基于深度学习的电力线路航拍图像检索分析方法  

Retrieval and Analysis Method for Aerial Images of Power Lines Based on Deep Learning

作　　者：刘主光 杨迁 吕家辉 LIU Zhuguang;YANG Qian;L Jiahui(Wenzhou Science and Technology Branch,Zhejiang Tusheng Power Transmission and Transformation Engineering Co.,Ltd.,Wenzhou 325000,China;Wenzhou Tusheng Holding Group Co.,Ltd.,Wenzhou 325000,China)

机构地区：[1]浙江图盛输变电工程有限公司温州科技分公司,浙江温州325000 [2]温州图盛控股集团有限公司,浙江温州325000

出　　处：《计算机测量与控制》2025年第3期295-301,共7页Computer Measurement &Control

基　　金：电网企业开放基金项目(CF058804062021011)。

摘　　要：电力行业的快速发展使得电力线路的巡检工作变得越来越重要;传统巡检方法存在着效率低、精度不高等问题,基于航拍图像的电力线路巡检已经成为重要的解决途径;在传统图像检索框架的基础上,创新提出一种基于深度学习的图像检索方法;通过构建深度残差注意力网络哈希图像检索模型,采用通道和空间注意力模块加强特征的细化,自动识别并聚焦于图像中的关键区域;此外,通过基准样本的三元组损失函数进行端到端学习,优化图像之间的距离度量,从而实现相似图像的精确区分;对比实验表明,该系统能够有效地对电力线路航拍图像进行检索和分析,为电力线路巡检工作提供强有力的支持。With the rapid development of the power industry,the inspection of power lines has become increasingly important.Traditional inspection methods have the characteristics of low efficiency and low precision,so power line inspections based on aerial images have become an important solution.Based on the traditional image retrieval framework,an innovative image retrieval method based on deep learning is proposed.By constructing a deep residual attention network Hash image model,channel and spatial attention modules are used to enhance feature refinement,automatically identifying and focusing on key regions of the images.Additionally,the system utilizes a triplet loss function with benchmark sample for end-to-end learning,optimizing the distance metrics between the images to accurately distinguish similar images.Comparative experiments show that the system can effectively retrieve and analyze aerial images of power lines,providing strong support for power line inspections.

关 键 词：电力巡检 图像检索 分析系统 深度学习 注意力机制 三元组损失函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]