Multi-polarization Features Fusion Detection of Marine Small Targets based on LSTM

This article introduces a novel detector based on long short-term memory (LSTM) networks which intelligently fuses multi-polarization features for detection of the marine small targets. Specifically, complex Doppler features of data with different polarization are designed to provide basis for differentiation of the target and the clutter. Then, LSTM cells are utilized to capture temporal dependencies of both amplitude and phase of the Doppler features, which have powerful sequence extraction capability. After that, the detector adaptively assigns the fusion weights into features of different polarizations to carry out weighted feature fusion. Finally, the detection results are given according to final binary classification results of clutter and target. Experiments on public IPIX database validate the effectiveness of the proposed fusion scheme and demonstrate that the proposed detector surpasses the detector that uses only single polarization features.