改进YOLOv8的输电线路异物检测方法  

Improved YOLOv8 foreign object detection method for transmission lines

作　　者：易磊 黄哲玮 易雅雯 Yi Lei;Huang Zhewei;Yi Yawen(School of Electrical&Electronic Engineering,Huazhong University of Science and Technology,Wuhan 430074,China)

机构地区：[1]华中科技大学电气与电子工程学院,武汉430074

出　　处：《电子测量技术》2024年第15期125-134,共10页Electronic Measurement Technology

基　　金：教育部2020年第二批新工科研究与实践项目(E-NYDQHGC20202219)资助。

摘　　要：针对无人机对输电线路异物检测准确度有限,模型计算复杂度高、计算速度有限的问题,提出一种改进YOLOv8的输电线路异物检测方法SC-YOLO。该方法引入StarNet以构造C2f_Star模块实现Neck网络轻量化,有效降低模型参数量与计算量,同时通过增加特征空间维度提升Neck部分特征提取能力;在骨干网络输出特征图后添加卷积注意力融合模块,提升骨干网络对输入特征图的初步特征提取能力,增强模型整体检测效果;将原检测头替换为动态检测头,提升模型对不同输入的动态调整能力与对关键信息的关注程度;使用WIoU作为边界框损失函数,EMA-Slide Loss作为分类损失函数,提升模型泛化能力与检测性能。实验结果表明,提出的SC-YOLO计算量较原始模型下降8.02%,mAP提升1.4个百分点,达到了95.2%的检测精度,在降低模型计算复杂度的同时实现了较高的检测准确率,具有高可行性与实用性。Aiming at the problems of limited accuracy of UAV detection of foreign objects on power transmission lines,high model computational complexity and limited computational speed,a power transmission line foreign object detection method SC-YOLO which improves YOLOv8 is proposed.This method introduces StarNet to construct C2f_Star module to realize the lightweight of Neck network,effectively reducing the number of model parameters and calculation amount,and at the same time improves the feature extraction ability of Neck by increasing the dimension of feature space;adds convolution attention fusion module after the backbone network outputs feature map to improve the backbone network′s preliminary feature extraction ability of input feature map,and enhance the overall detection effect of the model;replaces the original detection head with dynamic detection head to improve the model′s dynamic adjustment ability to different inputs and the degree of attention to key information;uses WIoU as the bounding box loss function and EMA-Slide Loss as the classification loss function to improve the model′s generalization ability and detection performance.Experimental results show that the proposed SC-YOLO has 8.02%fewer computational amount than the original model,and mAP is increased by 1.4 percentage points,reaching a detection accuracy of 95.2%.RC-YOLO reduces the computational complexity while achieving a high detection accuracy,and is highly feasible and practical.

关 键 词：输电线异物检测 YOLOv8 StarNet 卷积注意力融合模块 动态检测头 WIOU损失函数 EMA-Slide Loss损失函数 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TN60[自动化与计算机技术—计算机科学与技术]