铁道运输与经济2025，Vol.47Issue(11)：84-93,10.DOI:10.16668/j.cnki.issn.1003-1421.2025.11.07
基于改进VGG16的无人机钢轨缺陷识别算法研究
Research on UAV Rail Defect Identification Algorithm Based on Improved VGG16
摘要
Abstract
With the rapid development of the low-altitude economy,the application of unmanned aerial vehicle(UAV)inspection technology in railway infrastructure maintenance has seen significant growth.To address the inefficiency and high cost of traditional manual inspection,as well as the limited computational resources and low precision rate of existing deep learning models on UAV-mounted devices,this paper proposed an improved VGG16 network model for rail defect identification in UAV inspection scenarios.First,several standard convolutional layers in the conventional VGG16 were replaced with depthwise separable convolutional layers to reduce the parameter count while preserving feature extraction capability.Second,the convolutional block attention module(CBAM)was incorporated during the high-level feature extraction stage,enabling the model to focus on key characteristics of rail defects.Finally,global average pooling(GAP)was introduced to reduce feature dimensions,and a random forest classifier was employed for classification,effectively enhancing classification performance while achieving lightweight model.Experimental results show that on the same dataset,the proposed algorithm outperforms the original VGG16 model in terms of precision rate by 24.9%and detection rate by 25%,with an 88.83%reduction in model size.Compared with the network models Inceptionv3 and ResNet34,the detection rate is 17.5%and 19.5%higher,respectively.关键词
钢轨缺陷/VGG16算法/深度可分离卷积/随机森林/注意力机制Key words
Rail Defect/VGG16 Algorithm/Depthwise Separable Convolution/Random Forest/Attention Mechanism分类
交通运输引用本文复制引用
王志飞,李樊,杨博璇,刘天阳,魏奇..基于改进VGG16的无人机钢轨缺陷识别算法研究[J].铁道运输与经济,2025,47(11):84-93,10.基金项目
中国铁道科学研究院集团有限公司科研项目(2024YJ360) （2024YJ360）
