基于卷积神经网络的绝缘子目标检测研究  

Research on Insulator Target Detection Based on Convolutional Neural Network

作　　者：廖晓群 李元 沈树建 张汉瑾 付文旭 刘帅[1] LIAO Xiaoqun;LI Yuan;SHEN Shujian;ZHANG Hanjin;FU Wenxu;LIU Shuai(School of Communication and Information Engineering,Xi'an University of Science and Technology,Xi'an 710600)

机构地区：[1]西安科技大学通信与信息工程学院,西安710600

出　　处：《计算机与数字工程》2025年第2期409-414,516,共7页Computer & Digital Engineering

基　　金：中国高校产学研创新基金项目(编号:2021KSA05005)资助。

摘　　要：电网绝缘子的识别及定位是电网运行状态有效检测的前提,基于无人机拍摄的输电线路复合绝缘子图像,为解决传统人工巡检输电线路安全系数高、作业效率低下的问题,提出了基于改进的卷积神经网络绝缘子设备研究识别的方法。算法通过添加注意力机制CBAM和SENet,实现对输电线路器件定位的有效改进,实验结果表示,改进的模型较之前减少了运行时间,目标检测精度提高了8.2%,有效提高了对输电线路绝缘子检测的性能和鲁棒性,取得了比已有算法更有优势的检测结果。The identification and positioning of power grid insulators is the premise for effective detection of power grid opera⁃tion status.Based on the composite insulator images of transmission lines captured by drones,in order to solve the problems of high safety factor and low operation efficiency of traditional manual inspection of transmission lines,an improved method is proposed,which is convolutional neural network insulator device research recognition algorithm.By adding attention mechanism CBAM and SENet,the algorithm can effectively improve the positioning of transmission line devices.Experimental results show that the im⁃proved model reduces execution time compared to previous versions,and the target detection accuracy increases by 8.2%.This method significantly enhances the performance and robustness of the system for detecting insulators on transmission lines,providing more effective results compared to existing algorithms.

关 键 词：绝缘子 卷积神经网络 注意力机制 Faster RCNN 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]