With the continuous increasing of railway operation speed and traffic volume, the phenomenon of rail scratches caused by the interaction between wheels and rails has become more and more prominent. The acceleration of the axle box can effectively describe the short-wave irregularity of the track, and the scratches appear at fixed intervals. There will be periodic fluctuations. In view of the periodic characteristics of continuous rail scratches, we proposes an adaptive time-frequency analysis method for periodic track diseases. First, the adaptive short-time Fourier transform is performed on the acceleration data of the axle box, and the energy spectrum of time-frequency of the signal is obtained. After extracting the time energy signal using the $L^P$ norm criterion, then using the moving Gaussian weighted average filter and the least squares smoothing filter to smooth and detrend the time-energy signal. Finally, using a fixed window to smooth the signal perform post-intersection analysis and threshold judgment determine the continuous abrasion section. The algorithm is tested in the field, and the results show that the method of self-adaptive time-frequency analysis can effectively identify the continuous scratched section of the rail.