Sea/Land Clutter Recognition for Over-The-Horizon Radar via Deep CNN

摘要

Over-the-horizon radar (OTHR) is of significance in persistent surveillance. To localize targets, coordinate registration (CR) has to be carried out to transform the coordinates of the targets from the slant coordinate system to the ground coordinate system. An alternative way to improve the accuracy of CR is the utilization of sea-land transitions and islands. The key novelty of this paper is a solution of recognizing sea/land clutter based on the range-Doppler spectrum of OTHR. We propose a deep convolutional neural network (DCNN) with multiple hidden layers to learn features of different levels directly from the input R-D spectrum of sea/land. With the help of massive training data, the results of the experiment show that the proposed DCNN performs better than the support vector machine method and the least-mean-square method.