深度学习结合超分辨率重建技术的电力终端设备识别系统
Electric power terminal device recognition based on deep learning combined with super-resolution reconstruction technique
蒋跃宇 1张鹏程 1杨凯 1王霄聪 1王康1
作者信息
- 1. 国家电网江苏省电力有限公司常州供电分公司,江苏常州 213000
- 折叠
摘要
为了提高无线专网电力终端设备布设、故障处理的效率,设计一种基于深度学习结合超分辨率重建技术的电力终端设备识别系统.该系统利用超分辨率重建技术和Overlap切分法,改善识别大尺寸遥感图像时产生的漏检问题.加入由空间到深度(Space-to-Depth,SPD)层和非Stride卷积Conv层组成的SPD-Conv模块优化YOLOv5算法,以增强小目标检测的效果,并在自建的输电杆塔遥感图像数据集进行训练.实验结果表明,该系统对于单个设备的识别速度可达到毫秒级,mAP指标为0.833,能够有效提高电力终端设备检测的效率.
Abstract
To increase the efficiency of targeted deployment and fault handling in electric power wireless mesh networks,a power terminal device recognition system is designed based on the inte-gration of deep learning and super-resolution reconstruction technique.To solve the problem of missed detections in identifying large-scale remote sensing images,the overlap segmentation method and the super-resolution reconstruction technique are adopted.The SPD-Conv module,which is the combination of the SPD layer and non-Stride convolution layer,is included to optimize the YOLOv5 algorithm,so as to enhance the effectiveness of small object detection,and train it on the constructed dataset of transmission tower images from remote sensing satellite(TIRS).Experiment results dem-onstrate that the system achieves a recognition speed of millisecond level for individual devices,with a mean average precision(mAP)of 0.833,which can effectively increase the efficiency of power ter-minal device detection.
关键词
遥感图像/无线专网/深度学习/图像识别/输电杆塔遥感图像数据集Key words
remote sensing images/wireless private network/deep learning/image recognition/data-set of transmission tower images from remote sensing引用本文复制引用
基金项目
国家电网江苏省电力公司孵化项目(JF2022026)
出版年
2024
NETL
NSTL
万方数据