Radar Working Mode Recognition Method Based on Contrastive Pulse Description Graph Learning

With the development of radar technology, radar working modes are diverse and signal waveforms are complex and variable. However, existing radar working mode recognition methods heavily rely on prior knowledge and have poor generalization ability, making it difficult to adapt to increasingly complex real electromagnetic environment problems. In this paper, a contrastive pulse description graph learning (CPDGL) based approach for radar working mode recognition is proposed. The pulse description word (PDW) sequence is first intercepted by a certain sliding window length without the use of labels. Secondly, the radar PDW sequence that was intercepted is transformed into a pulse description graph (PDG), allowing for the successful conversion of working pattern recognition into picture categorization. Lastly, unsupervised learning of PDG data characteristics and working mode detection using contrastive learning method. According to simulation results, when complete sequence data is available, radar working mode detection accuracy can reach 96.71%. The approach suggested in this research can still achieve high accuracy in working mode recognition even in the presence of noisy pulses, poor interception conditions, and pulse loss. Simulating and comparing with common clustering techniques like Agglomerative clustering, DBSCAN, and K-means further highlights the superiority of the CPDGL-based radar working mode detection method.