A review on denoising methods for underwater explosion shockwave pressure signals

Abstract Underwater explosion shockwave pressure signals are critical for assessing damage effects of weapons. However, noise contamination renders these signals non-stationary, broadband, and transient characteristics, making high-fidelity denoising a bottleneck for accurate parameter extraction. This paper reviews recent denoising advances, classifying mainstream approaches into three categories: traditional filtering and transform-based methods, adaptive decomposition, and deep learning. Tracing the technological evolution, this review delineates core mechanisms and boundaries of representative methodologies: time-frequency analyses, Fourier transform (FT) and wavelet transform (WT); adaptive decomposition, empirical mode decomposition (EMD) and variational mode decomposition (VMD); and deep learning models, convolutional neural networks (CNN), Transformers, and diffusion models. Traditional methods, despite their interpretability and computational efficiency, struggle to balance transient preservation with noise suppression. Adaptive decomposition methods have data-driven advantages but are constrained by parameter sensitivity. Deep learning approaches exhibit strong nonlinear mapping capabilities, yet face data scarcity and limited interpretability. Future breakthroughs require: the development of mechanisms for high-fidelity preservation of transient features, few-shot and weakly supervised learning, physics-informed data-driven models, multi-scale signal modeling frameworks, and standardized datasets and evaluation systems. This review provides a systematic methodological reference for underwater explosion signal processing and damage assessment.