基于机器视觉的输电线路异物识别及检测算法研究  

Research on Machine Vision Based on Foreign Object Recognition and Detection Algorithm for Transmission Lines

作　　者：刘孝文 闫建豪 LIU Xiaowen;YAN Jianhao(College of Automation and Electronic Engineering,Qingdao University of Science and Technology,Qingdao,Shandong 266061,China)

机构地区：[1]青岛科技大学自动化与电子工程学院,山东青岛266061

出　　处：《自动化应用》2025年第4期70-74,79,共6页Automation Application

摘　　要：旨在解决无人机输电线路巡检中小目标识别的难题。由于小目标存在像素值少、特征不丰富、难提取、易受环境干扰等局限性,容易导致模型漏检、精度低、网络参数量过大等问题。通过对Yolov5目标检测模型进行优化,成功引入了卷积块注意力模块(CBAM)和高效交并比(EIoU)损失函数。实验结果表明,CBAM有效提升了模型对图像中细微特征的关注度;EIoU损失函数显著优化了预测框与真实框之间的匹配度,在处理小目标时表现出色;改进后的模型Precison值增添了1.5%,Reacll值增添了1.9%,mAP@.5值增添了1.2%。在无人机输电线路巡检中联合使用CBAM和EIoU,不仅能提升模型在小目标检测方面的精度和鲁棒性,还能增强其在实际巡检场景中的应用效果。This research aims to address the challenge of small object recognition in unmanned aerial vehicle transmission line inspections.The limitations of small targets,such as their low pixel count,scarcity of features,difficulty in extraction,and susceptibility to environmental interference,often lead to issues such as model missed detections,low accuracy,and excessive network parameter volume.By optimizing the Yolov5 object detection model,we successfully introduced the Convolutional Block Attention Module and the Efficient Intersection over Union(EIoU)loss function.Experimental results demonstrate that the CBAM effectively enhances the model's focus on subtle features in images.The EIoU loss function significantly optimizes the matching degree between predicted bounding boxes and ground truth boxes,exhibiting exceptional performance when processing small targets.The improved model shows an increase of 1.5%in Precision,1.9%in Recall,and 1.2%in mAP@.5.The combined use of CBAM and EIoU in UAV transmission line inspections not only improves the accuracy and robustness of the model in small object detection,but also enhances its application effect in actual inspection scenarios.

关 键 词：Yolov5 卷积块注意力机制 高效交并比损失函数 小目标识别 

分 类 号：TM75[电气工程—电力系统及自动化]