基于改进YOLOv5的铁道辙叉损伤检测研究  

Research on damage detection of railway crossing based on improved YOLOv5

作　　者：俞成海[1] 谭勇 叶泽支 卢智龙 YU Chenghai;TAN Yong;YE Zezhi;LU Zhilong(School of Computer Science and Technology(School of Artificial Intelligence),Zhejiang Sci-Tech University,Hangzhou 310018,China)

机构地区：[1]浙江理工大学计算机科学与技术学院(人工智能学院),杭州310018

出　　处：《智能计算机与应用》2025年第3期207-212,共6页Intelligent Computer and Applications

摘　　要：铁道辙叉表面损伤如裂纹和掉块会对铁路运输安全造成严重威胁。虽然现有的深度学习模型可有效检测铁道表面损伤,但对结构更复杂的辙叉损伤检测效果欠佳。为此,本文构建了一个包含不同天气环境的铁道辙叉损伤数据集,并基于YOLOv5提出了一种新的YOLOv5-CDC网络模型。模型在主干网络中引入了通道注意力机制,提高模型对辙叉缺陷的定位能力;在Neck模块中采用了改进的密集连接结构,增强了不同抽象层特征的融合;此外,在Neck和检测头间加入了通道块注意力模块,降低了复杂光照对检测性能的不利影响。实验结果表明,本文提出的改进模型相较原YOLOv5模型在铁道辙叉损伤检测任务上平均精度提升3.4%。经实验证明,本文方法可以有效提高铁道辙叉缺陷的检测性能,具有一定的实用价值。Surface defects on rail frogs such as cracks and spalling pose a serious threat to railway transportation safety.Although existing deep learning models are effective in detecting rail surface damage,their performance on the more complex structure of frog damage detection is poor.To address this problem,this paper constructs a rail frog damage dataset containing different weather environments,and proposes a new YOLOv5-CDC network model based on YOLOv5.The model introduces channel attention mechanisms in the backbone network to improve the model's localization ability for frog defects.An improved dense connection structure is adopted in the Neck module to enhance the fusion of features from different abstraction levels.In addition,channelspatial attention modules are added between the Neck and detection head to reduce the adverse effects of complex lighting on detection performance.The experimental results show that the improved model proposed in this paper achieves an average precision improvement of 3.4%over the original YOLOv5 model on rail frog damage detection.It is validated that the proposed method can effectively improve the detection performance on rail frog defects and holds certain practical value.

关 键 词：铁道辙叉 缺陷检测 YOLOv5 注意力机制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]