DIEFEN

DIEFEN: Differential Information-Enhanced Feature Exchange Network for Hyperspectral Change Detection

The proposed DIEFEN mainly consists of a channel-spatial interaction (CSI) module, a feature focus and long-range attention (FFLA) module and an enhanced differential multi-head attention (EDMA) module.

The CSI integrates channel exchange and spatial exchange mechanisms to align bitemporal features and to diminish the influence of noise. The FFLA consists of a feature compression layer, a large kernel attention and a light-weight layer, which facilitates the extraction of discriminative features while simultaneously reducing computational complexity. The EDMA module is constructed to guide the original HSIs using difference information, which can effective suppress unchanged pixels while simultaneously highlight changed pixels.