Abstract: To address the limited spatial resolution of hyperspectral satellite imagery,this study proposes a data fusion method driven by deep spatio-spectral-temporal features.The method integrates the rich spectral information of hyperspectral images with the fine spatial details of multi-spectral data,aiming to generate fused imagery with both high spectral and spatial resolution.The fusion network is built upon a generative adversarial network (GAN)architecture,with an optimized feature fusion strategy that significantly enhances the network's capability in handling multi-resolution data.Experiments conducted on a comprehensive dataset,comprising both spaceborne and airborne hyperspectral imagery,demonstrate that the proposed method notably improves image quality,outperforming conventional approaches in terms of spatial detail preservation and spectral consistency.Quantitative evaluation using multiple metrics further confirms the robustness and effectiveness of the method.This study provides essential technical support for the enhancement and application of hyperspectral remote sensing imagery,offering important theoretical and practical value.

Key words: hyperspectral image, multi-spectral image, data-level fusion, generative adversarial network, hyperspectral data fusion