The improved semi-supervised active learning for radar signal recognition

In the process of radar signal recognition, fewer training samples is a common and challenging problem. A novel algorithm named improved semi-supervised active learning is proposed for signal classification, which is based on pseudo-labels verification procedure. For the problem of low radar signal recognition in complex electromagnetic environment, the time-frequency analysis of radially Gaussian kernel is applied to radar signals. Through singular value decomposition of time-frequency distribution, it extracts its singular values as feature parameters for radar signal recognition. In order to overcome the shortcomings of the traditional semi-supervised active learning algorithm, a classifier is constructed using an improved semi-supervised active learning algorithm. The proposed algorithm enables a collaborative labeling procedure by both human experts and classifiers to acquire more confidently labeled samples to improve the final classification performance and realize the high probability of radar signal recognition when the number of available labeled samples is small. Simulation results show that the proposed feature recognition method can achieve higher radar signal recognition at low SNR.