Visualization and classification of Radar Emitter Pulse Sequences based on 2D feature map

Abstract

With the development of modern electronic countermeasure technology, radar reconnaissance equipment receives pulse sequences of increasing density and complexity. Achieving concise as well as robust feature representation of Radar Emitter Pulse Sequences (REPS), and improving the classification and recognition performance of REPS signals have always been the focus at present. Current research did not consider the distribution information and co-information buried in REPS data. To solve this problem, we propose to visualize and classify REPS signals based on 2D feature map. First, we extract the Sequence Distribution Property (SDP) of REPS for spatial coding to obtain an encoded feature vector that is deformed into a 2D matrix. Three types of informative 2D matrices generated from REPS are combined into a color 2D feature map. Different from previous feature extraction methods, the proposed method can not only express the single sequence information concisely and accurately but also express the correlation of multiple parameter sequences synchronously in the interaction of different color channels. Second, based on the 2D feature map, we design a ‘L&FCN’ framework for classification, in which we fed the 2D feature map into a network composed of lower convolution neural layers and fully connected layers to judge the category and output recognition result. Finally, we put forward a novel strategy with integrating global and local segments’ information of REPS to train the network, making it converge with smaller classification errors. Simulation results verify the effectiveness and superiority of our method.