Method and device for detecting tiny foreign matter particles in battery and medium

Abstract

The invention relates to a method and a device for detecting tiny foreign matter particles in a battery and a medium. The method comprises the following steps: acquiring slice images of different surfaces and sections after CT scanning of the battery; using a deep learning image segmentation method to classify the slice images and obtaining a battery contour in each slice image; corroding a plurality of pixels outwards by taking the battery contour of each slice image as a reference to form a new contour, and delimiting a region of interest by a minimum bounding rectangle of the new contour; detecting the region of interest by using a deep learning target detection algorithm to obtain a detection result; meanwhile, the invention discloses a device and a medium using the method. The product precision is improved, the working efficiency is improved, and the labor cost and the time cost are saved.

Description

Technical Field

[0001] The present invention relates to the field of battery detection, and in particular to a method, a device and a medium for detecting tiny foreign particles inside a battery. Background Art

[0002] The quality of mobile phone batteries has become a key factor that more and more consumers consider. Mobile phone batteries usually need to go through a variety of complex process flows during production, among which the production quality of the positive and negative electrodes inside the battery is a key link in the entire mobile phone battery production. The positive and negative electrodes are nested and stacked inside the battery, which can easily cause metal particles or other electrode material particles to fall off during the nested stacking. These particles adhere to the normal electrode area, causing the electrodes to be continuously corroded during subsequent use of the battery, which seriously reduces the battery performance and service life, which will seriously affect the overall quality of the product and the user experience. Therefore, how to accurately detect abnormal bright spot particles inside the battery is the key to ensuring product quality. Since abnormal bright spots can appear anywhere inside the battery and are tiny, in the actual production process, deviations in the placement of the battery and abnormal metal particles in the non-battery area will affect the CT scan results of the battery. It is necessary to accurately find the battery area from each cross-sectional view for detection. The existing detection methods cannot accurately detect all abnormal points, and the detection results will be interfered by the non-battery area during the detection process. This requires higher accuracy and intelligence for the abnormal bright spots of the positive and negative electrodes inside the battery. Not only can the abnormal bright spots be accurately detected, but also faster speed and robustness are required, and accurate recognition capabilities are required for different models of batteries and different forms of defects. Summary of the invention

[0003] In view of the existing deficiencies, the present invention provides a method, device and medium for detecting tiny foreign particles inside a battery.

[0004] The technical solution adopted by the present invention to solve the technical problem is: a method for detecting tiny foreign particles inside a battery,

[0005] Obtain slice images of different surfaces and cross sections after CT scanning the battery;

[0006] Using deep learning image segmentation method to classify slice images and obtain the battery contours in each slice image;

[0007] Each slice image is eroded outward by several pixels based on its battery outline to form a new outline, and the region of interest is delineated by the minimum circumscribed rectangle of the new outline;

[0008] The deep learning target detection algorithm is used to detect the region of interest and obtain the detection results.

[0009] Preferably, the threshold value of the battery contour segmentation is 0.5.

[0010] Preferably, the pixels eroded outwards are 2 pixels, and the area outside the region of interest is filled with pixels with a value of 0.

[0011] Preferably, the detection results include qualified and abnormal, and the deep learning target detection algorithm outputs a bounding box and a corresponding confidence level for each abnormal target point, and eliminates target points with a confidence level less than 0.6. The target points that are not eliminated in the abnormality are calculated using an area overlap algorithm and duplicate boxes are removed.

[0012] Preferably, the abnormal area where the target points not eliminated in the abnormality are located is divided into foreground and background by a clustering algorithm, and the target points not eliminated in the abnormality are divided into unqualified and pending by analyzing the gray thresholds of the foreground and background.

[0013] Preferably, the unqualified and pending classifications are classified using the mean difference between the grayscale thresholds of the foreground and the background as a classification standard.

[0014] Preferably, the abnormal detection result is back-calculated into the slice image before the clustering algorithm is calculated.

[0015] Preferably, the slice images include a frontal slice image, a cross-sectional slice image and a longitudinal slice image.

[0016] A device for detecting tiny foreign particles inside a battery, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is used to execute the computer program stored in the memory to implement a method for detecting tiny foreign particles inside a battery as described in any of the preceding items.

[0017] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the method for detecting tiny foreign particles inside a battery as described in any of the preceding items.

[0018] The beneficial effects of the present invention are as follows: the present invention first uses a deep learning image segmentation algorithm to extract the contours of each surface of the battery, fills and intercepts the region of interest (ROI area) of the battery contour in the image, and then uses a deep learning target detection algorithm to perform abnormal bright spot detection on the intercepted image to determine the detection result of the battery. It can adapt to the continuous changes in the shape and color of the battery contour in the CT scan, avoids the interference of non-battery areas on the detection results, and greatly improves the recognition accuracy and speed. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flowchart of a detection method according to an embodiment of the present invention;

[0020] Figure 2 is a pending detection result diagram detected by an embodiment of the present invention; DETAILED DESCRIPTION

[0021] In order to more clearly illustrate the purpose, technical solutions and advantages of the embodiments of the present invention, the present invention will be further described below in conjunction with the accompanying drawings and embodiments for a clear and complete description. Obviously, the embodiments described are partial embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work belong to the protection scope of the present invention.

[0022] Embodiments of the present invention Figure 1 As shown in the figure, a method for detecting tiny foreign particles inside a battery is

[0023] Obtain slice images of different surfaces and cross sections of the battery after CT scanning, that is, first scan the battery using CT scanning (computed tomography), and then slice different end faces or cross sections along different directions of the three-dimensional image obtained by the CT scanning to obtain slice images of the two-dimensional plane;

[0024] The deep learning image segmentation method is used to classify the slice images and obtain the battery contour in each slice image; at this time, the deep learning image segmentation method classifies the slice images according to the surfaces they are intercepted, and the slice images include front slice images, cross-sectional slice images and longitudinal slice images, that is, two-dimensional images of the front, cross-sectional and longitudinal surfaces are intercepted from the three-dimensional image obtained by CT scanning, and then the battery contour is obtained from the slice images of these surfaces. At this time, the threshold of battery contour segmentation is 0.5; the deep learning image segmentation method uses a convolutional neural network to extract features from the slice images, outputs the probability score of each category (the sum of the three probability scores is 1) and the probability map of the battery contour, and takes the category with the largest probability value as the interface category, thereby realizing automatic classification of the cross section (front, cross-sectional, longitudinal section) and battery contour extraction, where the threshold of contour segmentation on the probability map is 0.5, and the pixel value is set to 255 if it is greater than 0, otherwise it is 0;

[0025] Each slice image is eroded outward by several pixels based on its battery outline to form a new outline, and the region of interest is delineated by the minimum circumscribed rectangle of the new outline; the outward erosion is achieved using a traditional algorithm, that is, by sliding the structural element (i.e., a non-zero convolution with a convolution kernel of 2) on the image, as long as the structural element intersects with the battery outline in the image, the central pixel value is set as the foreground value, and this operation will make the white area in the image larger and the black area smaller. If a pixel's neighborhood contains any white pixels, the pixel will be expanded to white; 2 pixels are selected for outward erosion, and the area outside the region of interest is filled with a value of 0, that is, the area outside the region of interest (non-battery area) is filled with a black background to ensure that the battery is in the area to be detected and there is no other interference, so as to speed up the detection speed and improve the detection accuracy;

[0026] The deep learning target detection algorithm is used to detect the region of interest to obtain the detection result. The deep learning target detection algorithm uses a convolutional neural network to identify abnormal bright spots in the region of interest, determine their locations, and output the bounding box and confidence of the abnormal bright spots, that is, output the center point coordinates, length, width, confidence and category probability matrix of each abnormal region (the dimensions are (1, 2), and the sum is 1). If no abnormal bright spots are identified in the region of interest, the detection result is qualified, and if abnormal bright spots are identified, they are judged as abnormal and the next step of confirmation is carried out. That is to say, the detection results include qualified and abnormal results. For the abnormal detection results, the deep learning target detection algorithm outputs a bounding box and a corresponding confidence for each abnormal target point, and eliminates the target points with a confidence less than 0.6, that is, the confidence (conf) less than 0.6 is judged as qualified (OK), and the target point with a confidence greater than or equal to 0.6 is judged as abnormal; at this time, the target points that are not eliminated in the abnormality use the area overlap algorithm to calculate and remove the repeated boxes between each other, that is, the overlapping parts of the bounding boxes of the target points. The area overlap algorithm is implemented using existing methods, such as using OpenCV templates to calculate and remove. The abnormal area where the target points that are not eliminated in the abnormality are located is divided into foreground and background through a clustering algorithm. For example, the Kmeans clustering algorithm is used to quantify the grayscale value of the entire abnormal bright spot area, so as to determine the type of battery detection result. The target points that are not eliminated in the abnormality are divided into unqualified and pending through the analysis of the grayscale thresholds of the foreground and background; unqualified and pending use the mean difference of the grayscale thresholds of the foreground and background as the division standard. When the slice image is divided into a front slice image, a cross-sectional slice image, and a longitudinal slice image, the grayscale threshold of the unqualified (NG) front slice image is 42.77, and the grayscale threshold of the pending (Unsure) is 9.57; the NG grayscale threshold of the cross-sectional slice image is 50, and the Unsure grayscale threshold is 4; the NG grayscale threshold of the longitudinal slice image is 46.6, and the Unsure grayscale threshold is 11.8.

[0027] At the same time, the abnormal detection results are back-calculated into the slice image before the clustering algorithm is calculated, so as to improve the accuracy of detection.

[0028] A device for detecting tiny foreign particles inside a battery, comprising a memory and a processor, wherein a computer program is stored in the memory, and the processor is used to execute the computer program stored in the memory to implement a method for detecting tiny foreign particles inside a battery as described in any of the preceding items.

[0029] A computer-readable storage medium stores a computer program. The computer-readable storage medium may include: any entity or device, medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, and software distribution medium, etc., which can carry the computer program code. The computer program code can be in any existing code form. When the computer program is executed by a processor, the method for detecting tiny foreign particles inside a battery as described in any of the preceding items is implemented.

[0030] Although the present invention has been described in detail above with general descriptions and specific embodiments, it is obvious to those skilled in the art that some modifications or improvements can be made on the basis of the present invention. Therefore, these modifications or improvements made on the basis of not departing from the spirit of the present invention all belong to the scope of protection claimed by the present invention.

Claims

1. A method for detecting tiny foreign particles inside a battery, characterized in that: Obtain slice images of different surfaces and cross sections after CT scanning the battery; Using deep learning image segmentation method to classify slice images and obtain the battery contours in each slice image; Each slice image is eroded outward by several pixels based on its battery outline to form a new outline, and the region of interest is delineated by the minimum circumscribed rectangle of the new outline; The deep learning target detection algorithm is used to detect the region of interest and obtain the detection results.

2. The method for detecting tiny foreign particles inside a battery according to claim 1, characterized in that : The threshold value of the battery contour segmentation is 0.

5.

3. The method for detecting tiny foreign particles inside a battery according to claim 1, characterized in that : The pixels eroded outward are 2 pixels, and the area outside the area of ​​interest is filled with pixels with a value of 0.

4. The method for detecting tiny foreign particles inside a battery according to claim 1, characterized in that :The detection results include qualified and abnormal. The deep learning target detection algorithm outputs a bounding box and a corresponding confidence level for each abnormal target point, and removes target points with a confidence level less than 0.

6. The target points that are not removed from the abnormality are calculated using an area overlap algorithm and the duplicate frames between them are removed.

5. The method for detecting tiny foreign particles inside a battery according to claim 4, characterized in that: The abnormal area where the target points not eliminated in the abnormality are located is divided into foreground and background by a clustering algorithm, and the target points not eliminated in the abnormality are divided into unqualified and pending by analyzing the gray thresholds of the foreground and background.

6. The method for detecting tiny foreign particles inside a battery according to claim 5, characterized in that The unqualified and pending items are divided by using the mean difference of the grayscale thresholds of the foreground and background as the classification criteria.

7. The method for detecting tiny foreign particles inside a battery according to claim 4, characterized in that : The abnormal detection results are back-calculated into the slice image before the clustering algorithm is calculated.

8. The method for detecting tiny foreign particles inside a battery according to claim 1, characterized in that: The slice images include a front slice image, a cross-sectional slice image, and a longitudinal slice image.

9. A device for detecting tiny foreign particles inside a battery, characterized in that : comprising a memory and a processor, wherein the memory stores a computer program, and the processor is used to execute the computer program stored in the memory, so as to implement the method for detecting tiny foreign particles inside a battery as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for detecting tiny foreign particles inside a battery according to any one of claims 1 to 8 is implemented.

Images