A Hybrid Vision Transformer-based Capsule Network for Radar Automatic Modulation Recognition

Abstract

In radar Automatic Modulation Recognition (AMR), Single neural networks fail to achieve a satisfactory recognition accuracy especially in low SNR conditions. In this paper, we propose a novel AMR framework called A Hybrid Vision Transformer-based Capsule Network (HVTCN) that integrates Vision Transformers (ViT) with Capsule Networks (CapsNet) to enhance recognition performance. The ViT extracts global dependencies from radar spectrograms, while the CapsNet maintains spatial relationships, improving classification accuracy. Evaluations on benchmark datasets demonstrate superior performance under varying signal conditions.