Rail Anomaly Recognition Method Using Haralick Features and AFNN Detector

Abstract

Rail abnormalities can lead to traffic accidents causing personal and property losses, making timely detection of these anomalies crucial for rail transportation. However, the current discovery of rail abnormalities relies mainly on subjective human observation, lacking mature machine vision-based detection methods. This study proposes a new AFNN detector based on the extraction of Haralick features and the development of a fuzzy neural network for identifying rail abnormalities. Experimental results demonstrate that this method achieves rail abnormality recognition, with the identification performance of Haralick features surpassing that of colour features and HU features, achieving an accuracy of 0.9186.