PAN-Guided Band-Aware Multi-Spectral Feature Enhancement for Pan-Sharpening

Due to the physical hardware limits, multi-spectral (MS) images often suffer from low-spatial resolution, challenging their practical utility in real applications. Therefore, pan-sharpening technology has been widely explored as a popular tool to generate images with both high-spatial and high-spectral resolutions by integrating PAN and MS images. In this paper, we propose an effective pan-sharpening network, which consists of two core designs: a PAN-guided band-aware multi-spectral feature enhancement module and a multi-focus feature fusion module. To be specific, the former exploits the PAN features to perform band-aware multi-spectral feature modulation and selectively enhance the information of each spectral band while the latter covers various convolution kernels to extract multi-scale features, benefiting the fusion performance of the remote sensing scene. Equipped with the above core modules, our proposed framework is capable of achieving the best performance when compared with existing state-of-the-art Pan-sharpening methods over multiple satellite datasets. Extensive ablation studies are additionally conducted to verify the effectiveness of our key components both visually and quantitatively.