An Updateable Network Based on ReduNet for Radar Jamming Recognition

Radar jamming recognition provides prior information for anti-jamming measures, thus is a crucial part in modern battlefield. Feature extraction and SVM based algorithm is a mainstream method in practical applications. However, its robustness cannot be guaranteed when jamming parameters and external environments change. To address this issue, many methods based on deep neural networks have been studied. Nonetheless, deep neural networks often require a large number of samples, which is difficult to obtain in practice. Additionally, it is often impossible to obtain all data at once in practice, necessitating that the network has the ability to update in real-time without forgetting previously learned knowledge. This paper proposes an updatable neural network based on ReduNet. Simulation experiments demonstrate that this network achieves high accuracy with small samples. Furthermore, two update algorithms for this network is proposed, and experiment results show that these methods possess the ability of continual learning.