Investigating GF-5 Hyperspectral and GF-1 Multispectral Data Fusion Methods for Multitemporal Change Analysis

Abstract

Multitemporal change analysis is one of the essential purposes for discovering knowledge from various remote sensing terrestrial earth observation techniques. Particularly, the China Gaofen-5 (GF-5) hyperspectral imager provides a new data source for multitemporal change analysis. Its 330 bands, 60 km swath width and 5–10 nm spectrum resolutions make it captures subtle changes in spectrum responses of ground objects across different images. Unfortunately, its 30 spatial resolution still hinders its accurate geospatial location in some specific applications. Therefore, we explore state-of-the-art data fusion methods and seek an appropriate fusing method of GF-5 hyperspectral and GF-1 multispectral data to benefit multitemporal change analysis. We utilize four image fusion methods and implement six evaluation criteria to holistically evaluate the performance of different methods. Experimental results on three datasets of Taihu Lake and Poyang Lake in China show that the Modulation transfer functions-generalized Laplacian pyramid (MTF-GLP) has smaller spectral distortion, better spatial fidelity and requires moderate computational time than the other three methods. It accordingly can be a good choice for fusing GF-5 and GF-1 remote sensing data in both classification and multitemporal change analysis.