State Estimation of an Underwater Markov Chain Maneuvering Target Using Intelligent Computing

Area of Science:

• Marine robotics
• Artificial intelligence
• Signal processing

Background:

• Underwater state estimation is crucial for marine operations.
• Traditional methods like Kalman filters struggle with highly maneuvering targets.
• Deep learning offers potential for enhanced tracking accuracy.

Purpose of the Study:

• To propose a deep learning-based strategy for efficient real-time state estimation of underwater maneuvering objects.
• To investigate the robustness and precision of the NARX neural network for passive Markov chain targets.
• To evaluate the performance in various ocean conditions and noise levels.

Main Methods:

• Utilized a nonlinear autoregressive with exogenous input (NARX) neural network model.
• Modeled a continuous coordinated turning trajectory for an underwater object.
• Developed state estimation within a bearings-only tracking framework.
• Conducted Monte Carlo simulations with varying noise conditions.

Main Results:

• The NARX neural network accurately estimated the real-time position and velocity of the maneuvering object.
• Demonstrated superior performance over conventional generalized pseudo-Bayesian filtering algorithms (IMM-EKF, IMM-UKF).
• Validated competence in both ideal and complex ocean environments.

Conclusions:

• Deep learning-based NARX neural computing provides an efficient and accurate solution for underwater maneuvering object state estimation.
• The proposed method offers a robust alternative to traditional filtering techniques.
• NARX networks show significant potential for real-time applications in marine surveillance and navigation.