基于改进YOLOv4防震锤的定位识别与丢失检测  

LOCATION RECOGNITION AND LOSS DETECTION BASED ON IMPROVED YOLOV4 SHOCKPROOF HAMMER

作　　者：张元伟 陈春玲[1] 张楠楠 Zhang Yuanwei;Chen Chunling;Zhang Nannan(School of Information and Electrical Engineering,Shenyang Agricultural University,Shenyang 110866,Liaoning,China)

机构地区：[1]沈阳农业大学信息与电气工程学院,辽宁沈阳110866

出　　处：《计算机应用与软件》2025年第3期135-140,161,共7页Computer Applications and Software

基　　金：国家自然科学基金青年科学基金项目(61903264)。

摘　　要：针对高压线路巡检中防震锤的识别定位与丢失检测,提出一种基于改进YOLOv4的算法模型。首先根据收集而来的巡检图像做有目的地数据增强,扩大数据集。然后融入迁移学习思想,在模型训练过程中使用预权重以及进行冻结训练。最后将YOLOv4原始主干特征提取网络替换成轻量型网络MobileNet V2,将深度可分离卷积运用于网络中,大大减少参数量。对实验结果进行对比分析,改进后的算法模型综合性能表现良好,也符合巡检要求。A new algorithm model based on improved YOLOv4 is presented to identify,locate and detect the loss of shockproof hammer in high voltage line inspection.Purposeful data enhancements were made based on the collected patrol images to expand the dataset.The idea of transfer learning was incorporated,and pre-weights and freeze training were used during model training.The YOLOv4 original trunk feature extraction network was replaced by the lightweight network MobileNet V2,and the deep detachable convolution was applied to the network,which greatly reduced the amount of parameters.By comparing and analyzing the experimental results,the improved algorithm model performs well and meets the requirements of patrol inspection.

关 键 词：深度学习 目标检测 防震锤 YOLOv4 MobileNet V2 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]