基于GA-SVM的钢轨廓形类型在线识别算法研究
Research on Online Recognition Algorithm of Rail Profile Type Based on GA-SVM
叶志坚 1王菁 1吴越 1陈建政1
作者信息
- 1. 西南交通大学,轨道交通运载系统全国重点实验室
- 折叠
摘要
针对轨道交通日常运维中钢轨廓形自动化检测识别率不高的情况,提出了一种基于几何描述符和支持向量机(SVM)的高精度钢轨廓形在线识别算法.利用结构光传感器对钢轨廓形数据进行采集,采用几何去噪算法对廓形进行离群点剔除和重采样预处理.通过廓形几何描述符对不同类别钢轨廓形进行特征提取,制作廓形特征数据集用于训练SVM.采用遗传算法(GA)对SVM模型参数进行优化选取.将优化训练后的SVM模型用于钢轨廓形检测并和传统廓形识别方法进行对比.结果表明:提出的采用几何描述符的GA-SVM模型平均准确率达到99.62%,单帧廓形识别用时 6.43 ms,能有效提升廓形识别准确率与高速性,满足轨道车辆在线检测的需求,并为轨道自动化检测提供了理论和技术支撑.
Abstract
Aiming at the low recognition rate of rail profile automatic detection in daily rail transit operation and mainte-nance,a high-precision rail profile online recognition algorithm based on geometric descriptors and support vector machine(SVM)was proposed.Structure light sensor was used to collect rail profile data,and geometric denoising algorithm was used to re-move outliers and resampling the profile.The feature extraction of different types of rail profiles was carried out through the profile geometric descriptors,and the profile feature dataset was made for SVM training.Genetic algorithm(GA)was used to optimize the parameters of SVM model.The optimized SVM model was used for rail profile detection and compared with the traditional profile recognition method.The results show that the proposed GA-SVM model using geometric descriptors can achieve an average accu-racy rate of 99.62%and a single-frame contour recognition time of 6.43 ms.It can effectively improve the accuracy and high speed of profile recognition,meet the needs of online detection of rail vehicles,and provide theoretical and technical support for automated track detection.
关键词
轨道自动化检测/钢轨廓形/几何描述符/遗传算法/支持向量机Key words
automated rail inspection/rail profile/geometric descriptors/genetic algorithms/support vector machines引用本文复制引用
基金项目
国家自然科学基金项目(U21A20167)
出版年
2024
NETL
NSTL
万方数据