When laser-induced shock (LIS) wave characteristics testing is carried out using laser interference methods, the electrical noise, vibration noise, jitter or offset of the diffuse reflective surface during high-speed motion and other factors can lead to superposition or offset of the noise signal frequency with the low-frequency phase of the useful signal. As a result, problems such as phase shift of the useful signal and difficulty in extracting the initial position of the wavelet ridge may appear, and the initial position cannot be zeroed. By introducing a measurement factor through multi-resolution singular value decomposition and adaptively determining the optimal number of decomposition layers, zero-phase offset filtering is achieved and the signal-to-noise ratio (S/N) of the data is improved. According to the waveform decay characteristics of laser-induced shock wave velocity signals, the Shannon wavelet entropy method is proposed to accurately optimise the Morlet wavelet basis parameters to enhance the aggregation of wavelet scale spectrum time–frequency distribution. Simulation and test results show that: the proposed method overcomes the background noise interference and enables the initial velocity to be zeroed; the relative singular value points of the pulse are accurately symmetrically distributed and no phase shift occurs; the denoising performance is better than that of the wavelet method, the root mean square error is only 0.046, and the repeatability is strong; the error of the elastic wave velocity of the laser-induced shock wave at different power densities is within 3.2%; the relative velocity resolution remains constant at low/high frequencies, effectively avoiding false peaks and misclassification of the velocity profile.