YOLOv8-FW: Rail Surface Defect Detection Algorithm Integrating Frequency Domain Features

Rail defect detection plays a crucial role in preventing train derailments and ensuring passenger safety. However, the continuous operation of trains can lead to the compression and abrasion of rails, resulting in defects such as delamination, subsidence, and scratches, which significantly compromise train safety. To address the limitations of existing detection algorithms in recognizing subsidence and scratch defects, this paper proposes a YOLOv8-FW model that incorporates frequency domain features for automated rail surface defect detection. By introducing the Frequency Domain Attention (FDA) mechanism into the backbone network of YOLOv8, the model can effectively transform spatial domain features into the frequency domain, thereby improving the recognition capability of defects with similar background features, such as subsidence. Additionally, the integration of the Wavelet Hybrid Convolution (WHC) module reduces feature loss and enhances the model's ability to detect small-target defects, such as scratches. Experimental results demonstrate that the YOLOv8-FW model achieves a 3.2% improvement in mean average precision compared to the YOLOv8n model, effectively enhancing the model's detection capabilities for small targets and defects with similar features.