摘要
提出了一种基于改进YOLOX-S视觉检测算法的轨道交通车辆自动识别方法,通过优化网络结构和学习率策略,基于梯度差自适应学习率对YOLOX-S视觉检测算法进行改进,引入注意力机制和Transformer模块,增强YOLOX-S视觉检测算法对轨道交通车辆复杂特征的捕捉能力。实验结果表明,改进后的YOLOX-S视觉检测算法其检测精度均值达93.20%,算法体积减小至15 MB,检测速度提升至28.90帧/s,检测延迟减至80.90 ms,为轨道交通车辆智能化监测和管理提供了技术支持。
An automatic recognition method is proposed for rail transit vehicles based on the improved YOLOX-S visual detection algorithm.This method improves the YOLOX-S visual inspection algorithm based on the adaptive learning rate of gradient difference by optimizing the network structure and learning rate strategy.The attention mechanism and Transformer module are introduced to enhance the YOLOX-S visual detection algorithm's ability to capture the complex features of rail transit vehicles.The experimental results show that the average detection accuracy of the improved YOLOX-S visual detection algorithm reaches 93.20%,the algorithm volume is reduced to 15 MB,the detection speed is increased to 28.90 frames per second,and the detection delay is reduced to 80.90 ms.It provides technical support for the intelligent monitoring and management of rail transit vehicles.
作者
邢国栋
刘东凯
刘晓龙
Xing Guodong;Liu Dongkai;Liu Xiaolong(Locomotive Division of Guoneng Baoshen Railway Group Co.,Ltd.,Inner Mongolia Ordos,017000,China;Beijing Beijiufang Rail Transit Technology Co.,Ltd.,Beijing,100036,China)
出处
《机械设计与制造工程》
2026年第2期102-106,共5页
Machine Design and Manufacturing Engineering
关键词
YOLOX-S视觉检测算法
算法压缩
轨道交通
车辆自动识别
视觉检测
YOLOX-S visual inspection algorithm
algorithm compression
rail transit
automatic vehicle identification
visual detection
