Deep learning for vessel identification leveraging coefficient of variation spectrum

Identifying individual vessels based on their radiated noise is challenging due to two key factors: (1) inter-identity similarity (IIS), where differentvessels often exhibit similar acoustic feature patterns, and (2) intra-identity variation (IIV), where the same vessel shows different feature patterns at operating conditions such as marine area, vessel speed, and operating time. To address these challenges, we utilize time-averaged spectral features rather than instantaneous snapshots, thereby mitigating the effect of transfer function in the ocean waveguide and other time variabilities. Furthermore, the coefficient of variation (CV) spectrum is incorporated with the time-averaged features as an additional descriptor representing the quality of information. Using modeled datasets and publicly available datasets, we analyze the effectiveness of the proposed features and develop a vessel identification framework based on a deep learning network. [Work supported by the Korea Research Institute for Defense Technology Planning and Advancement (KRIT) grant funded by the Korea government Defense Acquisition Program Administration (No. KRIT-CT-22-023-431 02, 2022).]