Dual-Branch Fourier-Mixing Transformer Network for Hyperspectral Target Detection

In recent years, transformers have shown great potential in hyperspectral image processing and also been gradually applied target detection (HTD). Nonetheless, applying a typical transformer to HTD remains challenging. The heavy computation burden of the multi-head self-attention (MSA) limits its efficient HTD, while limited ability extract local spectral features can reduce discrimination learned features. To further explore for balance representation computational efficiency, we propose dual-branch Fourier-mixing network (DBFTTD). First, this work explores network. transformer-style replaces MSA sublayer with sublayer, which shows advantages improving efficiency learning valuable information effectively HTD. Second, proposes learnable filter ensembles Fourier domain that are inspired by ensemble improve performance. Third, simple but dropout strategy is proposed data augmentation. Sufficient balanced training samples constructed network, Experiments on four sets indicate our detector superior state-of-the-art detectors.