Development Of A Soft-pack Lithium Battery Detection System Based On Machine Vision

With the increasingly extensive application of mobile electronic devices,the demand for lithium batteries is increasing day by day with its advantages of large capacity,small volume and low density.The production of lithium batteries has basically realized automation.However,in the production process,scratches,pits,dents,pinholes,exposed aluminum and other defects will be formed on the surface of lithium battery shell due to various factors,which will not only hinder its appearance,but also affect the performance of lithium battery in serious cases.In industrial detection,the result of manual detection of lithium battery surface defects is not good,which is time-consuming and labor-consuming.Based on deep learning,this thesis studies the surface defect detection of soft-pack lithium battery,optimizes the object detection algorithm yolov5 s,and then puts forward a new network structure.The main work is as follows:（1）Collect a certain number of surface defect images of soft-pack lithium battery,and then the surface defects of lithium battery are classified according to the geometric characteristics and forms of defects.Due to the problem of insufficient data,the lithium battery defect image is expanded through random crop and data augmentation.The image is labeled with the data labeling software Label Img,and finally the data set of the surface defect image of soft-pack lithium battery is obtained.（2）In order to select the best defect detection algorithm of soft-pack lithium battery,the surface defects of soft-pack lithium battery are detected based on three object detection algorithms: Faster RCNN YOLOv3 and YOLOv5 s.The experimental results show that among the three algorithms,the map of YOLOv5 s is 93.1%,which is basically the same as that of YOLOv3 and 22.13% higher than that of Faster RCNN.At the same time,the FPS of YOLOv5 s is 92,followed by YOLOv3,whose FPS is 56,and finally Faster RCNN,whose FPS is 22.Under the condition of ensuring accuracy,the FPS of YOLOv5 s is much higher than that of YOLOv3.Therefore,YOLOv5 s is determined as the main research object of surface defect detection algorithm of soft-pack lithium battery.（3）Firstly,a lightweight network model based on G-YOLOv5 s is studied and applied to defect detection task.Experiments show that the module can maintain a certain detection accuracy and improve the flexibility of the algorithm while reducing a large number of parameters and calculations.However,in order to further improve the detection accuracy of lightweight networks,GCA-YOLOv5 s detection algorithm is proposed next.The experimental results show that the map of the improved GCA-YOLOv5 s algorithm is 93.1%and the FPS is 93,which are slightly higher than that of YOLOv5 s.Its Precision and Recall are 96.1% and 87% respectively,both higher than that of YOLOv5 s,the parameter quantity of YOLOv5 s is 7260488,and the parameter quantity of GCA-YOLOv5 s is only 2479176,which is far less than that of YOLOv5 s.This method can meet the requirements of high precision and high timeliness in actual production quality inspection.At the same time,its network model occupies less memory and is easier to be embedded into small equipment.Finally,GCA-YOLOv5 s is loaded into the system,and the interface of soft-pack lithium battery surface defect detection system is developed to visually show the function of lithium battery surface defect detection.（4）The software and hardware platform is designed according to the defect feature detection requirements of soft-pack lithium battery.Firstly,the appropriate camera,optical lens,light source,digital light source controller and other devices are selected to build the surface defect detection platform of s soft-pack lithium battery.Then the upper computer software is designed.Finally,GCA-YOLOv5 s is loaded into the system to develop the surface defect detection system of soft packed lithium battery,so as to visually show the function of surface defect detection of lithium battery.