Detection and Classification of Grassland with Stellera chamaejasme through HJ-1A Hyperspectral Image in Northern Tibet

Abstract

Stellera chamaejasme is the most typical invasive plant in North Tibet. Monitoring the distribution of the Stellera chamaejasme is the prerequisite for grassland quality evaluation or management. In this paper, the hyperspectral image of HJ-1A satellite (HJ-1A HSI) was used to monitor the Stellera chamaejasme in the northern region of Xainza. Linear Spectral Unmixing (LSU) algorithm was used to get the abundances of six landuse types(include grassland with Stellera chamaejasme) based on HSI. Feature extraction method of principle component analysis (PCA) was used to compress the HSI information into five bands. And then, maximum likelihood classifier(MLC) was applied to the six abundances and the first five principal components for getting the classification result. The result showed that the classification result of LSU-MLC was better than PCA5-MLC, the mapping accuracy of Stellera chamaejasme grassland based on both methods were all greater than 91%, the user accuracy was all greater than 66%. We concluded that HJ-1A HSI can be used to detect and classify Stellera chamaejasme in northern Tibet. In addition, we got the spatial distribution characteristics of Stellera chamaejasme by overlay analysis of classification result with the terrain factors: the Stellera chamaejasme mainly distributed in the area with elevation between 4500 and 4800m, slope less than 10 degrees and mostly in the Southwest aspect, where radiation and hydrothermal condition were favorable.