This paper proposes a method based on singular value decomposition technology in order to solve the hard target penetration overload signals de-noising problem. First, the signal reconstruction sub-matrix is established based on the principle of stability of the main singular components. Second, the "dominant of the former K singular values energy" rule is used to extract the effective order of singular value. Penetration signals are then decomposed based on the previous steps. Finally, the signal is reconstructed using the extracted effective singular values. Experiments show that the proposed method can effectively eliminate the vibration and noise hiding in the penetration process. The proposed method can get a better signal to noise ratio than when using the wavelet transform, as well as the exact penetration depths of the experiments. Ultimately, the proposed method is a feasible new method on penetration fuse signals processing.