Deep Learning-Based Source Depth Discrimination with Frequency-Wavenumber Features and a Single Hydrophone

Accurate discrimination of surface and underwater sources is essential for marine target detection, but remains challenging in single-hydrophone scenarios due to the lack of spatial diversity. Conventional methods like matched mode processing (MMP) rely on hydrophone arrays and precise environmental information, while synthetic aperture beamforming requires long source trajectories, limiting its practical applicability. To address these limitations, this paper proposes a deep learning-based approach that leverages the shape characteristics of wavenumber spectra across multiple frequencies as neural network inputs. This method enables effective source depth discrimination using a single hydrophone and short aperture, reducing dependence on array configurations and extensive source movements. Simulation results indicate that the proposed method outperforms conventional techniques across varying apertures, source starting distances, SNR levels, and environmental mismatches, demonstrating its robustness and versatility in real-world scenarios.