Bidirectional LSTM-Based Target Detection in Weibull and Gamma Clutter

Radar target detection performance depends on how accurately clutter can be characterized. However, depending on applications, it is difficult to accurately pre-dict clutter statistics and its parameters. The model-based detection algorithms that are optimized for one clutter scenario may fail to yield satisfactory results in another scenario. In this paper, we propose a bidirectional long short-term memory (LSTM) network to classify whether the received signal, in the presence of clutter, consists of target return or not. The key idea is to leverage the bidirectional LSTM's capability to learn from the time dependency of the input data sequence, both in forward and backward directions. The proposed network is trained using the test statistics of the clutter independent energy detector (ED), matched filter (MF) detector, and general-ized likelihood ratio test (GLRT) detector, as the features. Simulations results, conducted for Weibull and Gamma distributed clutter, show that the proposed bidirectional LSTM performs significantly better than the feed forward neural network (FFNN), especially when GLRT output is employed as a feature.