Based on the theory that the wavelet coefficients can signify the wavelet functions and analyze the degree of signal approximation, by superimposing the wavelet coefficients of different scales we compare nine kinds of wavelet functions and find that Mexican hat wavelet can be used to erase high frequency noise in GPR signals and amply effective signals, thereby providing basis for the judgment on suspected abnormal location by adopting the traditional data processing method to process the GPR signals.