Source localization in the deep ocean using a convolutional neural network.

Area of Science:

• Oceanography
• Acoustics
• Machine Learning

Background:

• Deep-sea source localization is crucial for various applications.
• Existing methods often struggle with accurate range and depth estimation, leading to significant errors.
• Limitations in current techniques necessitate advanced approaches for reliable underwater acoustic localization.

Purpose of the Study:

• To introduce a highly accurate convolutional neural network (CNN)-based method for deep-sea source localization.
• To address the limitations of existing methods in simultaneously estimating source range and depth.
• To solve the source localization problem as a regression task using a novel neural network architecture.

Main Methods:

• A convolutional neural network (CNN) was developed and trained using normalized acoustic matrices.
• The source localization problem was framed as a regression task for the CNN.
• The trained CNN was employed to predict the three-dimensional position of acoustic sources.

Main Results:

• The proposed CNN-based method demonstrated high accuracy in experimental data from the western Pacific.
• The mean absolute percentage error for range estimation was 2.10%.
• The mean absolute percentage error for depth estimation was 3.08%.

Conclusions:

• The CNN-based approach provides a significant improvement in deep-sea source localization accuracy.
• This method effectively overcomes the challenges associated with range and depth estimation errors in existing techniques.
• The study validates the performance of the proposed method using real-world experimental data, highlighting its practical applicability.