Abstract: The Sichuan-Tibet Railway features extremely complex geological conditions and construction environment, which greatly disturb the advanced geological forecast signals of ground penetrating radar (GPR). Therefore, it is necessary to adopt more effective signal noise reduction technology to conduct signal processing. With the tunnel section in Lalin of the Sichuan-Tibet railway as an example, this study applies the DTCWT to the advanced geological prediction of tunnels. Firstly, GPR signals were decomposed to some coefficients using the dual-tree complex wavelet transform (DTCWT), which were collected for the advanced geological prediction. Then high-frequency coefficients were shrunk for denoising based on four threshold processing selection rules (i.e., Sqtwolog, rigrsure, Heursure, and Minimaxi) combined with four thresholding schemes (i.e., hard, soft, firm shrinkage, nonnegative garrote shrinkage). Then signals were reconstructed using Wavelet coefficients through DTCWT inversion after the threshold shrinking. The wavelet denoising effects were assessed by calculating the normalized root-mean-square error (NRMSE), signal-to-noise ratio (SNR), and energy ratio (P). It is found that the best denoising effects can be obtained using the rigrsure threshold selection method combined with nonnegative garrote shrinkage and that the recognition accuracy of advanced geological forecast can be significantly improved accordingly.