Comparison of Image Fusion Methods for Gaofen-2 Panchromatic-Multispectral

To explore the fusion method suitable for GaoFen-2 satellite images, the efficiency of six state-of-the-art fusion methods, including Gram-Schmidt Transformation (GS), Adaptive GS Transform (GSA), Principal Component Analysis (PCA), Nearest-neighbor Diffusion-based Pan-Sharpening (NND), University of New Brunswick (UNB) method, and Subtractive Resolution Merge (SRM) method were compared. Meanwhile, a combination of direct and indirect evaluation methods including visual interpretation and quantitative index and classification accuracy were used to evaluate the fusion effects. The results showed that the types of ground cover in the study area were different, and the effects of different fusion methods were different. For instance, the GSA, GS and PCA methods performed well in spatial detail enhancement and spectral preserving, they were suitable for visual interpretation. From the classification point of view, SRM method was best for the application. UNB method had moderate effect in fusion, it can be an alternate method for fusing Gaofen-2 images. And NND method appeared a slight spectral distortion and had lower classification accuracy, so it was not recommended in the practical application.