Dual-Spectrum Imaging and Deep Learning for Enhanced Detection of Internal Defects and Surface Cracks in Railway Crossties

Detecting internal defects and surface cracks in railway crossties is crucial for track safety. Traditional inspection methods often fail to identify both simultaneously. This study integrates optical and thermal imaging with deep learning models to improve defect detection. Using a dual-spectrum camera, we captured optical and thermal images of crossties and applied image fusion techniques to enhance defect visibility. YOLOv10s and Faster R-CNN were evaluated on fused datasets, achieving F1-scores of 0.882 and 0.867, respectively. YOLOv10s demonstrated real-time inference (12 ms/frame), while Faster R-CNN excelled in precision. These results highlight the potential of dual-spectrum imaging and deep learning for efficient, accurate railway crosstie inspections.