针对目前热态重轨表面缺陷检测速度慢、精度低的问题，提出一种基于机器视觉的热态重轨表面缺陷检测系统。利用多线阵CCD摄像机采集图像，根据重轨几何特征及其缺陷高频区域特性，对重轨进行了六视角拍摄，然后在图像工作站中进行各种图像处理。系统采用改进的Hough变换提取特征缺陷，针对SVM算法训练速度慢的特点，利用模糊Kohonen神经网络对重轨表面缺陷进行分类。采用上述机器视觉检测关键技术对热态重轨表面进行缺陷识别，提高了检测速度，且正确率在85%以上。

Aiming at the low efficiency and precision of hot rail steel surface faults detecting at present,a suit of surface defect detection system of hot heavy rail based on the machine vision is put forward. Multi-CCD cameras are used to collect pictures. According to the geometric characteristics of the heavy rail and its defect characteristics of high-frequency region,six angle shot is used for heavy rail,and then various image processing technology are adopted in workstation. The system adopts improved Hough transform to get surface faults and Kohonen network to make a classification for the characteristics of low SVM training algorithm. The above key machine vision technology for detection of hot heavy rail surface defects greatly improves the speed and accuracy of testing and the detecting correct rate arrives over 85%.

国家自然科学基金委员会与中国工程物理研究院联合基金资助(10976034)