Swin–MRDB: Pan-Sharpening Model Based on the Swin Transformer and Multi-Scale CNN

Pan-sharpening aims to create high-resolution spectrum images by fusing low-resolution hyperspectral (HS) with panchromatic (PAN) images. Inspired the Swin transformer used in image classification tasks, this research constructs a three-stream pan-sharpening network based on and multi-scale feature extraction module. Unlike traditional convolutional neural (CNN) model, we use establish global connections combine it module extract local features of different sizes. The model combines advantages CNN, enabling fused maintain good detail linkage mitigating distortion In order verify effectiveness method, paper evaluates subjective visual quantitative indicators. Experimental results show that method proposed can better preserve spatial spectral information compared classical latest models.