摘要
[目的]钢轨的裂纹、掉块等伤损可能引发脱轨、断轨等事故,开展钢轨内部伤损检测工作具有重要的意义。钢轨探伤车超声B扫检测是最常用的检测方法,存在判伤时间长、人工工作量大、漏检等问题,为此,提出了一种基于改进YOLOv8算法的钢轨B扫图像自动缺陷识别方法。[方法]介绍了YOLOv系列算法的特点以及YOLOv8算法在图像分类检测中的优势;通过改进卷积层conv和损失函数WIoU优化YOLOv8算法;以上海轨道交通2014年至2022年全线实际伤损超声B扫图中800张图为数据集,通过试验验证改进YOLOv8算法的有效性。[结果及结论]与原始YOLOv8算法试验结果相对比,改进YOLOv8算法的平均精度达到了88.7,验证了改进YOLOv8算法对钢轨超声B扫图像识别的有效性和准确性。通过该方法能够优化通过人工回放钢轨B扫图像检测伤损时的漏检问题。
[Objective]Cracks,spalling,and other types of damage in rails may lead to derailment or rail breakage accidents,making internal rail defect detection critically important.Ultrasonic B-scan detection using rail flaw detection vehicles is the most commonly used detection method,but it presents issues such as lengthy damage assessment time,high manual workload,and missed detections.Therefore,an automatic defect recognition method for rail B-scan images based on an improved YOLOv8 algorithm is proposed.[Method]The characteristics of the YOLOv series algorithms and the advantages of YOLOv8 algorithm in image classification and detection are introduced.The YOLOv8 algorithm is optimized by improving the convolutional layer(conv)and the loss function(WIoU).A dataset of 800 actual ultrasonic B-scan images of rail defects collected from the entire Shanghai rail transit network from 2014 to 2022 is used to verify the effectiveness of the improved algorithm through experiments.[Result&Conclusion]Compared with the original YOLOv8 algorithm,the improved YOLOv8 algorithm achieves a mean average precision(mAP)of 88.7,validating its effectiveness and accuracy in identifying defects in rail ultrasonic B-scan images.This method improves the situation of missed detections during manual playback of rail B-scan image inspections.
作者
任恩璇
REN Enxuan(Track Works Branch of Shanghai Metro Maintenance Support Co.,Ltd.,200233,Shanghai,China)
出处
《城市轨道交通研究》
北大核心
2025年第S1期78-83,共6页
Urban Mass Transit
