Robust Underwater Direction-of-Arrival Tracking Based on AI-Aided Variational Bayesian Extended Kalman Filter
Xianghao Hou, Yueyi Qiao, Boxuan Zhang, Yixin Yang
Abstract
The AI-aided variational Bayesian extended Kalman filter (AI-VBEKF)-based robust direction-of-arrival (DOA) technique is proposed to make reliable estimations of the bearing angle of an uncooperative underwater target with uncertain environment noise. Considering that the large error of the guess of the initial mean square error matrix (MSEM) will lead to inaccurate DOA tracking results, an attention-based deep convolutional neural network is first proposed to make reliable estimations of the initial MSEM. Then, by utilizing the AI-VBEKF estimating scheme, the uncertain measurement noise caused by the unknown underwater environment along with the bearing angle of the target can be estimated simultaneously to provide reliable results at every DOA tracking step. The proposed technique is demonstrated and verified by both of the simulations and the real sea trial data from the South China Sea in July 2021, and both the robustness and accuracy are proven superior to the traditional DOA-estimating methods.