基于反向P-M扩散的钢轨表面缺陷视觉检测  

Research on Inverse P-M Diffusion-based Rail Surface Defect Detection

作　　者：贺振东[1,2] 王耀南[1] 毛建旭[1] 印峰[1] 

机构地区：[1]湖南大学电气与信息工程学院,长沙410082 [2]郑州轻工业学院电气信息工程学院,郑州450002

出　　处：《自动化学报》2014年第8期1667-1679,共13页Acta Automatica Sinica

基　　金：国家自然科学基金(60835004;61072121;61172160;61175075);河南省科技攻关计划(142102210514)资助~~

摘　　要：研制了一种基于反向P-M(Perona-Malik)扩散的钢轨表面缺陷视觉检测装置,该装置可自动获取钢轨表面图像,并实现实时检测与定位钢轨表面缺陷.钢轨图像具有光照变化、反射不均、特征少等特点,为了在运动过程中从复杂的钢轨表面图像提取缺陷,首先将图像进行反向P-M扩散,然后将扩散后的图像与原图像进行差分,从而减小了上述因素的影响,最后将差分图像进行二值化操作,根据缺陷边缘特性和面积进行滤波,分割出缺陷图像.实验仿真和现场测试结果表明,该方法能很好地识别块状缺陷和线状缺陷,并且检测速度、精度、识别率和误检率都能很好地满足要求.Abstract A vision machine is developed for rail surface defects detection based on the inverse P-M （Perona-Malik） diffusion. The rail surface defects images can be obtained through an image acquisition system. The rail surface images show illumination variation, reflection inequality, and heterogeneous texture, they make the automated visual inspection task extremely difficult. The faultless region of the rail surface image is preserved by an inverse P-M model, but the fault region is smoothed after diffusing by an inverse P-M model. Therefore, by subtracting the inverse diffused image from the original image, the defects can be distinctly enhanced in the difference image. The influence of illumination variation, reflection inequality, and heterogeneous texture can also be decreased. A simple binary thresholding, followed by filter operations based on the edge performance and the size of defects, can then easily segment the defect. The simulation and field experiments indicate that the inspection machine can detect the rail surface defects effectively and the detection speed, accuracy, detection ratio and the fault ratio also satisfy the needs of automated rail track.

关 键 词：反向P-M扩散 图像差分 钢轨表面缺陷 视觉检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]