AEWFNet: Adaptive Enhancement and Wavelet Convolution for Hyperspectral and Multispectral Image Fusion

Hyperspectral and multispectral image fusion (HMIF) represents a highly effective approach to enhancing the spatial resolution of hyperspectral images (HSIs). However, due to the inaccurate modeling of the spectral response function (SRF) and limited capability in capturing high-frequency details, many existing methods still struggle to maintain spectral consistency and effectively represent spatial structures. To address the aforementioned issues, this study proposes an Adaptive Enhancement and Wavelet Convolution-based HMIF framework, termed AEWFNet, which aims to improve the spatial structural representation and spectral reconstruction accuracy of the fused images. First, a Spatial-Aware Response Learning (SARL) mechanism is proposed to structurally optimize and guide the spectral degradation learning module, helping the model better understand the complex interactions between spatial structures and spectral information, thereby improving the fitting accuracy to the real degradation process. Second, an Attention-based Rotation Prediction Augmentation Network (ARPAN) is designed to optimize spectral information through an adaptive enhancement strategy, effectively improving the generalization performance and feature representation of images under various rotation conditions. Finally, wavelet convolution is employed to enhance the spatial feature representation across multiple scales. Experimental results on several datasets show that AEWFNet outperforms the compared methods in both quantitative and qualitative evaluations, achieving better preservation of the spatial-spectral characteristics of the images. The code for AEWFNet will be made publicly available at https://github.com/LRuiRui517/AEWFNet.