Unknown Radar Signal Recognition Method Based on SER Block and Semi-Hard Triplet Loss

Abstract

To solve the problem of misclassification about unknown radar categories, a method based on squeeze and excitation residual block (SER Block) and semi-hard triplet loss is proposed in this paper. First, the known radar signal is preprocessed to obtain its time-frequency images (TFIs). Second, the TFIs of the known radar signal is mapped to the 128-dimensional embedding space through the network built based on the SER Block. At the same time, use the semi-hard triplet loss function to adjust the network parameters, so that the signals of the same kind are more concentrated and the signals between classes are more dispersed in the embedded space. Finally, a method based on distance threshold is used to distinguish the unknown radar signals accurately, and to ensure the correct classification of the known radar signals. The experimental results show that the recognition accuracy of unknown radar signals is more than 89% at −8db, and that of known radar signals is more than 95%.