An Image Detection Method for Surface Defects of the Insulating Shell of a Lithium Battery Charger

Abstract

The present invention provides a method for detecting surface defects of the insulating housing of a lithium battery charger, belonging to the field of image processing; through multi-scale feature extraction, high-priority screening, and adaptive filtering techniques, efficient and accurate detection of surface defects of the insulating housing of a lithium battery charger is achieved; the method first takes and annotates defect images on the surface of the insulating housing in a production environment to form a data set; then, the defect images are input, three feature maps of different scales are generated, and features are screened according to the priority value confidence, and only the features corresponding to the high priority values are retained; next, through multi-scale feature priority processing, the flexibility and accuracy of feature processing are enhanced, and the features and boundaries of the region of interest are enhanced by adaptively filtering sparse defect components; finally, a defect recognition model is constructed according to the module; through deep learning and image processing techniques, the method significantly improves the efficiency and accuracy of surface defect detection of lithium battery chargers.

Description

Technical Field

[0001] The present invention belongs to the field of image processing, and particularly relates to a method for detecting surface defects of an insulating housing for a lithium battery charger. Background Art

[0002] Lithium batteries have been widely used in modern life and play an important role in fields such as electric vehicles, consumer electronics, and energy storage systems; as an important accessory for lithium batteries, the quality of lithium battery chargers directly affects the performance and safety of lithium batteries; the insulating housing of a lithium battery charger, as a protective component, its surface quality is particularly crucial; surface defects of the insulating housing, such as scratches, cracks, and deformations, not only affect the appearance but may also pose potential hazards to the function and safety of the charger; in order to ensure the reliability of the charger, a detection method for surface defects of the insulating housing has emerged and become an indispensable part of the production process.

[0003] Traditional surface defect detection methods mainly rely on manual inspection and simple machine vision technologies; however, manual inspection is inefficient and the accuracy is limited by the experience and attention of the operator; while simple machine vision technologies tend to be insufficient when faced with complex surface defects and are difficult to efficiently and accurately identify various types of defects; with the rapid development of deep learning and image processing technologies, multi-scale feature extraction and adaptive filtering technologies have gradually been applied to the field of defect detection, greatly improving the detection accuracy and efficiency; therefore, developing a method for detecting surface defects of the insulating housing of a lithium battery charger based on advanced image processing technologies has become a hot topic in the industry.

[0004] The present invention proposes a method for detecting surface defects of an insulating housing for a lithium battery charger, aiming to achieve efficient detection of surface defects of the insulating housing of a lithium battery charger through technical means such as constructing a multi-scale feature extraction module, a high-priority screening module, and an adaptive filtering sparse defect component; this method constructs a special surface defect dataset by collecting and annotating image data on the surface of the insulating housing in a production environment; subsequently, feature extraction is performed using a deep convolutional neural network, and through steps such as multi-scale feature fusion, priority screening, and adaptive filtering, the accuracy and efficiency of defect recognition are gradually improved; finally, the trained defect recognition model can accurately identify the location and type of surface defects of the charger insulating housing, providing strong technical support for the quality control of the charger. Summary of the Invention

[0005] The present invention proposes a method for detecting surface defects of an insulating housing for a lithium battery charger, aiming to improve the safety and reliability of lithium battery chargers during the production and detection processes through efficient and automated image processing technologies.

[0006] A method for detecting surface defects of the insulating shell of a lithium battery charger, comprising the following steps:

[0007] S1. Production of the surface dataset of the insulating shell of the lithium battery charger. In the production environment of the lithium battery charger, take an overall surface photo of the lithium battery charger with defects on the surface of the insulating shell, annotate the obtained image, and the annotation targets are the surface defect areas and five defect types of the insulating shell of the lithium battery charger. After the annotation is completed, the surface dataset of the insulating shell of the lithium battery charger is obtained;

[0008] S2. Construct a multi-scale feature extraction module. Input the defect image to obtain three feature maps of different scales;

[0009] S3. Construct a high-priority screening module. Perform query screening through the priority value confidence level, and only retain the queries corresponding to the priority values higher than the priority value confidence level;

[0010] S4. Construct a multi-scale feature priority processing module to make the priority processing module more flexible and accurate when processing multi-scale features, thereby improving the performance of defect recognition;

[0011] S5. Construct an AFSD adaptive filtering sparse defect component to enhance the features and boundaries of the region of interest;

[0012] S6. Construct a surface defect recognition model for the insulating shell of the lithium battery charger, including an input module, a multi-scale feature generation module, a high-priority screening module, a multi-scale feature priority processing module, a CCRS module, and an AFSD adaptive filtering sparse defect component;

[0013] S7. Train and detect the surface defect recognition model of the insulating shell of the lithium battery charger. Use the lithium battery charger for detection. Use the surface defect dataset of the insulating shell of the lithium battery charger charged by the lithium battery to train the surface defect recognition model of the insulating shell of the lithium battery charger. After the training is completed, input the surface image of the insulating shell of the new lithium battery charger to be detected into the surface defect recognition model of the insulating shell of the lithium battery charger to obtain the detection result. The detection result includes the defect position and type in the surface image of the insulating shell of the lithium battery charger.

[0014] Preferably, in the method for detecting surface defects of the insulating shell of a lithium battery charger, in step S1, for the five defect types on the surface of the insulating shell of the lithium battery charger, specifically include five major defect categories: scratches on the surface of the shell, color difference on the surface of the shell, cracks on the surface of the shell, bumps on the surface of the shell, and deformation of the surface of the shell.

[0015] Preferably, in the method for detecting surface defects of the insulating shell of a lithium battery charger, in step S2, for the multi-scale feature extraction module, specifically includes the following steps:

[0016] Receive the surface defect images of a group of lithium battery chargers, denoted as X, where the size of each image is H×W×C, where H is the height, W is the width, and C is the number of channels equal to 3; perform feature extraction on each image X and input it into a deep convolutional neural network, where the convolutional layer and pooling layer in the network extract the initial feature values of the image; perform forward propagation in the network to extract multi-layer feature maps, and through a feature fusion strategy, convert the initial feature maps into three-scale feature maps F l 、F m 、F h , specifically as follows F h =H×W×C 3 , where C 1 、C 2 and C 3 are the new numbers of channels; use interpolation and pooling operations to unify the scales of the multi-scale feature maps to obtain a set of feature maps with the same resolution; input the set of feature maps with unified scales into the feature selection module, screen out important features through the attention mechanism, and perform feature recombination to form the final feature representation.

[0017] Preferably, in the method for detecting surface defects of the insulating housing of a lithium battery charger, in step S3, the high-priority screening module is specifically as follows:

[0018] Input the three-scale feature maps F l 、F m 、F h generated by the multi-scale feature generation module, and each scale feature contains n query points Q i ={q i1 ,q i2 ,...,q in}, where i represents the scale level; for each query point Q i , calculate its position and related priority value in the power grid terminal defect picture, and the priority value calculation formula is changed to:

[0019]

[0020] where W and b are learnable parameters; by setting a dynamic priority value confidence interval and dynamically adjusting in the range of 0.6 to 1, screen out the query points with priority values higher than the confidence level to form a high-priority screened feature set S.

[0021] Preferably, in the method for detecting surface defects of the insulating housing of a lithium battery charger, in step S4, the specific design of the multi-scale feature priority processing module is as follows:

[0022] In the multi-scale feature prioritization module, the calculation method of the query priority value is redefined. The set of feature values S calculated in step S3 is judged, and the query with a value higher than the set threshold t is called a priority query. The threshold t is dynamically adjusted during the training process to adapt to different feature distributions;

[0023] An improved attention mechanism is applied to the priority query to enhance its feature representation ability. The formula is as follows:

[0024]

[0025] Where Q i is the query point, K i is the key, V i is the value, d k is the scaling factor, and T is the set threshold;

[0026] For the selected priority features S, they are given priority in the subsequent processing steps, thereby improving the accuracy and efficiency of defect recognition; other unselected queries remain unchanged and are used as auxiliary information in the subsequent processing stage, thereby improving the overall recognition accuracy and robustness.

[0027] Preferably, for the method for detecting surface defects of the insulating shell of a lithium battery charger, the AFSD adaptive filtering sparse defect component in step S5 is specifically designed as follows: First, the generation of the initial heat map is carried out. The specific formula is as follows:

[0028]

[0029] Specifically, when the defect is less than from the edge, assuming the margin is L, the window width from the defect to the non-edge side should be k - L; Therefore, we need to correct it and move the window center position to obtain a window of the same size. The specific formula is:

[0030]

[0031] In these two formulas, G is the two-dimensional matrix of the original wafer image; the value at the coordinate (i, j) in the G[i, j] heat map represents the defect density in this area; k is the size of the region of interest window; [m, n] are the coordinates in the region of interest window, traversing the pixel points near i and j; Then, the mapping operation of the density eigenvalue is carried out. The specific formula is:

[0032]

[0033] G[i, j] = (G[i, j] - min(G)) · Dist;

[0034] Among them, Dist is the density eigenvalue scaling factor. The threshold Q is determined by the cumulative number of regions of different types of defects in the production batch. Finally, sparse defect regions are filtered out, and the eigenvalues of dense defect regions are retained. The specific formula is as follows:

[0035]

[0036] Among them, G[i,j] is the final thermal value, and Q is the adaptive density threshold.

[0037] Preferably, in the method for detecting surface defects of an insulating shell of a lithium battery charger, in step S6, a surface defect recognition network model of the insulating shell of the lithium battery charger is constructed. First, the power grid terminal image is input into the multi-scale feature generation module to generate three different scales of features; then, these features are input into the high-priority screening module, and screening is performed among the three scales through the priority confidence; subsequently, the screened features are input into multiple cascaded multi-scale feature priority processing modules for multiple high-priority screenings and self-attention calculations; the detected defect regions are input into the AFSD adaptive filtering sparse defect component for defect type classification operations; finally, the surface defect position and defect type information of the insulating shell of the lithium battery charger are output.

[0038] The method for detecting surface defects of an insulating shell of a lithium battery charger proposed by the present invention compared with the prior art: The present invention adopts deep learning and multi-scale feature extraction technologies to construct a deep convolutional neural network for efficient extraction and accurate recognition of various defect types; the multi-scale feature extraction module of the present invention generates feature maps of different scales, enhancing the recognition ability for different defects; the high-priority screening module screens through the priority confidence to improve the detection accuracy; the AFSD adaptive filtering sparse defect component effectively filters out unimportant defect regions, improving the detection accuracy and efficiency; the present invention is significantly superior to the prior art in terms of detection efficiency, accuracy, and adaptability to complex environments, providing a more reliable defect detection solution. Description of the Drawings

[0039] Figure 1 It is a flow chart of a method for detecting surface defects of an insulating shell of a lithium battery charger.

[0040] Figure 2 It is a flow chart of multi-scale feature extraction, priority screening, and processing.

[0041] Figure 3 It is a schematic diagram of the network of the surface defect recognition model of the insulating shell of the constructed lithium battery charger.

[0042] Figure 4 It is a schematic diagram of the detection output result of the surface defect recognition model of the insulating shell of the lithium battery charger. Detailed implementation manners

[0043] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0044] Please refer to Figures 1-4 , the present invention provides a technical solution.

[0045] The present invention proposes a method for detecting surface defects of the insulating shell of a lithium battery charger, aiming to improve the safety and reliability of the lithium battery charger in the production and detection processes through efficient and automated image processing technologies.

[0046] A method for detecting surface defects of the insulating shell of a lithium battery charger, the process is as Figure 1 shown, and specifically includes the following steps:

[0047] S1. Production of the surface data set of the insulating shell of the lithium battery charger. In the production environment of the lithium battery charger, take a whole surface photo of the lithium battery charger with defects on the surface of the insulating shell, annotate the obtained image, and the annotation targets are the surface defect areas and five defect types of the insulating shell of the lithium battery charger. After the annotation is completed, the surface data set of the insulating shell of the lithium battery charger is obtained;

[0048] S2. Construct a multi-scale feature extraction module, input the defect image, and obtain three feature maps of different scales;

[0049] S3. Construct a high-priority screening module, perform query screening through the priority value confidence level, and only retain the queries corresponding to the priority values higher than the priority value confidence level;

[0050] S4. Construct a multi-scale feature priority processing module to make the priority processing module more flexible and accurate when processing multi-scale features, thereby improving the performance of defect recognition at the power grid terminal;

[0051] S5. Construct an AFSD adaptive filtering sparse defect component to enhance the features and boundaries of the region of interest;

[0052] S6. Construct a surface defect recognition model for the insulating shell of the lithium battery charger, including an input module, a multi-scale feature generation module, a high-priority screening module, a multi-scale feature priority processing module, a CCRS module, and an AFSD adaptive filtering sparse defect component;

[0053] S7. Train and detect the surface defect recognition model of the insulating housing of the lithium battery charger. Use the lithium battery charger for detection. Train the surface defect recognition model of the insulating housing of the lithium battery charger using the surface defect dataset of the insulating housing of the lithium battery charger. After training, input the surface image of the insulating housing of the new lithium battery charger to be detected into the surface defect recognition model of the insulating housing of the lithium battery charger to obtain the detection result. The detection result includes the defect position and type in the surface image of the insulating housing of the lithium battery charger.

[0054] Further, in step S1 of a method for detecting surface defects of the insulating housing of a lithium battery charger, the five types of surface defects of the insulating housing of the lithium battery charger specifically include five major defect categories: scratches on the housing surface, color difference on the housing surface, cracks on the housing surface, bumps on the housing surface, and deformation of the housing surface.

[0055] Further, in step S2 of a method for detecting surface defects of the insulating housing of a lithium battery charger, the multi-scale feature extraction module is as Figure 2 shown on the left, and specifically includes the following steps:

[0056] Receive a set of power grid terminal defect images, denoted as X, where the size of each image is H×W×C, where H is the height, W is the width, and C is the number of channels equal to 3; perform feature extraction on each image X and input it into a deep convolutional neural network. The convolutional layer and pooling layer in the network extract the initial feature values of the image; perform forward propagation in the network to extract multi-layer feature maps, and through the feature fusion strategy, convert the initial feature maps into three-scale feature maps F l 、F m 、F h , specifically F h =H×W×C 3 , where C 1 、C 2 and C 3 are the new number of channels; use interpolation and pooling operations to unify the scales of the multi-scale feature maps to obtain a set of feature maps with the same resolution; input the set of feature maps with unified scales into the feature selection module, screen out important features through the attention mechanism, and perform feature recombination to form the final feature representation.

[0057] Further, in step S3 of a method for detecting surface defects of the insulating housing of a lithium battery charger, the high-priority screening module is as Figure 2 shown in the middle, and the specific design is as follows:

[0058] Input F l 、F m 、Fh Feature maps of three scales, each scale feature contains n query points Q i ={q i1 , q i2 ,..., q in}, where i represents the scale level; for each query point Q i , calculate its position and related priority value in the power grid terminal defect image, and the priority value calculation formula is changed to:

[0059]

[0060] where W and b are learnable parameters; by setting the dynamic priority value confidence interval, which is dynamically adjusted in the range of 0.6 to 1, filter out the query points with priority values higher than the confidence level to form the feature set S after high-priority screening.

[0061] Furthermore, in step S4 of a method for detecting surface defects of an insulating shell of a lithium battery charger, the multi-scale feature priority processing module is as shown Figure 2 on the right, and the specific design is as follows:

[0062] In the multi-scale feature priority processing module, the calculation method of the query priority value is redefined, and the feature value set S calculated in step S3 is judged. The query with a value higher than the set threshold t is called a priority query, and the threshold t is dynamically adjusted during the training process to adapt to different feature distributions;

[0063] Apply an improved attention mechanism to the priority query to enhance its feature representation ability. The formula is as follows:

[0064]

[0065] where Q i is the query point, K i is the key, V i is the value, d k is the scaling factor, and T is the set threshold;

[0066] For the selected priority features S, give priority consideration in the subsequent processing steps, so as to improve the accuracy and efficiency of defect recognition; other unselected queries will remain unchanged and be used as auxiliary information in the subsequent processing stage, so as to improve the overall recognition accuracy and robustness.

[0067] Furthermore, in step S5 of the method for detecting surface defects of an insulating shell of a lithium battery charger, the AFSD adaptive filtering sparse defect component is specifically designed as follows: First, generate the initial heat map. The specific formula is as follows:

[0068]

[0069] Specifically, when the distance of the defect from the edge is less than , assuming the edge margin is L, the window width from the defect to the non-edge side should be k - L; therefore, we need to correct it and move the window center position to obtain a window of the same size. The specific formula is:

[0070]

[0071] In these two formulas, G is the two-dimensional matrix of the original wafer image; G[i, j] represents the value at the coordinate (i, j) in the heat map, indicating the defect density of the area; k is the size of the region of interest window; [m, n] are the coordinates in the region of interest window, traversing the pixel points near i and j; then a mapping operation of the density eigenvalue is performed. The specific formula is:

[0072]

[0073] G[i, j] = (G[i, j] - min(G)) · Dist;

[0074] Among them, Dist is the density eigenvalue scaling factor. The threshold Q is determined by accumulating the number of regions of different types of defects in the production batch. Finally, the sparse defect regions are filtered out, and the eigenvalues of the dense defect regions are retained. The specific formula is as follows:

[0075]

[0076] Among them, G[i, j] is the final heat value, and Q is the adaptive density threshold.

[0077] Furthermore, in step S6 of a method for detecting surface defects of an insulating housing of a lithium battery charger, as Figure 3 shown is a schematic diagram of the surface defect recognition network model of the insulating housing of the designed lithium battery charger. The specific process is as follows: First, the grid terminal image is input into the multi-scale feature generation module to generate three features of different scales; then, these features are input into the high-priority screening module, and screening is performed among the three scales through the priority value confidence; subsequently, the screened features are input into multiple cascaded multi-scale feature priority processing modules for multiple high-priority screenings and self-attention calculations; the detected defect regions are input into the AFSD adaptive filtering sparse defect component for defect type classification operations; finally, the surface defect position and defect type information of the insulating housing of the lithium battery charger are output.

[0078] Furthermore, the detection result of the surface defects of the insulating housing of the lithium battery charger is as Figure 4As shown, it can be seen that the surface defect detection model of the lithium battery charger can well locate the surface defect area of the insulation shell of the lithium battery charger and output accurate defect classification results at the program end.

Claims

1. A method for detecting surface defects of an insulating shell of a lithium battery charger, characterized in that: The following steps are involved: S1. Preparation of the surface dataset of the insulating shell of a lithium battery charger. In the production environment of a lithium battery charger, the entire surface of a lithium battery charger with defects on the surface of the insulating shell is photographed, and the images obtained by the photography are annotated. The annotated targets are the surface defect areas and five defect types of the insulating shell of the lithium battery charger. After the annotation is completed, the surface dataset of the insulating shell of the lithium battery charger is obtained. S2, construct a multi-scale feature extraction module, input the defect image, and obtain feature maps of three different scales; S3. Construct a high priority screening module, perform query screening by priority value confidence, and only retain queries corresponding to priority values ​​higher than the priority value confidence; the steps of the high priority screening module are: Input F obtained by the multi-scale feature generation module l 、F m 、F h Feature maps of three scales, each scale feature contains n query points Q i = {q i1 ,q i2 , ..., q in }, calculate its position and related priority value in the surface defect image of the insulating shell of the lithium battery charger, and the priority value calculation formula is changed to: Among them, β and γ are learnable parameters; by setting the dynamic priority value confidence interval, dynamically adjusting it in the range of 0.6 to 1, the query points with priority values ​​higher than the confidence value are screened out to form a high-priority screened feature set S; S4, construct a multi-scale feature priority processing module, redefine the query priority value calculation method, judge the feature value set S calculated in step S3, and call the query higher than the set threshold t as the priority query; S5. Construct an adaptive filtering sparse defect AFSD component to enhance the features and boundaries of the region of interest; first generate a thermal map of the initial features; when the defect distance edge is less than the preset value, it needs to be corrected and the window center position is moved to obtain a window of the same size; then perform a mapping operation of the density feature value; finally filter out the sparse defect area and retain the feature value of the dense defect area; S6. Construct a surface defect recognition model for the insulation shell of a lithium battery charger, including an input module, a multi-scale feature generation module, a high priority screening module, a multi-scale feature priority processing module, a CCRS module, and an adaptive filtering sparse defect AFSD component; S7. Train and detect the surface defect recognition model of the insulating shell of the lithium battery charger. Use the lithium battery charger for detection. Use the insulating shell surface defect dataset of the lithium battery charger to train the insulating shell surface defect recognition model of the lithium battery charger. After the training is completed, input the new insulating shell surface image of the lithium battery charger to be tested into the insulating shell surface defect recognition model of the lithium battery charger to obtain the test results. The test results include the defect location and type in the insulating shell surface image of the lithium battery charger.

2. The method for detecting surface defects of an insulating shell of a lithium battery charger according to claim 1, characterized in that: In step S1, the five types of surface defects of the insulating shell of the lithium battery charger include scratches on the shell surface, color difference on the shell surface, cracks on the shell surface, bumps on the shell surface and deformation of the shell surface.

3. The method for detecting surface defects of an insulating shell of a lithium battery charger according to claim 2, characterized in that: In step S2, the multi-scale feature extraction module specifically includes the following steps: receiving a group of images of surface defects of the insulating shell of the lithium battery charger, denoted as X, wherein the size of each image is H×W×C, wherein H is the height, W is the width, C is the number of channels and the number of channels is equal to 3; extracting features from each image X, inputting it into a deep convolutional neural network, and extracting initial feature values ​​from the image by the convolution layer and the pooling layer in the network; performing forward propagation in the network, extracting multi-layer feature maps, and converting the initial feature maps into three scale feature maps F through a feature fusion strategy. l 、F m 、F h , specifically F h =H×W×C3, where C1, C2 and C3 are the new number of channels; the multi-scale feature maps are scaled uniformly by using interpolation and pooling operations to obtain a set of feature maps with the same resolution; the scaled feature map set is input into the feature selection module, important features are screened out through the attention mechanism, and features are reorganized to form the final feature representation.

4. The method for detecting surface defects of an insulating shell of a lithium battery charger according to claim 3, characterized in that: The multi-scale feature priority processing module described in step S4 is specifically designed as follows: In the multi-scale feature priority processing module, the threshold t is dynamically adjusted during the training process to adapt to different feature distributions; An improved attention mechanism is applied to the priority query to enhance its feature representation capability. The formula is as follows: Among them, Q i is the query point, K i is the key, V i is the value, d k is the scaling factor; The feature value set S calculated in step S3 is given priority in the subsequent processing steps; other unselected queries will remain unchanged and used as auxiliary information in the subsequent processing stages.

5. The method for detecting surface defects of an insulating shell of a lithium battery charger according to claim 4, characterized in that: The specific design of the adaptive filtering sparse defect AFSD component described in step S5 is as follows: First, an initial thermal map is generated, and the specific formula is as follows: When the defect is closer to the edge When , assuming the margin is L, the window width from the defect to the non-edge side should be wL; therefore, we need to correct it and move the window center position to obtain a window of the same size. The specific formula is: In these two formulas, G is the two-dimensional matrix of the original image; G[a,b] represents the thermal value at the coordinate (a,b) in the thermal map; w is the size of the interest window; [x,y] is the coordinate in the interest window, traversing the pixels near (a,b); Then the density eigenvalue mapping operation is performed, and the specific formula is: G1[a,b]=(G[a,b]-min(G))·Dist; Among them, Dist is the density eigenvalue scaling factor. The threshold Q is determined by the cumulative number of regions with different types of defects in the production batch. Finally, the sparse defect areas are filtered out and the eigenvalues ​​of the dense defect areas are retained. The specific formula is as follows: Among them, G2[a, b] obtained by this formula is the final thermal value at the coordinate (a, b), and Q is the adaptive density threshold.

6. The method for detecting surface defects of an insulating shell of a lithium battery charger according to claim 5, characterized in that: The network model for identifying the surface defects of the insulating shell of the lithium battery charger constructed in step S6 first inputs the surface image of the insulating shell of the lithium battery charger into the multi-scale feature generation module to generate features of three different scales; then, these features are input into the high priority screening module to screen them in three scales by the priority value confidence; then, the screened features are input into the multi-scale feature priority processing module in series to perform multiple high priority screening and self-attention calculations; the detected defect area is input into the adaptive filtering sparse defect AFSD component to perform defect type classification operations; and finally, the defect position and defect type information of the surface of the insulating shell of the lithium battery charger are output.

Images