Radar Signal Recognition Method Based on Knowledge Distillation and Attention Map

In order to solve the problem that traditional radar signal recognition methods can not effectivelyexpand the recognition types, a radar signal recognition method based on knowledge distillation and attentionmap is proposed. Firstly, the Smooth Pseudo Wigner-Ville Distribution (SPWVD) of the radar signal is used as input;Then, the incremental learning network structure based on residual network is designed, and the loss functionbased on knowledge distillation and attention map is used to alleviate the catastrophic forgetting in the processof category increment; Finally, a method based on the mean distance of sample features is used to manage thedata set, which reduces effectively the occupied storage resources. Experiments show that this method canquickly complete the training of the extended classification signal under the condition of limited storageresources, and has good recognition accuracy for the original classification and the extended classification signal.