铁道科学与工程学报2017，Vol.14Issue(12)：2554-2562,9.
基于多特征融合与AdaBoost算法的轨面缺陷识别方法
Rail surface defect recognition method based on multi feature fusion and adaboost algorithm
摘要
Abstract
In view of the detection of surface defects are vulnerable to the vibration of the acquisition device and foreign interference, the rail image acquisition device was designed by analyzing the position of the defect. Firstly, according to the shape characteristics of rail, the rail surface area was extracted by combining Hough transform and least square method. Second, the excess entropy theory and the fuzzy theory were combined to divide rail surface defects. Then, by establishing the positive and negative sample databases, the sample feature database was established by extracting the Harr-like features and low-level features of the samples. Finally, the defect classifier was designed with C4.5 and AdaBoost algorithm, and the non defect regions were excluded and the defect regions were classified. By identifying 500~1000 lx, 1000~10000 lx, 10000~100000 lx in three different light intensity ranges of concrete sleeper and sleeper track rail surface defects, the average recognition time is 698 ms, the average recognition rate is 97.02%. Compared with the traditional recognition method, it has obvious advantages.关键词
振动/轨面提取/Hough变换/图像特征/AdaBoost/光照强度/缺陷识别Key words
vibration/rail surface extraction/Hough transform/image feature/AdaBoost/light intensity/defect recognition分类
信息技术与安全科学引用本文复制引用
闵永智,程天栋,马宏锋..基于多特征融合与AdaBoost算法的轨面缺陷识别方法[J].铁道科学与工程学报,2017,14(12):2554-2562,9.基金项目
国家自然科学基金资助项目(61663022,61461023) （61663022,61461023）
长江学者和创新团队发展计划资助项目(IRT_16R36) （IRT_16R36）
甘肃省高原信息工程及控制重点实验室开放课题基金资助项目(20161105) （20161105）
兰州交通大学优秀科研团队资助项目(201701) （201701）
