改进YOLOv7的输电线路变尺度目标检测  

Improved YOLOv7 for Variable Scale Target Detection of Transmission Lines

作　　者：周景[1] 李英杰 周蓉[1] 崔灿灿 ZHOU Jing;LI Ying-Jie;ZHOU Rong;CUI Can-Can(School of Control and Computer Engineering,North China Electric Power University,Beijing 102206,China)

机构地区：[1]华北电力大学控制与计算机工程学院,北京102206

出　　处：《计算机仿真》2024年第8期228-233,共6页Computer Simulation

基　　金：国家自然科学基金项目(52179014)。

摘　　要：针对输电线路无人机巡检图像中小目标检测精度低下的问题,提出一种改进型YOLOv7的输电线路变尺度多目标的检测方案。方案首次将基于YOLO7的目标检测模型应用到输电线路目标检测中,引入Transformer注意力工作机制,使用g^(n)Conv代替高效聚合网络中的卷积层提取巡检图像特征,经过RFPN网络将不同分辨率特征进行融合后,分别进行不同尺度目标的预测,提高对小目标的检测精度,达到了93.68%的平均检测精度,也可以检测到被遮挡的目标,具有一定的泛化能力。结果表明,上述模型能够有效检测出巡检图像中的防震锤和绝缘子,为后续故障诊断提供了理论依据。Aiming at the problem of low detection accuracy of small targets in UAV inspection images of transmission lines,an improved YOLOv7 variable-scale multi-target detection scheme for transmission lines is proposed.This solution applies the YOLOv7-based target detection model to transmission line target detection for the first time,introducing the Transformer attention mechanism,usingg"Conv to replace the convolution layer in the efficient aggregation network to extract inspection image features,and fusing features of different resolutions through RFPN network to predict targets of different scales.The detection accuracy of small targets is improved,reaching an average detection accuracy of 93.68%.It can also detect occluded targets and has a certain generalization ability.The results show that the model can effectively detect the anti-vibration hammer and insulator in the inspection image,which provides a theoretical basis for subsequent fault diagnosis.

关 键 词：深度学习 目标检测 输电线路 

分 类 号：TP391[自动化与计算机技术—计算机应用技术] TM726[自动化与计算机技术—计算机科学与技术]