Wavelet-Enhanced YOLOv8 for Real-Time Railway Surface Defect Detection

Timely and accurate detection of rail surface defects is crucial for ensuring railway operation safety. Methods based on deep learning have shown good results, but they are often limited by the scarcity of high-quality labeled defect data and the computational complexity of the models, which hinders their deployment in real-time detection systems. To address these issues, we propose an improved algorithm based on YOLOv8 for railway surface damage detection. By integrating wavelet delay using Discrete Wavelet Transform and Haar Wavelet Transform into the YOLOv8 model, it enhances the encoding ability for visual information such as image edges and lines, improving the capability to capture fine damages. This fusion enriches feature representation, providing a more comprehensive description of defect features across different scales. The choice of Haar wavelet is due to its computational efficiency and simplicity, allowing for fast processing without complex numerical operations. Experimental results on a rail defect dataset show that the proposed method achieves higher detection accuracy and robustness compared to traditional methods, with a 10% increase in mAP50, while significantly reducing computational overhead, making it suitable for real-time detection applications.