Research on Denoising Methods for Infrasound Leakage Signals Using Improved Wavelet Threshold Algorithm

Abstract

Infrasound leakage signals, with low propagation energy loss, are ideal for long-distance and small leakage detection but suffer severe background noise interference. Existing wavelet denoising methods using traditional soft/hard threshold functions face critical limitations: soft thresholds introduce constant deviation, while hard thresholds cause discontinuities, both leading to suboptimal noise reduction for infrasound signals—this gap hinders accurate leakage detection. To address this, we propose a wavelet denoising method with an improved threshold function, analyze its process via the Mallat algorithm, and prove its continuity and convergence. Comparative experiments on infrasound leakage data show that, at the optimal decomposition level, our method reduces RMSE by 41.19% and increases SNR by 5.1326 dB compared to the soft threshold method; versus the hard threshold method, RMSE decreases by 34.65% and SNR increases by 4.2148 dB. It also separates background noise more thoroughly in time–frequency domains, demonstrating strong feasibility for pipeline infrasound leakage detection.