A Sustainable Learning Method for Radar Jamming Recognition Based on Open Set

As the modern electromagnetic countermeasure environment becomes increasingly complex, more and more types of jamming are emerging. Current jamming recognition methods mainly rely on extensive pre-training data and prior knowledge, and lack the capability to identify unknown jamming and do not have the ability to learn dynamically. In this paper, a sustainable learning method for radar jamming recognition based on open set is proposed, which consists of three parts: “open set recognition + intra-class classification + pseudo-sample generation”, to realize effective recognition and sustainable learning of unknown jamming. Firstly, a jamming open set recognition network based on OpenMax is proposed by combining the residual structure and the attention module. Secondly, Siamese network is used to measure the similarity between the query sample and the obtained unknown jamming sample, in order to correct the misjudgment of the open set recognition network. Then, the intra-class classification of unknown jamming is performed to obtain specific unknown jamming labels. Finally, the pseudo samples of jamming are generated based on unknown jamming labels by generative learning, so as to provide sufficient training data of unknown jamming for sustainable learning tasks. The experimental results of the classification of jamming signals show that the method can effectively identify unknown jamming signals and has excellent generalization performance. When the JNR is 0 dB, the presented network achieves an average jamming recognition accuracy of 97.4% for signals containing two types of unknown jamming. Meanwhile, the sustainable learning system can attain an average jamming recognition accuracy of 94.9% for signals containing three types of unknown jamming.