Memory-enhanced temporal feature learning framework for improved underwater maneuvering target tracking

Area of Science:

• Robotics and Control Systems
• Signal Processing
• Ocean Engineering

Background:

• Tracking underwater passive maneuvering targets presents significant challenges due to rapid trajectory shifts and high measurement noise.
• Existing methods struggle to maintain accurate state estimation under dynamic conditions and acoustic interference.

Purpose of the Study:

• To develop an innovative framework for robust underwater passive target tracking.
• To enhance the prediction of target location, velocity, and trajectory using memory-augmented temporal features.

Main Methods:

• Implementation of a Long Short-Term Memory (LSTM) model for learning temporal relationships in target state formation.
• Configuration of the LSTM model using state-space physics.
• Evaluation under varying levels of Gaussian measurement distortion and comparison with standard estimators.

Main Results:

• The LSTM-based model significantly reduces state estimation errors compared to generalized pseudo-Bayesian estimators like the Interacting Multiple Model Extended Kalman Filter and Unscented Kalman Filter.
• Demonstrated superior performance in maintaining accurate predictions during high maneuverability and high noise levels.
• Achieved a notable decrease in mean squared error (MSE) for state estimation.

Conclusions:

• The proposed memory-augmented LSTM framework offers a feasible and effective solution for real-time passive tracking in challenging underwater acoustic environments.
• The model exhibits significant flexibility for diverse maneuvering behaviors, outperforming conventional tracking algorithms.