一种基于LSKNet的绝缘子缺陷检测方法研究
Research on an insulator defect detection method based on LSKNet
范美楷 1方志 1晏宇 1刘苈乐 1黄鹏程 1钟剑丹1
作者信息
- 1. 成都信息工程大学通信工程学院,四川成都 610025
- 折叠
摘要
目前,电力网络缺陷检测主要通过无人机航拍完成.对当前公开的数据集进行筛选,发现绝缘子的标注误差较大且正负样本失衡;同时,巡检图像中存在许多小尺度和细长类型的目标,使用现有的算法很难达到高精度的检测效果.针对上述问题,通过雾化算法构建一个新的数据集,采用大型选择核网络(LSKNet),引入暗通道先验算法,提出针对电力网络缺陷的LSK绝缘子图像去雾算法.实验结果表明,在SFID-PRO数据集上的mAP达到 85.90%,其中缺陷绝缘子的召回率达到了99.6%,能够对细长物体和小尺寸物体进行精准的检测.
Abstract
Currently,power network defect detection is mainly accomplished by UAV(Unmanned Aerial Vehicles)aerial photogra-phy.Screening the current publicly available dataset,it is found that the labeling error of insulators is large and the positive and nega-tive samples are out of balance;At the same time,there are many small-scaled and slender-type targets in the inspection images,and it is difficult to achieve high-precision detection using the existing algorithms.To address the above problems,a new dataset is con-structed by fogging algorithm,a large selective kernel network(LSKNet)is used,and the Dark Channel Prior algorithm is introduced to propose a LSK insulator image defogging algorithm for defects in power networks.The experimental results show that the mAP on the SFID-PRO dataset reaches 85.90%,in which the recall rate of defective insulators reaches 99.6%,and it is able to accurately detect elongated objects and small-sized objects.
关键词
小目标检测/绝缘子缺陷/暗通道先验/LSKNet/深度学习Key words
Small target detection/Insulator defects/Dark channel prior/LSKNet/Deep learning引用本文复制引用
基金项目
大学生创新创业训练计划(202210621145)
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据