摘要
针对现有输电铁塔鸟巢检测准确率较低的问题,提出一种基于轻量化残差卷积注意力网络(LRCAN)的无人机(UAV)输电线路鸟巢检测系统。在分析工作流程基础上,提出一个基于LRCAN的输电线鸟巢检测模型,使网络更关注所需的细节特征,并抑制其他无用信息的干扰;基于深度可分离卷积层修改特征融合网络中的正常卷积,减少网络参数数量。仿真结果表明,与参数数量相似的YOLOX-S相比,所提模型mAP提高了5.4%;与具有相同水平mAP的YOLOX-L和YOLOX-X相比,所提模型参数数量减少至它们的1/5和1/10。
Aiming at the problem of low accuracy in existing bird nest detection on transmission towers,a UAV(Unmanned Aerial Vehicle)-based bird nest detection system for power transmission lines based on the Lightweight Residual Convolutional Attention Network(LRCAN)is proposed.On the basis of analyzing the workflow,a bird nest detection model for power transmission lines based on LRCAN is proposed,which enables the network to focus more on the required detail features and suppress the interference of other irrelevant information.The normal convolution in the feature fusion network is modified by using depthwise separable convolution layers to reduce the number of network parameters.Simulation results show that compared with YOLOX-S,which has a similar number of parameters,the proposed model has increased the mAP(mean Average Precision)by 5.4%.Compared with YOLOX-L and YOLOX-X,which have the same level of mAP,the number of parameters in the proposed model is reduced to 1/5 and 1/10 of theirs,respectively.
作者
舒恺
张洁
范天成
刘玉婷
张蔡洧
SHU Kai;ZHANG Jie;FAN Tiancheng;LIU Yuting;ZHANG Caiwei(Ningbo Electric Power Design Institute Co.,Ltd,Ningbo Zhejiang 315000,China;Ningbo Power Supply Company,State Grid Zhejiang Electric Power Co.,Ltd,Ningbo Zhejiang 315000,China)
出处
《太赫兹科学与电子信息学报》
2025年第6期648-654,共7页
Journal of Terahertz Science and Electronic Information Technology
关键词
电力系统
电力巡检
无人机(UAV)
深度学习
注意力机制
参数优化
power system
electric power inspection
Unmanned Aerial Vehicle(UAV)
deep learning
attention mechanism
parameter optimization
