科技创新与应用2025，Vol.15Issue(13)：50-54,5.DOI:10.19981/j.CN23-1581/G3.2025.13.012
基于改进YOLOv5s的SAR影像铁道检测技术研究
摘要
Abstract
Gaofen-SAR satellites for railroad inspection have the advantages of wide coverage,all-weather and metal sensitivity,but need to solve the problem of instant detection of railway targets.For this reason,an improved model called Lightweight-YOLOv5s is proposed based on YOLOv5s,which is especially suitable for detection of railroad targets in SAR image.By reducing the Mobile-Darknet backbone feature extraction network layers,we optimized the network structure;by adding HDC and CBAM mechanisms,we adjusted the small-target sensory field weights and strengthen the small-target line facility feature extraction;by using FPGM pruning,we eliminated the redundant feature modules and achieve a lightweight model;by using Varifocal Loss as the loss function,we equalized the positive and negative categories and highlight the contribution of positive examples.The results show that,the accuracy of Lightweight-YOLOv5s model achieves 97.6%,and the inference time reduces to 6.87 ms.Compared with the classical algorithms for detecting linear targets in remote sensing images,the performance is greatly improved for instant detection of railroad targets.关键词
SAR/铁路巡检/铁道目标检测/YOLOv5s/轻量化网络Key words
SAR/railroad inspection/rail target detection/YOLOv5s/lightweight network分类
信息技术与安全科学引用本文复制引用
张经纶,徐天涛,李建国,王振,郑允,杨鹤猛..基于改进YOLOv5s的SAR影像铁道检测技术研究[J].科技创新与应用,2025,15(13):50-54,5.基金项目
中国国家铁路集团有限公司科技研究开发计划项目(N2023T013) （N2023T013）
