基于改进YOLOv4的柱形锂电池缺陷检测研究  

Surface Defect Detection of Cylindrical Lithium Battery Based on Improved YOLOv4

作　　者：田杰[1] 胡昊[1] 周华健 邹润 TIAN Jie;HU Hao;ZHOU Huajian;ZOU Run(School of Mechanical Engineering,Hefei University of Technology,Hefei 230009,China)

机构地区：[1]合肥工业大学机械工程学院,合肥230009

出　　处：《机械工程师》2023年第3期16-18,共3页Mechanical Engineer

摘　　要：为了实现柱形锂电池缺陷检测的实时性与高精度,提出一种基于改进YOLOv4的柱形锂电池表面缺陷检测算法。将主干网络由CSPDarkNet53替换为轻量化网络Mobile Netv1,使用K-means++算法对锂电池缺陷数据集先验框进行重新聚类,同时构建新的注意力机制ECSA模块关注重要信息。改进后的模型检测精度与检测速度均得到提升。In order to achieve real-time and high-precision defect detection of cylindrical lithium battery, this paper proposes a surface defect detection algorithm of cylindrical lithium battery based on improved YOLOv4. The backbone network CSPDarkNet53 is replaced by MobileNetv1, which is a lightweight network. K-means ++ algorithm is used to recluster the prior frames of lithium battery defect data set. At the same time, a new attention mechanism ECSA module is constructed to pay attention to important information. The detection accuracy and detection speed of the improved model are improved.

关 键 词：柱形锂电池 缺陷检测 深度学习 YOLOv4 

分 类 号：TM912[电气工程—电力电子与电力传动]