基于改进YOLOv8的铁路异物侵限检测方法  

A Railway Foreign Object Intrusion Detection Method Based on Improved YOLOv8

作　　者：陈伟迅 柯旭能 孟思明 Chen Weixun;Ke Xuneng;Meng Siming(Guangzhou Railway Polytechnic,Information Engineering Institute,Guangzhou 511300,China)

机构地区：[1]广州铁路职业技术学院信息工程学院,广州511300

出　　处：《机电工程技术》2024年第11期211-214,共4页Mechanical & Electrical Engineering Technology

基　　金：广东省普通高校创新团队项目(自然科学)(2021KCXTD068)。

摘　　要：快速、准确地从监控影像中检测侵入铁路的异物对保障列车运行安全具有重要意义。针对列车运行过程中出现的小型目标和与背景相似度较高的物体难以被准确识别的问题,提出一种改进的YOLOv8算法,可快速、准确地检测铁路图像中的侵限。首先,在Backbone网络主干层引入CBAM注意力机制,提高轨道异物特征提取速度,令模型更加关注于图像中的关键特征,同时抑制不相关的铁路背景信息。其次,针对模型中CIoU损失函数在宽高比上的模糊定义问题,将EIOU损失函数替代原有的损失函数,最小化目标框与锚盒的宽度和高度之差,提高边界框回归的精度的同时加快模型的收敛。最后利用分组卷积对传统目标检测头进行优化,在不损耗模型精度的情况下提高模型的效率,令模型在实际应用中具有更好的性能。实验结果表明,改进的YOLOv8算法在数据集上的mAP值达到96.2%,在检测精度上达到较高的水准,证明该模型在现实生活中具有应用价值。Efficient and accurate detection of foreign objects invading railways from surveillance images is of great significance for ensuring the safety of train operation.Aiming at the problem that small targets and objects with high similarity to the background appearing during train operation are difficult to be recognized accurately,an improved YOLOv8 algorithm that can detect the encroachment limits in railroad images quickly and accurately is proposed.Firstly,the CBAM attention mechanism is introduced in the Backbone network backbone layer to improve the speed of track foreign object feature extraction,so that the model pays more attention to the key features in the image,while suppressing the irrelevant railroad background information.Secondly,to address the fuzzy definition of the CIoU loss function in the model in terms of the width-to-height ratio,the EIOU loss function is used to replace the original loss function,minimize the difference between the width and height of the target box and the anchor box,and improve the accuracy of the bounding box regression while accelerating the convergence of the model.Finally,the group convolution is used to optimize the traditional target detection head,which improves the efficiency of the model without losing the accuracy of the model,and makes the model have better performance in practical applications.The experimental results show that the improved YOLOv8 algorithm achieves a mAP value of 96.2%on the dataset,which is a high level of detection accuracy and proves that the model has value for application in real life.

关 键 词：异物侵限检测 深度学习 目标检测 YOLOv8算法 CBAM注意力机制 EIoU RepConv 

分 类 号：U298[交通运输工程—交通运输规划与管理] TP391.41[交通运输工程—道路与铁道工程]