摘要
针对复杂环境下输电线路树障检测识别准确率较低的问题,提出了一种基于卷积神经网络的D-LinkNet模型语义分割技术。算法采用编码器-解码器结构,利用扩展卷积扩大感受野的同时引入特征提取模块,通过像素间的关联信息矩阵来构建网络权值矩阵,提高了网络对边界模糊区域的分割能力。仿真实验结果表明,所提算法将树障检测的准确度提高至97.87%,相较于FCN模型预测准确率提高了12.23%,且在有效提高识别精度的同时兼顾了运算速度,具有更高的实用价值。
To solve the problem of lower accuracy of transmission line tree obstacle detection and recognition in complex environment,a D-LinkNet model semantic segmentation technology based on convolution neural network was proposed.A coder-decoder structure was adopted by the algorithm.The structure took advantage of extended convolution to expand the receptive field and introduce feature extraction module.The network weight matrix was constructed by the correlation information matrix between pixels,for the improvement of network segmentation ability within the fuzzy boundary area.The simulation results show that the as-proposed algorithm improves the accuracy of tree obstacle detection to 97.87%,and the prediction accuracy is 12.23%higher than that of FCN model.The algorithm not only effectively improves the recognition accuracy,but also considers the operation speed,and has higher practical value.
作者
蔡文彪
吴怀诚
李立学
董云鹏
张嘉杨
CAI Wenbiao;WU Huaicheng;LI Lixue;DONG Yunpeng;ZHANG Jiayang(School of Power Transmission and Transformation Technology,Northeast Electric Power University,Jilin 132012,Jilin,China;Ultra High Voltage Branch,State Grid Liaoning Electric Power Co.,Ltd.,Shenyang 110003,Liaoning,China;School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200241,China)
出处
《沈阳工业大学学报》
CAS
北大核心
2024年第6期766-771,共6页
Journal of Shenyang University of Technology
基金
辽宁省自然科学基金项目(2020-MS-239)
国网辽宁省电力有限公司科技项目(5222JX20005A)。
关键词
输电线路
树障
语义分割
卷积神经网络
特征识别
关联信息
权值矩阵
边界模糊区域
transmission line
tree obstacle
semantic segmentation
convolution neural network
feature recognition
correlation information
weight matrix
fuzzy boundary area
