A Novel Three-Branch FIMA-Net for Millimeter-Wave Radar Gesture Recognition

Gesture recognition using millimeter-wave radar has become a significant research area in human-computer interaction. However, existing gesture recognition algorithms are usually developed based on large datasets with high complexity. In this paper, a novel three branch feature enhanced inception multi-head attention network (termed as FIMA-Net) for millimeter-wave radar gesture recognition is proposed. In order to improve the model’s recognition ability, a feature augmentation algorithm is also proposed based on logarithmic compression and softmax normalization. Experimental results demonstrate the effectiveness of the proposed method in various gesture recognition tasks.