Deblurring of Sound Source Orientation Recognition Based on Deep Neural Network

Area of Science:

• Information disciplines
• Acoustic signal processing
• Artificial intelligence in engineering

Background:

• Traditional underwater target detection faces limitations due to sonar array errors and signal instability.
• Conventional beamforming methods struggle with low signal-to-noise ratios (SNR < -43.05 dB), offering only vague directional identification.
• Existing high-resolution array signal processing methods are impractical for underwater applications.

Purpose of the Study:

• To develop an advanced beamforming method for enhanced underwater target detection and identification.
• To overcome the limitations of traditional methods in low SNR environments.
• To achieve high-resolution recognition and prediction of underwater targets using deep learning.

Main Methods:

• A deep neural network (DNN) based beamforming method is proposed.
• Underwater target sound signals are preprocessed to convert space-time data into an angle-time domain representation.
• The DNN is trained using extensive sample datasets for robust performance.

Main Results:

• The proposed DNN beamforming method achieves high-resolution recognition and prediction of underwater targets.
• The method demonstrates effectiveness in a test dataset, successfully detecting targets at a minimum SNR of -48 dB.
• This represents a significant improvement over conventional methods in challenging low-SNR conditions.

Conclusions:

• Deep neural network beamforming offers a superior solution for underwater target detection and identification.
• The proposed method significantly enhances detection resolution and accuracy, particularly in low SNR environments.
• This research advances the application of AI in underwater acoustic signal processing and engineering.