A rail corrugation recognization and characteristic parameter calculation method based on image spatial gray distribution processing

Abstract To solve inefficient rail corrugation detection and poor periodic feature recognition in intelligent models, this paper proposes a detection model and parameter calculation method using spatial gray distribution processing of rail images. The approach boosts detection speed, accuracy, and physical characterization. Rail surface images are analyzed by calculating vertical average pixel gray values. This extracts spatial gray distribution features for dimensionality reduction. The derived 1D features undergo two-step refinement: sliding window detrending eliminates baseline drift; Savitzky–Golay filtering smoothens signals. Processed data feeds a TimesNet-based detection model. Harmonic fitting is used for the detected corrugation to calculate physical parameters, yielding corrugation wavelengths and operational frequencies. Tests reveal 9 ms inference speed and 98.2% accuracy. Field data confirms wavelength/frequency consistency with observations, supporting rapid corrugation diagnosis.