铁道科学与工程学报2025，Vol.22Issue(4)：1828-1839,12.DOI:10.19713/j.cnki.43-1423/u.T20241118
基于改进YOLO v8的铁路人员入侵检测方法研究
Research on railway intrusion detection method based on improved YOLO v8
摘要
Abstract
China's railway transportation scale is huge and the operation environment is complex.Once intruders invade the railway operation area,it would seriously threaten the safety of railway transportation.This paper aimed to use intelligent algorithms to analyze and detect images from cameras along the railway and determine whether there are potential safety risks in real time,thus effectively protecting the safety of railway transportation.However,due to the uneven distribution of cameras along the railway,when there is a long-distance intruder,it is presented as a small target in the video image,which is easy to cause false and missed detection.Targeting such problems,an improved YOLOv8 intrusion detection method was proposed.First,the coordinate attention was added to the Backbone layer of the YOLOv8 model,which improved the feature extraction ability of the model for small targets and enhanced its robustness in complex background.Second,the detection head in head layer was changed to the dynamic head integrating size perception,space perception and task perception,and multiple perception modules were integrated into one detection head so as to make the model more accurate and efficient in detecting different size targets.Finally,the loss function was changed to Wise Intersection over Union(Wise-IoU),enabling dynamic focus adjustment during training to prioritize critical features and enhance model generalization.After experimental verification,each improvement significantly improved the detection effect of the model.Compared with the original YOLOv8n model,precision and recall of the improved model increased by from 2.2%and 2.3%to 91.6%and 83.7%,respectively;and mean Average Precision 50 and mean Average Precision 50-95 increased by from 2.6%and 2.9%to 89.9%and 60.9%,respectively.The research results also have a higher average detection accuracy as compared to that of other advanced target detection methods,which proves that it has a higher application value in railway intrusion detection and could provide more reliable technical support for railway transportation safety.关键词
YOLOv8算法/人员检测/注意力机制/检测头/损失函数Key words
YOLOv8 algorithm/personnel detection/attention mechanism/detection head/loss function分类
交通工程引用本文复制引用
吴浩楠,史宏,王瑞,杨文,胡昊..基于改进YOLO v8的铁路人员入侵检测方法研究[J].铁道科学与工程学报,2025,22(4):1828-1839,12.基金项目
中国国家铁路集团有限公司重大课题(K2023T003) （K2023T003）
