摘要
为解决输电线路所在环境较复杂多变、线路上异物目标体积较小难以识别等问题,提出将BCA YOLO网络针对小目标检测进行优化,将YOLO v5中的CSP2_X替换为CSP_CA,再添加一层小目标检测层,将原网络中的FPN结构替换为计算量小的BiFPN;针对一般输电线路异物数据集中图片较少的问题,提出通过场景增强、Mixup和加入噪声模拟对数据集进行有效扩充.试验结果证明,相较于传统的YOLO v5网络,mAP_0.5提高了3.8%、查全率提升6%、查准率提高了6.1%,更好地满足了隐患检测的工程实际需求.
This paper aims to solve problems of small volume of foreign object targets on transmission lines,which are difficult to identify,and lack of pictures in general foreign object data sets of transmission lines,which do not cover all kinds of scenes and environments.Thus,data sets are proposed to expand through scene enhancement,Mixup and noise simulation,and BCA-YOLO network is optimized for small target detection.Then,CSP2_X in YOLO v5 is replaced by CSP_CA,a layer of small target detection is added and FPN structure in the original network is replaced with BiFPN with small computation.The experimental results show that compared with traditional YOLO v5 network,mAP_0.5,recall rate and precision rate increase by 3.8%,6%,and 6.1%,respectively.What is more,it can meet actual needs of hidden danger detection.
作者
邹辉军
焦良葆
张智坚
汤博宇
刘子恒
ZOU Hui-jun;JIAO Liang-bao;ZHANG Zhi-jian;TANG Bo-yu;LIU Zi-heng(Institute of Artificial Intelligence Industry Technology,Nanjing Institute of Technology,Nanjing 211167,China)
出处
《南京工程学院学报（自然科学版）》
2022年第3期7-14,共8页
Journal of Nanjing Institute of Technology(Natural Science Edition)
基金
国家自然科学基金青年科学基金项目(62002160)。
