Particle Filter-Guided Online Neural Networks for Multi-Target Bearing-Only Tracking in Passive Sonar Systems.

Area of Science:

• Acoustic signal processing
• Machine learning for target tracking
• Passive sonar systems

Background:

• Multi-target tracking in passive sonar is challenging due to signal instability.
• Existing methods struggle with low signal-to-noise ratios (SNR) and target trajectory crossings.
• Accurate tracking is crucial for naval operations and underwater surveillance.

Purpose of the Study:

• To develop a robust method for stable multi-target bearing-only tracking in passive sonar.
• To improve tracking accuracy and reduce the minimum SNR required for reliable performance.
• To enhance the continuity and smoothness of target trajectories.

Main Methods:

• A particle filter-guided on-site training mechanism simplifies multi-classification to binary classification.
• An independent tracker is assigned to each target for simultaneous training and deployment.
• A hybrid Convolutional Neural Network (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) network is utilized.
• CNN enhances feature extraction and non-target discrimination; BiLSTM models spatiotemporal dependencies.

Main Results:

• The proposed method reduced the minimum required SNR for stable tracking to -31.78 dB, compared to -29.69 dB for pure particle filtering.
• Average tracking error decreased from 0.61° to 0.34°.
• Stable tracking was maintained even during target trajectory crossings in simulations and sea trials.

Conclusions:

• The hybrid CNN-BiLSTM network effectively addresses instability in multi-target bearing-only tracking.
• The method significantly enhances tracking accuracy and robustness in complex underwater acoustic environments.
• This approach offers a substantial advancement for passive sonar multi-target tracking capabilities.