.
[0159] The abnormal alarm unit is used to output the current battery surface defect position coordinates and damage degree information to the management personnel when the risk assessment report triggers the abnormal alarm condition.
[0160] It should be noted that the unit can provide an alarm display interface. When the risk assessment report triggers an abnormal alarm condition, the interface will display the location coordinates and degree of damage of the current battery surface defect. The interface may include alarm notification, coordinate information, description of the degree of damage, etc. The abnormal alarm unit can automatically detect any abnormality detected by the system and immediately notify the management personnel. This is crucial to take timely measures to prevent defective batteries from entering the market. Operators can take immediate action, such as stopping the production line, isolating defective batteries, or conducting further inspections.
[0161] In summary, in Example 2, by citing the analysis of the interface layout and effects of each unit of the above system, the operator can more effectively monitor the defects on the battery surface and take corresponding quality control measures. This system design helps to improve the quality control level in the battery production process and reduce the generation of defective products. The specific implementation process and principle have been introduced in detail in Example 1 and will not be repeated in this embodiment. Embodiment three:
[0162] The present application also provides a storage medium, which is a type of computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements any of the battery surface defect detection methods described above.
[0163] Preferably, in the embodiment of the present application, the computer readable storage medium can be used to store a computer program for implementing the battery surface defect detection method. Such storage medium can be in various forms, including but not limited to a hard disk drive, a solid state drive, a USB flash drive, an optical disk, a cloud storage service, etc.
[0164] In this embodiment, the software of the entire battery surface defect detection system, including all necessary algorithms, databases and user interfaces, can be stored in a hard disk drive (HDD); a solid-state drive (SSD) can be used as the system's boot disk to store the operating system and the battery surface defect detection program to ensure quick startup and efficient operation; a USB flash drive can be used to store updated versions of the battery surface defect detection or programs with specific functions to facilitate updates and maintenance by on-site technicians; a CD can be used as a software distribution medium to store the initial version or important updates of the battery surface defect detection system; and a cloud storage service can be used to store the latest version of the battery surface defect detection system for remote access by multiple users or devices.
[0165] In summary, these storage media can be selected and used according to actual application requirements to ensure the effective operation of the battery surface defect detection system and the safe storage of data. By storing these programs on computer-readable storage media, the battery surface defect detection system can be easily installed, updated and maintained while ensuring the stability and reliability of the system.
[0166] The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc. The memory is used to store an operating system, an application program, a boot loader, data, and other programs, such as the program code of the computer program, etc. The memory may also be used to temporarily store data that has been output or is to be output.
[0167] In several embodiments provided in the present application, it is understood that each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved.
[0168] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.
[0169] The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the scope of protection of the present application. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of protection of the present application.
Claims
1. A method for detecting battery surface defects, characterized in that: include: S100: starting a process to be detected, and collecting battery surface image data according to the process to be detected; S200: performing battery position segmentation on the battery surface image data to extract the battery surface area; performing imaging angle correction on the battery surface area image to obtain a battery position feature image; S300: extracting features from the battery position feature image to construct a feature sequence, classifying defective pixels in the feature sequence according to a preset defect probability scoring rule, and outputting a pixel distribution map that meets a preset probability threshold; S400: after calculating the defect hash value according to the pixel distribution map, eliminating foreign matter interference pixels according to the defect hash value to obtain a pixel distribution map of the defect area; In the step S300, the battery position characteristic image includes a first image output and displayed in RGB format; the pixel distribution map includes a first pixel distribution map; Extract all pixels in the first image and convert them into a first preset dimensional feature map, and perform first feature extraction through a first preset defect label; Converting the output display form of the extracted pixel feature information from the first preset dimension to restore the original dimension, and flattening it into a first feature sequence; In the step S300, the battery position characteristic image further includes a second image outputted and displayed in grayscale form; the pixel distribution map further includes a second pixel distribution map; Extract all pixels in the second image and convert them into a second preset dimensional feature map, and perform second feature extraction through a second preset defect label; The output display form of the extracted pixel feature information is converted from the second preset dimension to the original dimension and flattened into a second feature sequence.
2. A method for detecting battery surface defects according to claim 1, characterized in that: In step S100, The process to be inspected is generated by inputting the battery information of the current batch; wherein the process to be inspected includes a plurality of inspection items, and the inspection items are configured by the battery information of the current batch; The battery surface image data includes several frames of upper surface data and lower surface data of any battery.
3. A method for realizing battery surface defect detection according to claim 2, characterized in that: In the step S200, the imaging angle correction is performed on the battery surface area map, specifically: Comparing a preset standard template area with the battery surface area to obtain an imaging angle difference; The battery surface area is spatially transformed in coordinate projection according to the imaging angle difference to correct the imaging angle and obtain the battery position characteristic image.
4. A method for realizing battery surface defect detection according to claim 3, characterized in that: In step S300, the preset defect probability scoring rule is specifically: Calculating a defect probability score for each pixel in the first feature sequence according to the activation function and the first preset defect label, and generating a first probability score map according to the defect probability score; Performing a first defect pixel classification according to the first probability score map and a preset probability threshold; the preset probability threshold includes a plurality of first preset probability thresholds corresponding to different defect types; When the score of any pixel in the first probability score map exceeds any of the first preset probability thresholds, defining the pixel output as a first target defect type corresponding to the first preset probability threshold; wherein the first target defect type includes a first target feature code; The first pixel distribution map is output according to the first target feature encoding of each pixel.
5. A method for realizing battery surface defect detection according to claim 4, characterized in that: In step S300, the preset defect probability scoring rule is specifically: Calculating a defect probability score for each pixel in the second feature sequence according to the activation function and the second preset defect label, and generating a second probability score map according to the defect probability score; Performing a second defect pixel classification according to the second probability score map and a preset probability threshold; the preset probability threshold also includes a plurality of second preset probability thresholds corresponding to different defect types; When the score of any pixel in the second probability score map exceeds any of the second preset probability thresholds, defining the pixel output as a second target defect type corresponding to the second preset probability threshold; wherein the second target defect type includes a second target feature code; The second pixel distribution map is output according to the second target feature encoding of each pixel.
6. A method for realizing battery surface defect detection according to claim 5, characterized in that: The step S400 is specifically as follows: The defect hash value includes a first defect hash value and a second defect hash value; Perform a first hash encoding value on the first target feature code in the first pixel distribution map to calculate a first defect hash value; Perform a second hash encoding value on the second target feature code in the second pixel distribution map to calculate a second defect hash value; Adding the first defect hash value to the second defect hash value to obtain a third defect hash value, and generating a third pixel distribution map; When it is detected that the third defect hash value meets the preset hash threshold, the foreign matter interference pixels are eliminated from the third pixel distribution map to obtain a defect area pixel distribution map.
7. A method for realizing battery surface defect detection according to claim 6, characterized in that: The method further comprises: S500: constructing a defect position distribution heat map according to the defect area pixel distribution map, and outputting a risk assessment report according to the defect position distribution heat map; wherein, according to the defect area pixel distribution map, the groups are divided according to the preset areas and defect types; Calculate the number of defect occurrences and defect rates according to any of the groups, output the calculation results, and construct a defect location distribution heat map according to the calculation results; The defect location distribution heat map is compared with a preset standard battery process data map, and the risk assessment report is output according to a preset defect degree threshold.
8. A battery surface defect detection system, used to implement a battery surface defect detection method as described in any one of claims 1 to 7, characterized in that: include: Image detection starting unit, image preprocessing unit, image defect detection processing unit and image defect optimization processing unit; The image detection starting unit is used to start the process to be detected and collect battery surface image data according to the process to be detected; The image preprocessing unit is used to perform battery position segmentation on the battery surface image data to extract the battery surface area; and obtain a battery position feature image after performing imaging angle correction on the battery surface area image; The image defect detection processing unit is used to extract features from the battery position feature image and construct it into a feature sequence, and classify defective pixels of the feature sequence according to a preset defect probability scoring rule, and output a pixel distribution map that meets a preset probability threshold; The image defect optimization processing unit is used to calculate the defect hash value according to the pixel distribution map, and then eliminate the foreign matter interference pixels according to the defect hash value to obtain a defect area pixel distribution map; And, the image defect position prediction and analysis unit is used to construct a defect position distribution heat map according to the defect area pixel distribution map, and output a risk assessment report according to the defect position distribution heat map; The image defect detection processing unit extracts features from the battery position feature image and constructs it into a feature sequence, and classifies defective pixels of the feature sequence according to a preset defect probability scoring rule, and outputs a pixel distribution map that meets a preset probability threshold: the battery position feature image includes a first image output and displayed in RGB format; the pixel distribution map includes a first pixel distribution map; Extract all pixels in the first image and convert them into a first preset dimensional feature map, and perform first feature extraction through a first preset defect label; Converting the output display form of the extracted pixel feature information from the first preset dimension to restore the original dimension, and flattening it into a first feature sequence; The image defect detection processing unit extracts features from the battery position feature image and constructs it into a feature sequence, and classifies defective pixels of the feature sequence according to a preset defect probability scoring rule, and outputs a pixel distribution map that meets a preset probability threshold, wherein the battery position feature image also includes a second image output in grayscale form; the pixel distribution map also includes a second pixel distribution map; Extract all pixels in the second image and convert them into a second preset dimensional feature map, and perform second feature extraction through a second preset defect label; The output display form of the extracted pixel feature information is converted from the second preset dimension to the original dimension and flattened into a second feature sequence.
9. A battery surface defect detection system according to claim 8, characterized in that: Also includes: Image defect position prediction and analysis unit and abnormality alarm unit; The image defect position prediction and analysis unit is used to construct a defect position distribution heat map according to the defect area pixel distribution map, and output a risk assessment report according to the defect position distribution heat map; The abnormal alarm unit is used to output the current battery surface defect position coordinates and damage degree information to the management personnel when the risk assessment report triggers the abnormal alarm condition.
10. A storage medium, which is a computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the battery surface defect detection method as described in any one of claims 1 to 7 is implemented.
Citation Information
Patent Citations
Far-end chip defect detection system based on deep learning technology
CN113870202A
Strip steel surface defect detection method and system
CN115272293A
Optical fiber end face detection device and method and optical fiber end face detection model training method
CN115797746A
Inspection apparatus and an inspection method
US20140319351A1
