Optimizing Color Model Selection for Accurate Target Detection in Rail Anomaly

• Published in: Springer Proceedings in Mathematics and Statistics
• Main Author: Liyuan Y.; Osman G.; Hong L.
• Format: Conference paper
• Language: English
• Published: Springer 2024
• Online Access: https://www.scopus.com/inward/record.uri?eid=2-s2.0-85209572736&doi=10.1007%2f978-981-97-3450-4_19&partnerID=40&md5=c3395f2508a47526a792fdcfe778f1fa

Rapid and accurate detection of anomalies is essential for intelligent railway operation and maintenance. Target detection of rails is susceptible to interference from sunlight, leading to decreased accuracy in detection. The main objective of this study is to compare six (6) color models that can overcome the interference of sunlight based on feature analysis and to identify the most contributory color model for detecting anomalies in railway tracks. Firstly, the HSI, HSV, YCbCr, Lab, YIQ, and LUV color models were analyzed, and the color features, Contour Feature, and Haralick’s features for the co-occurrence matrix of each component without “luminance” were extracted. Secondly, the performances of 252 images of different abnormal types in different color model components and different features were compared and analyzed, and SVM was used to detect and recognize the results. Finally, the features of each component without “luminance” were integrated, and the color model that contributed the most was determined. The experimental results revealed that the highest recognition accuracies of the color feature, Couture Feature, and Haralick’s features for the co-occurrence matrix were 95.63%, 98.39%, and 98.13%, respectively. Moreover, after the feature fusion process, the HSV color model was found to be the best, achieving an exceptional recognition accuracy of 98.79%. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.