摘要
针对传统钢轨检测技术的效率低下、精度不足、安全隐患等问题,提出了基于曲率滤波和改进高斯混合模型（GMM）的钢轨表面缺陷检测方法。首先,提出了基于垂直投影的区域定位算法和灰度对比算法,克服现场工况复杂、轨面反射不均、信道噪声干扰的难题;考虑到图像信号受强工况噪声干扰,研究了具有隐式计算和曲面保持特性的曲率滤波法进行图像去噪;建立了基于马尔科夫随机场（MRF）的高斯混合模型完成表面缺陷的精确快速分割。最后,设计了＂区域定位-灰度均衡化-滤波-分割＂的实验流程,实验结果验证了算法的有效性,检测性能达到了精确度92.0%,相比其他方法更加精确、快速,具有更好的鲁棒性。
This paper proposes a visual detection method for rail surface defect based on curvature filtering and improved Gaussian mixture model（ GMM） aiming atthe problems of low efficiency,lack of precision and safety hazard for traditional rail inspection technique. First of all,this paper proposes a ROI location algorithm based on vertical projection and gray contrast algorithm,which overcome the difficulties of.complex field condition and rail surface reflectance inequality and signal channel noise interference,Considering the situation that the image signal is interfered by strong noise,a curvature filteringmethod with implicit computing and surface preserving power is studied to conductthe image denoising and keep the image details. Next,an improved Gaussian mixture model based on Markov random field（ MRF） is established to achievethe accurate and rapid surface defect segmentation. In the end,the experiment process of region location-gray equalization-filtersegmentation was designed. The experiment results verify the effectiveness of theproposed method,the detection accuracy of 92. 0% is reached,and the proposed method is more accurate,faster and more robust than other methods.
作者
张辉
金侠挺
Wu Q.M.Jonathan
贺振东
王耀南
Zhang Hui;Jin Xiating;Wu Q. M. Jonathan;He Zhendong;Wang Yaonan(College of Electrical and Information Engineering, Changsha University of Science and Technology, Changsha 410114, China;College of Electrical and Information Engineering, Hunan University, Changsha 410082, China;University of Windsor, Windsor N9B3P4, Canada)
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2018年第4期181-194,共14页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(61401046)
国家科技支撑计划(2015BAF11B01)
湖南省教育厅科学研究项目(17C0046)资助
关键词
钢轨表面缺陷
视觉检测
曲率滤波
马尔科夫随机场
改进高斯混合模型
rail surface defect
visual detection
curvature filtering
Markov random field
improved Gaussian mixture model
