中国铁道科学2025，Vol.46Issue(6)：34-43,10.DOI:10.3969/j.issn.1001-4632.2025.06.04
基于改进YOLOv9的钢轨B显图像损伤识别算法
Research on Track B Image Damage Recognition Algorithm Based on Improved YOLOv9
摘要
Abstract
Target Detection in B-scan images is currently a relatively effective method for rail flaw detection.To address the problem that traditional object detection algorithms tend to produce misjudgments and omissions for different types of flaws when detecting B-scan images,this paper proposes a flaw recognition algorithm for rail B-scan images based on the improved YOLOv9.Firstly,aiming at the problem of omissions and misjudgments,a novel multi-level feature fusion mechanism is proposed to improve model accuracy and reduce omissions and misjudgments.Then,aiming at the problems that traditional algorithms rely heavily on high-quality datasets and have a large model computational load,the loss function is optimized and the depthwise separable convolution is introduced to improve model training efficiency.Finally,verification of the improved YOLOv9 algorithm is conducted by testing various detection metrics of the YOLOv9 algorithm(before and after improvement)on the B-scan image dataset.The results show that the detection average precision(AP)of the YOLOv9 model after various improvements reaches 99.5%,and its performance outperforms other models.关键词
钢轨/探伤/B显图像/YOLOv9/目标检测/损伤识别Key words
Rail/Flaw detection/B-scan image/YOLOv9/Object detection/Flaw recognition分类
交通工程引用本文复制引用
白堂博,刘伟峰,宫明明,张玉华,黄筱妍..基于改进YOLOv9的钢轨B显图像损伤识别算法[J].中国铁道科学,2025,46(6):34-43,10.基金项目
北京市自然科学基金资助项目(L211007,L221027) （L211007,L221027）
