Enhancing Hyperspectral Image Classification and Segmentation with GNN and Ensemble U-Net

Hyperspectral image (HSI) analysis presents unique challenges due to the high dimensionality and complex spatial and spectral data involved. To address these, we propose a novel framework that integrates Graph Neural Networks (GNNs) with a Clustering Ensemble U-Net (CEU- Net) to enhance both classification and segmentation performance in HSI tasks. The GNN component captures intricate spatial relationships by modeling contextual dependencies between pixels, while the CEU-Net efficiently processes spectral information by leveraging a clustering-based ensemble of UNets. Using the Sentinel-2 dataset, the proposed framework achieves a classification accuracy of 95.4% and segmentation accuracy of 92.7%, significantly outperforming existing methods. This hybrid approach provides a robust and efficient solution, advancing the field of hyperspectral image analysis with applications spanning from precision agriculture to urban land cover mapping.