An Underwater Acoustic Signal Denoising Algorithm Based on U-Net

Abstract

Active sonar signal detection is crucial for sonar systems. However, developments in submarine stealth and vibration reduction have resulted in a continuous decrease in the signal-to-noise ratios (SNRs) of the active sonar received signals, which poses significant challenges for traditional denoising methods. To address the issue, this paper proposes an underwater acoustic signal denoising algorithm based on U-Net. The algorithm first introduces a multi-scale convolution module that extracts features from different receptive fields to improve the network's feature extraction capability. Then, an attention mechanism is integrated into U-Net that enables the network to selectively focus on the content and location within the feature maps. Furthermore, the proposed method incorporates an interference signal with a frequency close to the target signal, which enables the network to automatically learn the distinction between the target signal and the interference signal during training, thereby enhancing its anti-interference capability. The simulation results in both Gaussian white noise background and ocean ambient noise background demonstrate that the proposed algorithm outperforms traditional denoising methods and classic image denoising algorithms in terms of denoising performance.