Enhancing feature-aided data association tracking in passive sonar arrays: An advanced Siamese network approach

Area of Science:

• Signal Processing
• Machine Learning
• Ocean Acoustics

Background:

• Traditional multi-target tracking in passive sonar arrays relies on kinematic data, which is insufficient in complex marine environments.
• Existing feature-aided methods struggle with low signal-to-noise ratios and close-proximity targets due to raw feature utilization.

Purpose of the Study:

• To develop an advanced feature-aided data association method for passive sonar multi-target tracking.
• To improve tracking accuracy and robustness in challenging underwater acoustic scenarios.

Main Methods:

• Proposes BiChannel-SiamDinoNet, a Siamese network integrated into a joint probability data association framework.
• Utilizes an embedding space for acoustic target features to enhance discrimination.
• Refines feature extraction for underwater acoustic signals and employs knowledge distillation for improved feature consistency assessment.

Main Results:

• The BiChannel-SiamDinoNet demonstrates enhanced robustness to variations and complex target relationships.
• The method effectively discriminates between measurements and targets, improving data association.
• Performance validated through simulations and marine experiments.

Conclusions:

• The proposed BiChannel-SiamDinoNet significantly improves feature-aided multi-target tracking for passive sonar arrays.
• The deep learning approach offers a robust solution for complex underwater acoustic environments.
• This method enhances the reliability of tracking systems in challenging marine scenarios.