Detection Technology On Rail Surface Defects Based On Image Features

In view of the poor reliability,low adaptability and less automation existing in the process of rail surface detection defecting,this paper has designed a system based on image characteristics,by combining image processing and pattern recognition methods.The paper firstly introduced the latest development of the rail surface detection technology,analyzed some kind of defects and it’s reasons,designed a kind of rail surface detection system based on image features.This system mainly included three parts,image preprocessing,feature extraction and description,and classifier design.On the step of preprocessing,firstly the pretreatment stage,first of all,by improving the average projection method to extract the rail area,and simple test,select images may contain defects of rail,using adaptive median filter and wavelet denoising method of combining the realize rail image denoising.Aiming at the existence of rail image gray uniform distribution,the single threshold indivisible problem,put forward a kind of block adaptive fuzzy enhancement algorithm,based on sub-block entropy value judgment,fuzzy enhancement for defect potential sub-block,and through the OSTU threshold segmentation method to achieve image segmentation.Extracting defect image gray scale,geometry,frequency domain characteristics,design training BP neural network model,realize the recognition of the rail image classification.By analyzing the results,the system can realize crack,scratch,scar,identification and classification of defects such as peeling,overall to miss rate 8%,misjudgment rate index88.5%,shows that the system of practical application has certain reference value.