关键词: 高光谱影像模拟, 光谱匹配, 场景光谱库, 归一化植被指数

Abstract: Hyperspectral image can obtain fine spectral information of ground objects, and is a data source for fine identification of ground objects and high-precision inversion of parameters. However, due to its own characteristics, hyperspectral sensors often have low resolution and long coverage periods, which limits their application and promotion. In order to improve the timeliness of hyperspectral images, researchers have carried out a lot of research on hyperspectral image simulation. However, most of the existing methods are based on the ideal spectrum library, which is quite different from the actual scene spectra. A hyperspectral image simulation method is constructed based on a scene spectral library, and a spectral matching algorithm is proposed based on normalized difference vegetation index (NDVI), which greatly improves the speed and accuracy of simulation. The method was tested in the southwest of Dezhou City, Shandong Province, China, and the GF-1 WFV multispectral data were used as the template images to simulate the GF-5 AHSI load data, and at last, the simulation results was compared with the traditional simulation methods. The comparison results show that the simulation results of this method are good, with an average R2 of 0.69 for 283 effective bands, which is 0.16 higher than that of the traditional simulation method based on category retrieval. However, the simulation effect of each band is different, among which the best simulation effect is at band 71, where R2 reaches 0.81, which is 0.18 more than the simulation method based on category matching retrieval. In terms of operation efficiency, the simulation time of the new method is greatly shortened and the simulation speed is increased by 75%.

Key words: hyperspectral image simulation, spectral matching, scene spectrum library, normalized difference vegetation index