학술논문
基于YOLOv5的铁路接触网异物检测模型初步研究 / Preliminary study on YOLOv5-based object detection models for foreign objects attached to railway overhead line equipment
Document Type
Academic Journal
Author
Source
铁路计算机应用 / Railway Computer Application. 33(2):13-18
Subject
Language
Chinese
ISSN
1005-8451
Abstract
接触网上附着的异物是影响铁路列车运行安全的一大隐患,在开行列车前需要检查接触网上是否有异物附着.目前,接触网异物检测主要依靠人工巡检,工作效率低,人力物力消耗大.文章通过建模实验,初步探讨利用基于深度学习的目标检测技术实现铁路接触网异物检测的可行性;构建了 3种接触网异物检测模型:YOLO(You Only Look Once)v5模型、YOLOv5+坐标注意力(CA,Coordinate Attention)改进模型和YOLOv5+ConvNext Block改进模型,利用包含鸟窝和轻质异物两种常见异物的接触网图像数据集,对这 3种模型进行实验分析.实验结果表明,相比YOLOv5算法,对于检测鸟窝和轻质异物两种常见的接触网异物,YOLOv5+CA改进模型和YOLOv5+ConvNext Block改进模型具有更好的效果,且YOLOv5+ConvNext Block改进模型检测小尺寸目标的能力更强.
Foreign object attached to the overhead line equipment is one of major hazards affecting the safety of railway train operation.Before operating the trains on a railway line,it is necessary to check whether there are foreign objects attached to the overhead line equipment.At present,the detection of foreign objects attached to the overhead line equipment mainly relies on manual inspection,which has low work efficiency and high consumption of manpower.This article explores the feasibility of using deep learning based object detection technology to realize automatic detection of foreign objects attached to the overhead line equipment through modeling experiments.Three foreign object detection models were constructed:YOLOv5 model,YOLOv5+CA improved model,and YOLOv5+ConvNext Block improved model.Using a data set of overhead line equipment images containing two common foreign objects,i.e.bird nests and lightweight foreign objects,experimental analysis was conducted on the three models.The experimental results show that compared to the YOLOv5 model,the YOLOv5+CA improved model and the YOLOv5+ConvNext Block improved model have better performance in detecting the two common foreign objects.Moreover,the YOLOv5+ConvNext Block improved model has stronger capability to detect small objects.
Foreign object attached to the overhead line equipment is one of major hazards affecting the safety of railway train operation.Before operating the trains on a railway line,it is necessary to check whether there are foreign objects attached to the overhead line equipment.At present,the detection of foreign objects attached to the overhead line equipment mainly relies on manual inspection,which has low work efficiency and high consumption of manpower.This article explores the feasibility of using deep learning based object detection technology to realize automatic detection of foreign objects attached to the overhead line equipment through modeling experiments.Three foreign object detection models were constructed:YOLOv5 model,YOLOv5+CA improved model,and YOLOv5+ConvNext Block improved model.Using a data set of overhead line equipment images containing two common foreign objects,i.e.bird nests and lightweight foreign objects,experimental analysis was conducted on the three models.The experimental results show that compared to the YOLOv5 model,the YOLOv5+CA improved model and the YOLOv5+ConvNext Block improved model have better performance in detecting the two common foreign objects.Moreover,the YOLOv5+ConvNext Block improved model has stronger capability to detect small objects.