Abstract: Detection of the rail surface defects is important to the safety of railway transportation．Leveraging the techniques of image processing and pattern recognition to detect and locate defects is a viable and rapidly developing research techniques．In previous work,a Rail Surface Defect Detection (RSDD) was proposed by our research group．RSDD first enhances the contrast of the rail image,on this basis locates and detects suspicious defects．It has a high detection performance for conventional rail image,but misses some defects in the cases of rail image that contains multiple defects and has high gray value difference among them．This paper put forward an Improved Rail Surface Defect Detection (I-RSDD),which fills the detected defect areas with mean gray value of original rail image and then detects the intermediate result image again．In the rebuilt contrast image,thereby,the contrast value of the defect areas which are not obvious in the original image is enhanced and the possibility for the defect areas to be detected is increased．Our experimental results demonstrate that I-RSDD has high detect property:in the case of no noticeable increase of total time consumption,the average rate of accuracy of detection is 90.8%,and the average detection error rate is 4.0%,which has substantially improvement compared with RSDD.

Key words: Rail surface defect,Detection,Algorithm improvement,Spatial filtering