Angle-distance decomposition based on deep learning for active sonar detection

Area of Science:

• Marine sciences and engineering
• Underwater acoustics
• Signal processing

Background:

• Traditional active sonar target detection methods struggle with noise, reverberation, and interference in complex underwater environments.
• Effective underwater acoustic data is limited, posing challenges for training deep learning models.
• Accurate underwater target detection is crucial for marine sciences and engineering applications.

Purpose of the Study:

• To develop a robust deep learning-based method for active sonar target detection.
• To address the limitations of traditional signal processing in challenging underwater conditions.
• To improve the accuracy and reliability of underwater target localization.

Main Methods:

• A deep learning approach is proposed, decomposing target detection into angle and distance estimation.
• Deep learning models are employed to predict target distance and angle.
• Transfer learning and simulation techniques are utilized to overcome limited underwater acoustic data.

Main Results:

• The proposed method effectively detects underwater targets in complex acoustic environments.
• Experimental results demonstrate robust performance under challenging conditions, including noise and interference.
• Integrating estimated distance and angle provides accurate final target positioning.

Conclusions:

• Deep learning offers a promising solution for enhancing active sonar target detection capabilities.
• The proposed method provides a more effective and robust alternative to traditional signal processing techniques.
• Transfer learning and simulation are viable strategies for addressing data scarcity in underwater acoustics research.