End-to-end underwater acoustic source separation model based on EDBG-GALR

Area of Science:

• Underwater acoustics
• Signal processing
• Machine learning

Background:

• Overlapping acoustic signals from multiple vessels in close proximity cause interference, reducing vessel identification accuracy.
• Existing underwater acoustic source separation algorithms may have limitations in expressiveness and local modeling capabilities.

Purpose of the Study:

• To propose an improved end-to-end underwater acoustic source separation algorithm, EDBG-GALR.
• To enhance the capability of processing temporal signals and improve separator input efficiency.
• To improve local modeling ability of sequence features while reducing computational requirements.

Main Methods:

• Developed an improved GALR algorithm named EDBG-GALR.
• Introduced a deep encoder, ECA-DE (Efficient Channel Attention-Deep Encoder), to enhance temporal signal processing.
• Integrated Bi-GRU (Bidirectional Gated Recurrent Unit) into the separation block for improved local modeling.

Main Results:

• The EDBG-GALR method effectively separates mixed multi-target underwater acoustic signals.
• Achieved a maximum scale-invariant signal-to-noise ratio (SI-SNR) improvement of 3.32 dB.
• Achieved a maximum signal distortion ratio (SDR) improvement of 3.38 dB compared to baseline methods.

Conclusions:

• The proposed EDBG-GALR algorithm demonstrates significant improvements in underwater acoustic source separation.
• The enhancements in the encoder and separation blocks contribute to better performance.
• EDBG-GALR shows practical applicability in complex underwater environments for accurate vessel identification.