Source depth estimation using spectral transformations and convolutional neural network in a deep-sea environment.

Area of Science:

• Oceanography
• Acoustic signal processing
• Deep-sea exploration

Background:

• Accurate source depth estimation is crucial for underwater acoustic applications.
• Traditional methods like Multispectral Transformation for Depth Estimation (MSTDE) face limitations with real-world sound-speed profiles and source bandwidths.
• Noise and environmental variability impact the precision of depth estimation techniques.

Purpose of the Study:

• To evaluate and compare different depth estimation methods in deep-ocean environments.
• To introduce and assess a novel deep learning approach for enhanced depth estimation accuracy.
• To investigate the effectiveness of transfer learning in mitigating noise interference.

Main Methods:

• Derivation of the Multispectral Transformation for Depth Estimation (MSTDE) method.
• Development of a Convolutional Neural Network (CNN) based approach combined with Conventional Beamforming (CBF) preprocessing.
• Application of transfer learning to improve robustness against noise.
• Validation using at-sea experimental data.

Main Results:

• MSTDE provides depth estimation but exhibits increasing error with distance.
• Calculated factors can moderately compensate for MSTDE errors.
• The deep learning approach with CBF preprocessing demonstrates superior performance compared to MSTDE and traditional CNN methods.
• Transfer learning effectively addresses noise-related estimation degradation.

Conclusions:

• The proposed deep learning method significantly advances underwater acoustic source depth estimation.
• Conventional beamforming preprocessing is key to the success of deep learning in this domain.
• Future research should focus on refining deep learning models for even greater accuracy and robustness in diverse oceanographic conditions.