Localization of Immersed Sources by Modified Convolutional Neural Network: Application to a Deep-Sea Experiment.

Area of Science:

• Signal Processing
• Machine Learning
• Acoustics

Background:

• Accurate source ranging using acoustic data is crucial for underwater applications.
• Traditional methods often face limitations in reliability and precision.
• Vertical arrays are commonly used for acoustic data acquisition.

Purpose of the Study:

• To develop a more reliable method for acoustic source ranging.
• To enhance the performance of convolutional neural networks (CNNs) for ranging tasks.
• To improve the accuracy of distance estimation using acoustic field data.

Main Methods:

• A modified convolutional neural network (CNN) architecture was developed.
• The CNN's output layer was adapted to produce Gauss regression sequences.
• The model was trained and validated using deep-sea experimental acoustic data.

Main Results:

• The modified CNN with Gauss regression output outperformed traditional methods.
• The mean relative error between predicted and actual distances was approximately 2.77%.
• High positioning accuracy was achieved, with 99.56% at 10% error and 90.14% at 5% error.

Conclusions:

• The proposed CNN with Gauss regression offers enhanced reliability for acoustic source ranging.
• This approach provides a significant improvement over single regression and classification outputs.
• The method demonstrates strong potential for practical underwater acoustic positioning applications.