Radar signal pattern recognition method based on multimodal feature fusion

With the rapid development of modern radar technology, radar signal recognition faces new challenges and opportunities. This study proposes a method of parallel residual shrinkage network with multimodal feature extraction and fusion (PRSNet-MFM). First, a parallel residual and shrinkage convolution network (PRSNet) is introduced to extract intra- and inter-pulse features of radar signals independently. The attention mechanism in the residual and shrinkage module (RSM) is utilized to extract valid features and suppress noisy information. Finally, the multimodal feature fusion module (MFM) dynamically adjusts the fusion weights according to the importance of the features to integrate and utilize the modal information more effectively. Experimental results show that the proposed method significantly outperforms signal recognition methods based on single features or simple feature fusion, particularly in addressing parameter overlap and noise interference in complex electromagnetic environments, achieving an average recognition accuracy of over 95% under low signal-to-noise ratio conditions.