Electromagnetic signal recognition model based on multi-channel dilated dense convolutional networks

This paper addresses the issue of low accuracy in electromagnetic signal recognition when applying existing deep learning networks. It studies classic deep learning network solutions for electromagnetic signal recognition both domestically and internationally, comparing and analyzing the strengths and weaknesses of various approaches. Subsequently, it proposes an electromagnetic signal classification method based on multi-channel feature extraction and dilated convolutional neural networks. By extracting the signal graph features, spectrum graph features, and double-layer CNN learned features input into a dilated dense convolutional network model for classification and recognition. By constructing and training a model on the RADAR dataset, the experiment achieved a high recognition rate. Additionally, through ablation experiments, the importance and effectiveness of each component of the model were validated. Finally, this paper discusses the limitations of the model in complex electromagnetic environments and the directions for future improvement.