In order to eliminate the strong noise in 3D GPR data collected in urban underground pipeline detection, a method combining singular value decomposition(SVD) and wavenumber filtering is proposed, and a quantitative criterion for selecting singular value number threshold is given. The study shows that SVD method can distinguish the target signal from noise signal accurately by selecting appropriate singular values. In addition, the wavenumber domain filter based on Gaussian low pass can suppress clutter and noise effectively. The method combining the above-mentioned SVD and wavenumber filtering shows its ability to effectively improve the signal-to-noise (SNR) ratio of 3D GPR depth slice signal, making effective reflection signals clearer whilst preserving the integrity of underground target information.The proposed method can enhance the efficiency and accuracy of target recognition by 3D GPR, making it more suitable for urban underground pipeline detection.