Lithium battery pole piece number statistical method and system based on multi-task segmentation

Abstract

The invention provides a lithium battery pole piece number statistical method and system based on multi-task segmentation, and relates to the technical field of lithium battery detection. The method comprises the following steps: performing feature extraction on a target lithium battery X-ray image to obtain a target lithium battery feature image; performing feature segmentation on the target lithium battery feature image to obtain a plurality of feature daughters; performing feature fusion on the plurality of feature daughters to obtain a lithium battery pole piece feature image; and taking the lithium battery pole piece feature image as an acceleration parameter for lithium battery pole piece detection through the detection model to obtain a lithium battery pole piece detection result, and extracting the number information of the lithium battery pole pieces. The lithium battery pole piece feature image is obtained after feature extraction, feature segmentation and feature fusion are carried out on the image, the lithium battery pole piece detection and positioning efficiency and precision of the detection model are improved through the lithium battery pole piece feature image, the lithium battery pole piece number information is obtained, and the lithium battery quality detection efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of lithium battery detection, and particularly to a method and system for counting the number of lithium battery electrode sheets based on multi-task segmentation. Background Art

[0002] Due to advantages such as high specific energy, stable performance, and long service life, lithium batteries are widely used in technical fields such as intelligent electronic devices and electric vehicles.

[0003] Detecting the position and number of electrode sheets is one of the steps for quality inspection of lithium batteries. Generally, non-destructive testing techniques are used for quality inspection of lithium batteries. One of the current methods for detecting the quality of lithium batteries is to collect internal images of lithium batteries through an X-ray device and further analyze the collected images manually to obtain quality inspection information of the lithium batteries. This method has the disadvantage of low detection efficiency. Summary of the Invention

[0004] In view of the above problems, the present application provides a method and system for counting the number of lithium battery electrode sheets based on multi-task segmentation, which counts the number of electrode sheets of a lithium battery based on deep learning technology and improves the efficiency of lithium battery quality inspection.

[0005] As one aspect of the present application, there is provided a method for counting the number of lithium battery electrode sheets based on multi-task segmentation, including: Obtaining an X-ray image of a target lithium battery, inputting the X-ray image of the target lithium battery into a feature extraction model for feature extraction to obtain a feature image of the target lithium battery; Inputting the feature image of the target lithium battery into a feature segmentation model for feature segmentation to obtain a plurality of feature sub-volumes; Inputting the plurality of feature sub-volumes into a feature fusion model for feature fusion to obtain a feature image of lithium battery electrode sheets; Inputting the X-ray image of the target lithium battery into a detection model, using the feature image of lithium battery electrode sheets as an acceleration parameter for detecting lithium battery electrode sheets through the detection model to obtain a detection result of lithium battery electrode sheets, and extracting the number information of lithium battery electrode sheets from the detection result of lithium battery electrode sheets.

[0006] Further, the step of inputting the X-ray image of the target lithium battery into a feature extraction model for feature extraction to obtain a feature image of the target lithium battery includes: Performing grayscale processing on the X-ray image of the target lithium battery input into the feature extraction model to obtain a grayscale image of the target lithium battery; Determine an increment step and an initial feature box according to the size information of the target lithium battery grayscale image. Take the center point of the target lithium battery grayscale image as the center point of the initial feature box, and obtain the grayscale histogram corresponding to the initial feature box area image; Based on the increment step, sequentially increase the initial feature box, and obtain the grayscale histograms corresponding to multiple enlarged initial feature box area images; Determine a target segmentation threshold according to multiple grayscale histograms, and perform binary processing on the target lithium battery grayscale image based on the target segmentation threshold to obtain the target lithium battery feature image.

[0007] Further, the determining the increment step and the initial feature box according to the size information of the target lithium battery grayscale image includes: Extract the target size from the size information of the target lithium battery grayscale image, and reduce the target size based on a preset ratio to obtain the initial feature box; Calculate the size difference information between the target size and the initial feature box, and determine the increment step according to the size difference information and the preset increment times; The determining the target segmentation threshold according to multiple grayscale histograms includes: For any grayscale histogram, determine the first peak and the second peak, and use the valley value between the first peak and the second peak as the segmentation threshold corresponding to the grayscale histogram; Map multiple grayscale histograms to the same curve graph, screen out the segmentation thresholds located between any first peak and any second peak to obtain a target threshold set, and use the element with the least number of corresponding pixel points in the target threshold set as the target segmentation threshold.

[0008] Further, the inputting the target lithium battery feature image into a feature segmentation model for feature segmentation to obtain multiple feature sub-bodies includes: The feature segmentation model performs feature segmentation on the target lithium battery feature image based on the OpenCV contour extraction algorithm to obtain multiple independent feature sub-bodies.

[0009] Further, the inputting multiple feature sub-bodies into a feature fusion model for feature fusion to obtain a lithium battery electrode feature image includes: For multiple feature sub-bodies, count the number of pixel points corresponding to each feature sub-body, and determine the central reference point of each feature sub-body according to the contour information of each feature sub-body; Sort multiple feature sub-bodies according to the number of pixel points corresponding to each feature sub-body; Map multiple said feature sub-bodies to the target lithium battery feature image. Taking the center point of the target lithium battery grayscale image as the center point of the initial feature frame, establish a feature set based on each said feature sub-body whose central reference point is located in the initial feature frame area; Determine a segmentation reference range according to the number of pixel points corresponding to each said feature sub-body in the feature set. Screen multiple sorted said feature sub-bodies according to the segmentation reference range to obtain multiple target feature sub-bodies, and form the lithium battery electrode sheet feature image with the multiple target feature sub-bodies.

[0010] Further, the step of inputting the target lithium battery X-ray image into the detection model, using the lithium battery electrode sheet feature image as an acceleration parameter for lithium battery electrode sheet detection by the detection model to obtain the lithium battery electrode sheet detection result, further includes: The detection model is a pre-trained YOLO-v3 model for detecting lithium battery electrode sheets. Using the position information of multiple target feature sub-bodies in the lithium battery electrode sheet feature image as prior information on the position of the lithium battery electrode sheet to accelerate the process of the detection model detecting the lithium battery electrode sheet.

[0011] As another aspect of the present application, there is provided a lithium battery electrode sheet quantity statistics system based on multi-task segmentation, including: An image acquisition module, configured to acquire a target lithium battery X-ray image through an X-ray device; A feature extraction module, configured to extract features from the target lithium battery X-ray image to obtain a target lithium battery feature image; A feature segmentation module, configured to perform feature segmentation on the target lithium battery feature image to obtain multiple feature sub-bodies; A feature fusion module, configured to perform feature fusion on multiple said feature sub-bodies to obtain a lithium battery electrode sheet feature image; A quantity statistics module, configured to process the lithium battery electrode sheet feature image and the target lithium battery X-ray image through a detection model to obtain a lithium battery electrode sheet detection result, and extract lithium battery electrode sheet quantity information from the lithium battery electrode sheet detection result.

[0012] Further, for the detection model, it further includes: the detection model is a YOLO-v3 model for detecting lithium battery electrode sheets.

[0013] The beneficial effects of the present invention are as follows: By performing feature extraction, feature segmentation, and feature fusion on the X-ray image of the target lithium battery, a feature image of the lithium battery electrode is obtained. The efficiency and accuracy of detecting and positioning the lithium battery electrode are improved through the feature image of the lithium battery electrode, and the number information of the lithium battery electrode is extracted from the detection results of the detection model, thereby improving the efficiency of lithium battery quality detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on the structures shown in these drawings.

[0015] Figure 1 It is a schematic flowchart of a method for counting the number of lithium battery electrodes based on multi-task segmentation in an embodiment of the present invention.

[0016] Figure 2 It is a schematic structural diagram of a method for counting the number of lithium battery electrodes based on multi-task segmentation in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] In order to make the objectives, technical solutions, and advantages of the present application more clearly understood, the following further details some embodiments of the present application in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. However, those of ordinary skill in the art can understand that in the various embodiments of the present application, many technical details are provided for the convenience of readers to better understand the present application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solutions required to be protected by the present application can still be implemented.

[0018] Embodiment 1: Refer to Figure 1 , Embodiment 1 of the present invention provides a method for counting the number of lithium battery electrodes based on multi-task segmentation, including: S1. Obtain an X-ray image of a target lithium battery, and input the X-ray image of the target lithium battery into a feature extraction model for feature extraction to obtain a feature image of the target lithium battery; In this embodiment, the X-ray image of the target lithium battery is an internal structure image of the lithium battery to be detected obtained by X-ray shooting. The function of the feature extraction model is to perform preprocessing such as grayscale conversion and noise reduction on the X-ray image of the target lithium battery to achieve preliminary feature extraction of the X-ray image of the target lithium battery.

[0019] S2. Input the target lithium battery characteristic image into the feature segmentation model for feature segmentation to obtain multiple feature sub - bodies; In this embodiment, after obtaining the target lithium battery characteristic image, the feature segmentation model specifically performs feature segmentation on the target lithium battery characteristic image through edge extraction technology, thereby obtaining multiple independent feature sub - bodies; specifically, in this embodiment, the feature segmentation model performs feature segmentation on the target lithium battery characteristic image based on the OpenCV contour extraction algorithm to obtain multiple independent feature sub - bodies.

[0020] S3. Input the multiple feature sub - bodies into the feature fusion model for feature fusion to obtain the lithium battery electrode characteristic image; In this embodiment, the feature fusion model processes the multiple feature sub - bodies, analyzes and obtains the feature sub - bodies close to the lithium battery electrode among the multiple feature sub - bodies and performs feature fusion to obtain the lithium battery electrode characteristic image.

[0021] S4. Input the target lithium battery X - ray image and the lithium battery electrode characteristic image into the detection model to obtain the lithium battery electrode detection result, and extract the lithium battery electrode quantity information from the lithium battery electrode detection result.

[0022] In this embodiment, the detection model is specifically the pre - trained YOLO - v3 model for detecting lithium battery electrodes. The trained YOLO - v3 model is used for lithium battery electrode detection, and the lithium battery electrode characteristic image is used as prior information and input into the detection model together with the target lithium battery X - ray image. The position information of multiple target feature sub - bodies in the lithium battery electrode characteristic image is used as an acceleration parameter to obtain the lithium battery electrode detection result. As a one - stage model, the YOLO - v3 model can quickly achieve target detection, but the detection accuracy is insufficient and it cannot accurately locate the detected target well. The lithium battery electrode characteristic image obtained through multi - task operations such as feature extraction, feature segmentation, and feature fusion of the target lithium battery X - ray image can assist the YOLO - v3 model in lithium battery electrode detection and positioning, which can further improve the speed and accuracy of the YOLO - v3 model in detecting lithium battery electrodes. The position information and quantity information of the lithium battery electrodes are extracted from the lithium battery electrode detection result to complete the quantity statistics of the lithium battery electrodes.

[0023] In one of the embodiments, for step S1, inputting the target lithium battery X - ray image into the feature extraction model for feature extraction to obtain the target lithium battery characteristic image specifically includes: Perform grayscale processing on the target lithium battery X - ray image input into the feature extraction model to obtain the target lithium battery grayscale image; In this embodiment, after grayscale processing the X-ray image of the target lithium battery, denoising and enhancement processing can also be performed on the grayscale image of the target lithium battery. For example, denoising can be performed through a filter, etc., which will not be elaborated here.

[0024] Determine the increment step size and the initial feature box according to the size information of the grayscale image of the target lithium battery. Use the center point of the grayscale image of the target lithium battery as the center point of the initial feature box, and obtain the grayscale histogram corresponding to the image in the initial feature box area. In this embodiment, the target size is extracted from the size information of the grayscale image of the target lithium battery, that is, the length and width information of the grayscale image of the target lithium battery. Then, the size of the initial feature box is determined according to a preset ratio. Specifically, taking the preset ratio as 0.7 as an example, multiply 0.7 by the target size of the grayscale image of the target lithium battery to obtain the initial feature box. Then, calculate the size difference information between the target size and the size of the initial feature box, and determine the increment step size according to the size difference information and the preset number of increments. Suppose the size of the grayscale image of the target lithium battery is 500*300, then the size of the initial feature box is 350*210, and the size difference information is (500 - 350, 300 - 210), that is, (150, 90). When the preset number of increments is 30, the increment step sizes are 5 and 3, that is, the long side of the box increases by 5 and 3 respectively during each increment process.

[0025] Successively increase the initial feature box based on the increment step size, and obtain the grayscale histograms corresponding to the images in multiple enlarged initial feature box areas. Determine the target segmentation threshold according to multiple grayscale histograms, and perform binarization processing on the grayscale image of the target lithium battery based on the target segmentation threshold to obtain the feature image of the target lithium battery.

[0026] In this embodiment, the grayscale histograms corresponding to the initial feature box and the incremented initial feature box areas are obtained respectively. By analyzing multiple grayscale histograms, the target segmentation threshold can be determined. Compared with directly analyzing the histogram of the grayscale image of the target lithium battery, the target segmentation threshold for binarizing the grayscale image of the target lithium battery can be determined more accurately, reducing the influence of background information.

[0027] Specifically, the target segmentation threshold according to multiple grayscale histograms includes: For any grayscale histogram, determine the first peak value and the second peak value, and use the valley value between the first peak value and the second peak value as the segmentation threshold corresponding to this grayscale histogram. Map multiple grayscale histograms to the same curve graph, screen out the segmentation thresholds located between any first peak value and any second peak value to obtain a set of target thresholds, and use the element with the smallest number of corresponding pixel points in the set of target thresholds as the target segmentation threshold.

[0028] It should be noted that when multiple grayscale histograms are respectively converted into curve graphs and mapped onto the same curve graph, for the segmentation threshold corresponding to each grayscale histogram, the individuals located between any one of the first peaks and any one of the second peaks are screened out, and the optimal individual is selected as the target segmentation threshold. Specifically in this embodiment, the individual corresponding to the least number of pixel points is selected as the target segmentation threshold.

[0029] In one embodiment, multiple feature sub-volumes are input into a feature fusion model for feature fusion to obtain a lithium battery electrode sheet feature image, including: For multiple feature sub-volumes, the number of pixel points corresponding to each feature sub-volume is counted, and the central reference point of each feature sub-volume is determined according to the contour information of each feature sub-volume; Specifically, for any one feature sub-volume, the central reference point of the feature sub-volume can be determined according to the position information of the edge points in the four directions of up, down, left, and right of the feature sub-volume. For example, the center point of the rectangular frame tangent to the edge points in the four directions of up, down, left, and right of the feature sub-volume is used as the central reference point of the feature sub-volume.

[0030] The multiple feature sub-volumes are sorted according to the number of pixel points corresponding to each feature sub-volume; The multiple feature sub-volumes are mapped into the target lithium battery feature image. Taking the center point of the target lithium battery grayscale image as the center point of the initial feature frame, a feature set is established based on each feature sub-volume whose central reference point is located in the initial feature frame area; Specifically, for any one feature sub-volume that has an intersection area with the initial feature frame area, if the central reference point of a certain feature sub-volume is located in the initial feature frame area, then the feature sub-volume is used as an element in the feature set, thereby obtaining the feature set.

[0031] The segmentation reference range is determined according to the number of pixel points corresponding to each feature sub-volume in the feature set, and the sorted multiple feature sub-volumes are screened according to the segmentation reference range to obtain multiple target feature sub-volumes, and the lithium battery electrode sheet feature image is composed of the multiple target feature sub-volumes; Specifically, the feature sub-volumes in the feature set are sorted according to the number of pixel points, a preset screening ratio is set, and a part of the feature sub-volumes with the lowest number of corresponding pixel points in the feature set are screened out according to the screening ratio. The range of the number of pixel points of the remaining feature sub-volumes is used as the segmentation reference range, and the multiple independent feature sub-volumes are screened through the segmentation reference range. Based on all the feature sub-volumes located within the segmentation reference range, a lithium battery electrode sheet feature image is formed, and a lithium battery electrode sheet feature image for improving the detection speed and accuracy of the YOLO-v3 model for lithium battery electrode sheets is obtained.

[0032] Embodiment 2: Refer to Figure 2, on the basis of Embodiment 1, Embodiment 2 of the present invention further provides a lithium battery pole piece quantity statistical system based on multi-task segmentation, including: An image acquisition module, configured to acquire an X-ray image of a target lithium battery through an X-ray device; A feature extraction module, configured to extract features from the X-ray image of the target lithium battery to obtain a feature image of the target lithium battery; A feature segmentation module, configured to segment the features of the feature image of the target lithium battery to obtain a plurality of feature sub-volumes; A feature fusion module, configured to fuse the features of the plurality of feature sub-volumes to obtain a feature image of the lithium battery pole piece; A quantity statistics module, configured to process the feature image of the lithium battery pole piece and the X-ray image of the target lithium battery through a detection model to obtain a detection result of the lithium battery pole piece, and extract the quantity information of the lithium battery pole piece from the detection result of the lithium battery pole piece.

[0033] In one of the embodiments, for the detection model, it further includes: the detection model is a YOLO-v3 model for detecting lithium battery pole pieces.

[0034] It should be understood that those of ordinary skill in the art can make improvements or transformations according to the above description, and all such improvements and transformations should fall within the protection scope of the appended claims of the present invention. The parts not described in detail in this specification belong to the prior art well known to those of ordinary skill in the art.

Claims

1. A lithium battery pole piece counting method based on multi-task segmentation, characterized in that: include: Acquire an X-ray image of a target lithium battery, input the X-ray image of the target lithium battery into a feature extraction model to perform feature extraction, and obtain a feature image of the target lithium battery; Inputting the target lithium battery feature image into a feature segmentation model to perform feature segmentation to obtain a plurality of feature sub-bodies; Inputting the plurality of feature sub-bodies into a feature fusion model for feature fusion to obtain a lithium battery pole piece feature image; The target lithium battery X-ray image is input into the detection model, and the lithium battery pole piece characteristic image is used as an acceleration parameter for lithium battery pole piece detection through the detection model to obtain the lithium battery pole piece detection result, and the lithium battery pole piece quantity information is extracted from the lithium battery pole piece detection result.

2. A method for counting the number of lithium battery pole pieces based on multi-task segmentation as claimed in claim 1, characterized in that: The step of inputting the target lithium battery X-ray image into a feature extraction model to extract features and obtain a target lithium battery feature image includes: Grayscale the target lithium battery X-ray image input into the feature extraction model to obtain a target lithium battery grayscale image; Determine the incremental step size and the initial feature frame according to the size information of the target lithium battery grayscale image, take the center point of the target lithium battery grayscale image as the center point of the initial feature frame, and obtain the grayscale histogram corresponding to the image in the initial feature frame area; The initial feature frame is enlarged in sequence based on the increasing step size, and the grayscale histograms corresponding to the images of the enlarged initial feature frame regions are obtained; A target segmentation threshold is determined according to a plurality of grayscale histograms, and a target lithium battery grayscale image is binarized based on the target segmentation threshold to obtain the target lithium battery characteristic image.

3. A lithium battery pole piece counting method based on multi-task segmentation as claimed in claim 2, characterized in that: The step of determining the incremental step size and the initial feature frame according to the size information of the target lithium battery grayscale image comprises: Extracting a target size from the size information of the target lithium battery grayscale image, and reducing the target size based on a preset ratio to obtain an initial feature frame; Calculating the size difference information of the target size and the initial feature frame, and determining the increment step length according to the size difference information and a preset increment number; The target segmentation thresholds according to multiple grayscale histograms include: For any grayscale histogram, determine the first peak value and the second peak value, and use the valley value between the first peak value and the second peak value as the segmentation threshold corresponding to the grayscale histogram; A plurality of grayscale histograms are mapped to the same curve graph, and a segmentation threshold between any first peak and any second peak is screened out to obtain a target threshold set, and the element with the least number of corresponding pixels in the target threshold set is used as the target segmentation threshold.

4. A method for counting the number of lithium battery pole pieces based on multi-task segmentation as claimed in claim 3, characterized in that: The step of inputting the target lithium battery characteristic image into a feature segmentation model for feature segmentation to obtain a plurality of characteristic sub-bodies comprises: the feature segmentation model performs feature segmentation on the target lithium battery characteristic image based on an OpenCV contour extraction algorithm to obtain a plurality of independent characteristic sub-bodies.

5. A lithium battery pole piece counting method based on multi-task segmentation as claimed in claim 4, characterized in that: The step of inputting the plurality of feature sub-bodies into a feature fusion model for feature fusion to obtain a lithium battery pole piece feature image includes: For the plurality of feature sub-volumes, counting the number of pixel points corresponding to each feature sub-volume, and determining the central reference point of each feature sub-volume according to the contour information of each feature sub-volume; sorting the plurality of feature sub-volumes according to the number of pixels corresponding to each of the feature sub-volumes; Mapping the plurality of feature sub-bodies to the target lithium battery feature image, taking the center point of the target lithium battery grayscale image as the center point of an initial feature frame, and establishing a feature set based on each feature sub-bodies whose center reference point is located in the initial feature frame area; The segmentation reference range is determined according to the number of pixel points corresponding to each feature sub-body in the feature set, and the sorted multiple feature sub-bodies are screened according to the segmentation reference range to obtain multiple target feature sub-bodies, which form the lithium battery pole piece feature image.

6. A lithium battery pole piece counting method based on multi-task segmentation as claimed in claim 5, characterized in that: The target lithium battery X-ray image is input into the detection model, and the lithium battery pole piece feature image is used as an acceleration parameter for lithium battery pole piece detection through the detection model to obtain the lithium battery pole piece detection result, and also includes: the detection model is a pre-trained YOLO-v3 model for detecting lithium battery pole pieces, and the position information of multiple target feature sub-bodies in the lithium battery pole piece feature image is used as lithium battery pole piece position prior information to accelerate the process of detecting lithium battery pole pieces by the detection model.

7. A lithium battery pole piece counting system based on multi-task segmentation, characterized in that: The system is used to implement a method for counting the number of lithium battery pole pieces based on multi-task segmentation as described in any one of claims 1 to 6, comprising: An image acquisition module, used to acquire X-ray images of target lithium batteries through X-ray equipment; A feature extraction module is used to extract features from the target lithium battery X-ray image to obtain a target lithium battery feature image; A feature segmentation module is used to perform feature segmentation on the target lithium battery feature image to obtain a plurality of feature sub-bodies; A feature fusion module is used to perform feature fusion on a plurality of the feature sub-bodies to obtain a feature image of a lithium battery pole piece; The quantity statistics module is used to process the lithium battery pole piece characteristic image and the target lithium battery X-ray image through a detection model to obtain a lithium battery pole piece detection result, and extract the lithium battery pole piece quantity information from the lithium battery pole piece detection result.

8. A lithium battery pole piece counting system based on multi-task segmentation as claimed in claim 7, characterized in that: For the detection model, it also includes: the detection model is a YOLO-v3 model used to detect lithium battery pole pieces.

Images