Identification of the Optimal Color Model for Rail Anomaly Detection Based on Multi-Detector Performance Evaluation

The application of machine vision-based methods for rail anomaly detection facilitates timely maintenance and the intelligent management of railway infrastructure. However, the performance of detection systems is often compromised by variations in lighting, particularly due to the exposure of rails to outdoor environmental conditions. This study aims to identify the most effective color model for rail anomaly detection by comparing multiple models. The RGB images are first converted into six color models with luminance components—HSV, HSI, YCbCr, Lab, YIQ, and LUV. These images are then processed using classifiers (AFNN and RFNN) and detectors (YOLOV5 and YOLOV8). The results are quantitatively compared to determine the optimal color model. Experimental findings indicate that the YCbCr model achieves the highest detection performance, with a mAP of 0.921 under the YOLOV8 algorithm, followed by HSV. Conversely, the YIQ model demonstrated the poorest performance.