Hyperspectral Image Classification Based on Multi-Source Cross-Domain Few-Shot Learning

Hyperspectral remote sensing images (HSI) face classification challenges due to sensor and environmental impacts, limited training sets, spectral homogeneity, and difficulties integrating multi-source data. This project aims to study HSI classification methods based on multi-source assisted few-shot learning. Targeting the Hughes phenomenon caused by limited training samples, we will investigate semi-supervised cross-domain few-shot learning methods to enhance the generalization of classification models.