Radar HRRP Target Recognition with Recurrent Convolutional Neural Networks

Conventional radar automatic target recognition (RATR) methods using High-Resolution Range Profile (HRRP) sequences require carefully designed feature extraction techniques and plenty of HRRP waveforms, which result in insufficient recognition rate and limit in realtime recognition. To address these issues a modified end-to-end architecture consisting of a convolutional neural network (CNN) followed by a recurrent neural network (RNN) is proposed. In this model the local features of HRRPs extracted by a CNN are passed to a RNN, which avoids manual feature extraction and takes advantage of its shared parameters mechanism which enables single HRRP recognition in real-time. The effectiveness of this model is shown in this paper with numerical results.