Abstract
Traditional noise reduction methods often struggle to balance noise suppression with the preservation of transient features in acceleration signals, especially when dealing with high-speed transient data. This study proposes a novel noise reduction method combining ensemble empirical mode decomposition (EEMD), sample entropy (SE), and improved wavelet threshold denoising (IWTD) to address the issue. The method utilizes EEMD to decompose the signal into intrinsic mode functions (IMFs) and a residual term. By setting an SE threshold (SE = 0.3), it effectively differentiates noise-dominated components from those containing significant transient features. IWTD is then applied to the noise-dominated components, and the processed components are reconstructed to yield the denoised signal. A baseline signal is generated in the lab, and noise is added to create the test set. The results show that this method achieves optimal noise reduction performance. Its effectiveness is validated through the output signal-to-noise ratio, root mean square error, and correlation coefficient. Overall, this method enhances noise reduction performance while preserving transient features. The method has been validated using real multi-layer penetration acceleration signals, supporting subsequent penetration layer identification and inversion analysis of the penetration process.