软包装锂离子电池表面凹坑缺陷检测方法  

Surface pit defects detecting method of pouch Li-ion battery

作　　者：何涛[1,2] 张成娟 雷卓 王正家 HE Tao;ZHANG Chengjuan;LEI Zhuo;WANG Zhengjia(School of Mechanical Engineering,Hubei University of Technology,Wuhan,Hubei 430068,China;Hubei Provincial Key Laboratory of Modern Manufacturing Engineering,Wuhan,Hubei 430068,China)

机构地区：[1]湖北工业大学机械工程学院,湖北武汉430068 [2]现代制造工程湖北省重点实验室,湖北武汉430068

出　　处：《电池》2024年第3期358-363,共6页Battery Bimonthly

基　　金：国家自然科学基金项目(51275158)。

摘　　要：软包装锂离子电池表面凹坑缺陷对比度低、缺陷区域过小且存在反光,传统方法很难进行准确检测。提出一种基于图像增强和改进DeepLabV3网络的软包装锂离子电池表面凹坑缺陷检测方法。通过分析表面凹坑缺陷图像特征,采用图像增强算法对图像进行预处理,以增强凹坑缺陷对比度。对DeepLabV3网络进行改进,使用ResNet101作为特征提取网络,同时引入位置注意力模块,使得模型更加关注于凹坑缺陷相关特征,提升网络的检测精度。改进后的网络在自制数据集上的平均交并比达到85.98%,缺陷检测准确率达到98.33%。The traditional method is difficult to accurately detect the surface pit defects of pouch Li-ion battery due to their low contrast,small defect area and reflection.A method for detecting the pit defect on the surface of the pouch Li-ion battery based on image enhancement and improved DeepLabV3 network is proposed.The image features of pit defects on the surface are analyzed,the image enhancement algorithm is used to preprocess the image to enhance the contrast of pit defects.The DeepLabV3 network is improved,ResNet101 is used as the feature extraction network.The position attention module is introduced to make the model pay more attention to the related features of pit defects and improve the detection accuracy of the network.The median intersection-overunion of the improved network on the self-made data set reaches 85.98%,the accuracy of defect detection reaches 98.33%.

关 键 词：图像增强 深度学习 DeepLabV3 缺陷检测 软包装锂离子电池 表面凹坑 

分 类 号：TM912.9[电气工程—电力电子与电力传动]