Abstract:

Aiming at the actual radar reconnaissance process that will receive unknown radar signals that are not available in a large number of sample libraries, this paper designs density based spatial clustering of applications with noise based on particle swarm optimization (PSO-DBSCAN) and a semi-supervised conditional generation adversarial network (SCGAN) to process the monopulse unknown radar signals. First, the optimal input parameters of the noisy density clustering algorithm are obtained through the particle swarm optimization (PSO), and then the unknown radar signals are clustered, and the distance filtering algorithm is used to filter out more credible samples from the clusters output by the clustering algorithm, which are extended to the radar sample library. When too many types of unknown radar signals are added, the feature extraction network needs to be expanded and trained, and the small amount of data in the sample library is difficult to support the feature extraction network for effective expansion training. Therefore, a semi-supervised conditional generation confrontation network is designed to realize the training and classification of unknown signals in the case of small samples. The simulation results show that this method performs well in the recognition of unknown radar signals.

Key words: unknown radar signal recognition, particle swarm optimization (PSO), density-based spatial clustering of applications with noise (DBSCAN), distance filtering algorithm, semi-supervised conditional generation adversarial network (SCGAN)