Anomaly Detection on Railway Track Using Haralick Feature for Color Mapping Co-Occurrence Matrix

Surface anomalies on railway tracks pose safety hazards to the operation of high-speed trains. Employing advanced target detection techniques to identify and track anomalies revolutionizes traditional manual detection methods, enhancing detection speed and objectivity. However, the outdoor deployment of railway tracks renders them susceptible to variations in sunlight, leading to a decrease in the accuracy of target detection. To mitigate the impact of lighting changes on the recognition of surface anomalies on railway tracks, this study initially transforms RGB images of tracks into six other color models, including HSV, HSI, YCbCr, Lab, YIQ, and LUV, and extracts color model components unrelated to brightness. Haralick Feature for color mapping co-occurrence matrix is then utilized as an anomaly recognition feature, followed by input into three detectors-SVM, Random Forests, and AdaBoost-and evaluation using metrics such as TP, TN, FP, and FN to assess the recognition results of different color models across various detectors. The results demonstrate the effectiveness of the proposed method in identifying track anomalies, with the Cb-Cr component of the YCbCr color model exhibiting superior performance among all color models, achieving accuracies of 0.976, 0.988, and 0.988 across the three detectors, respectively.