摘要
针对无人机输电线路巡检图像的复杂背景目标检测失准、故障小目标难以被准确检测的问题,提出一种基于深度展开超分网络(deep unfolding super-resolution network,USRNet)与改进YOLOv5x算法的输电线路绝缘子故障检测方法。首先,使用USRNet对原始图像进行超分辨率重建,以降低复杂背景干扰实现测试数据集优化;然后,以YOLOv5x检测模型为基础,利用K-means++对标记框进行聚类,生成匹配输电线路故障目标尺寸的锚框;同时,通过更改多尺度特征融合模块结构,在预测端引入一个包含更大特征图的检测头以检测故障小目标;最后,使用有效交并比损失(efficient intersection over union loss, EIOU_Loss)函数优化模型整体性能,并设置对比实验对所提方法进行验证。结果表明,所提方法的均值平均精度(mean average precision, mAP)值达到98.8%,可使输电线路故障检测精度提高到95.4%,从而具有更好的复杂背景目标以及小目标检测性能。
Aiming at the problem that targets with complex background and small targets in UAV aerial inspection images are difficult to be accurately detected, we proposed a transmission line insulator fault detection method based on USRNet and improved YOLOv5x algorithm. Firstly, the interference of complex background was reduced by USRNet to reconstruct the test set images with super resolution. Then, K-means++ was used to cluster the marker frames to generate anchor frames matching the size of transmission line fault targets. Meanwhile, a detection head with a larger feature map was introduced at the prediction side to detect faulty small targets by changing the structure of the Neck part. Finally, the overall performance of the model was optimized using the EIOU loss function, and comparison experiments were designed to validate the proposed method. The results show that the mAP value of the proposed method reaches 98.8%, and the fault detection accuracy of transmission lines can be improved to 95.4%, which has better detection performance of complex background targets and small targets.
作者
黄悦华
刘恒冲
陈庆
陈照源
张家瑞
杨楚睿
HUANG Yuehua;LIU Hengchong;CHEN Qing;CHEN Zhaoyuan;ZHANG Jiarui;YANG Churui(College of Electrical Engineering and New Energy,China Three Gorges University,Yichang 443002,China)
出处
《高电压技术》
EI
CAS
CSCD
北大核心
2022年第9期3437-3446,共10页
High Voltage Engineering
基金
国家自然科学基金(52007103)
湖北省科技重大专项(2020AEA012)。
