A detection method for rail tread defects based on the Fourier transform

Abstract Traditional rail tread defect detection methods based on two-dimensional grayscale images struggle to quantify defect extent and are easily influenced by tread contamination. To overcome these issues, a line—structured light sensor was utilized to collect rail tread depth data, and a Fourier—transform—based detection method was proposed. This method transforms tread depth images into the frequency domain, conducts filtering operations, and then reverts them to the image domain, effectively highlighting defects like scratches and spalling, which are identified via feature screening. Considering the curvature characteristics of different rail top regions, a region—specific defect detection strategy divides the detection into three subtasks: central defect detection, abnormal rail gap width detection, and edge defect detection, with tailored methods for each. To address measurement errors from single—direction projection, a sectional—projection—based three-dimensional defect size measurement method was designed. Experimental results show that the proposed approach can accurately quantify defects. It achieves a 0.2 mm accuracy for tread defect depth and 2 mm accuracy for rail gap width. Under dynamic conditions, it attains an 89.2% detection rate and an 8.3% false positive rate, demonstrating strong resistance to common interferences. This method is crucial for enhancing in—service rail tread detection and supporting maintenance decisions.