Using ultrasonic non-destructive testing technology,when detecting the surface crack of the shaft,the micro crack echo signal is often covered by various noises,which causes the crack to be unrecognizable and positioned. In order to solve this problem,in this paper,a general cross-validation thresholding in synchrosqueezing wavelet domain approach is proposed to analyze the ultrasonic echo signals,and the characteristics of the crack echo signals are extracted from the time-frequency domain and localized. Based on synchrosqueezing transformation,the paper uses the general cross-validation method to estimate the threshold level of noise reduction,and does not depend on any prior knowledge. Specifically,by adding a preprocessing step based on kurtosis measurement and a post-processing step based on adaptive hard threshold processing,the efficiency of the threshold processing and the noise reduction effect in the time-frequency domain are improved,thereby realizing a useful distinction between noise and characteristic signals. Finally,the method is applied to the feature recognition of microcrack ultrasonic echo signals,and compared with the results of continuous wavelet transform. The results show that the method can identify the crack more accurately and extract the time point of crack occurrence,and then determine the specific location of the micro crack.