Underwater Passive Target Tracking Based On CNN-LSTM-Attention

Abstract

In order to solve the problem of large measurement error and mismatching of target motion model caused by complex underwater environment, we propose a CNN-LSTM-Attention (CLA) network based target tracking algorithm. First, CNN is employed to extract target features from multivariate time sequences. Then, the target trajectory is derived via LSTM due to its excellent representation of the time dependence. Further, an attention layer is added to model the important spatiotemporal features of moving target to improve tracking the accuracy. The experiments and analyses of trajectories with different starting states, speeds and turning rates show that our proposed algorithm can obtain the minimum RMSE. Besides, compared with the traditional model-based target tracking method, our proposed CLA does not require the target motion model in advance, and can make it better suited to complex noise interference. Furthermore, our proposed CLA algorithm performs better than the LSTM based target tracking algorithm.