Wide-Band Self-Noise Suppression Method for The underwater Glider by Using Deep Learning

As an underwater unmanned systems (UAS) that can operate without propellers, unmanned gliders with integrated hydrophones have been widely used in long-term and large-scale ocean exploration missions. However, during the detection process, the target signal is often interfered by environmental noise or self-noise, which affects the detection effect. Therefore, it becomes very important to suppress the interference noise. Aiming at the suppression of glider self-noise, this paper discusses a method to suppress the self-noise of underwater gliders. In particular, a convolution network based on the Demucs model is established to separate and reduction the wide-band self-noise with continuous spectrum. The experimental data collected in China South Sea are used as training and test datasets. The effect of processing is verified by comparing the output before and after noise suppression. The results demonstrate that the proposed algorithm can suppress the wide-band self-noise without losing target features.