北京交通大学学报2024，Vol.48Issue(5)：59-68,10.DOI:10.11860/j.issn.1673-0291.20240023
基于轻量化YOLOv8s的轨道扣件状态检测方法
Lightweight detection method for track fastener status based on improved YOLOv8s
摘要
Abstract
Railway infrastructure is continuously impacted by vehicle loads and external environmental factors,causing issues such as the loss,displacement,and damage of track fasteners along railway lines.These problems pose significant threats to the safe operation of railways.To address the low de-tection efficiency,high omission rates,and lack of real-time detection capabilities on edge devices as-sociated with traditional manual visual inspections and subjective sampling methods,this paper pro-poses a lightweight detection model for track fastener status,FTEL-YOLO,based on YOLOv8s.The model is designed to enhance detection accuracy and real-time performance.First,the C2f-Faster module,inspired by the FasterNet-Block concept,is introduced to reduce the model's parameters.Second,to mitigate the decline in detection accuracy caused by network lightweighting,a Triplet At-tention Mechanism is incorporated after the Spatial Pyramid Pooling Fast(SPPF)layer,and EIoU is utilized as the bounding box regression loss function,enhancing the model's feature extraction capabil-ity for track fastener conditions in complex backgrounds.Finally,Layer-Adaptive Magnitude-based Pruning(LAMP)is applied to the improved model to further compress it,reducing redundancy and en-hancing its deployment capability on edge devices.Experimental results demonstrate that the improved FTEL-YOLO model achieves a minimal detection accuracy loss of 0.3%,while the computation,pa-rameters,and model size are reduced by 63.1%,65.6%,and 66.2%,respectively,achieving light-weight design without compromising accuracy.关键词
深度学习/故障检测/轨道扣件/YOLOv8s/三元注意力机制/模型轻量化Key words
deep learning/fault detection/track fasteners/YOLOv8s/triplet attention/model light-weighting分类
信息技术与安全科学引用本文复制引用
武福,蒋鹏民,李忠学,杨喜娟,吕金旺..基于轻量化YOLOv8s的轨道扣件状态检测方法[J].北京交通大学学报,2024,48(5):59-68,10.基金项目
国家自然科学基金(56062028) （56062028）
甘肃省教育厅产业支撑计划项目(2021CYZC-11) （2021CYZC-11）
甘肃省教育厅创新基金项目(2022A-036) National Natural Science Foundation of China(56062028) （2022A-036）
Gansu Provincial Department of Education Industry Support Program Project(2021CYZC-11) （2021CYZC-11）
Gansu Provincial Department of Education Innovation Fund Project(2022A-036) （2022A-036）
