Study On Land Cover Classification Based On Decision Tree Method Using Aerial Hyperspectral Imagery

Land cover classification is the basis of surveying, planning and dynamic monitoring land resource, and is the effective means of studying environment effect, ecosystem security and global changing, therefore it has the important signality. There is abundant spectral information including expressing biologic, physical and chemical character of ground objects in every pixel of hyperspectral remote sensing imagery. Studying objects can be divided and recognized clearly and gradually by selecting different remote sensing data source and feasible data analysis methods.The paper takes Yixing area, Jiangsu province, China as an example. The workflow and method of hyperspectral remote sensing land cover classification using decision tree technology is presented in detail. Spectral characters of typical land cover types in research area are analyzed. Common reflectivity inversion methods of imaging spectral data obtained by OMIS /is summarized, and empirical line method is experimented according to the existing data. Post processing method is put forward in order to removal serration of spectral curve after calibration. Next, the feature is selected and extracted, and at the same time, several kinds of traditional classification methods(Maximum Likelihood Classification . Minimum Distance Classification Spectral Angle Mapping and so on) on pre-processed data are experimented on and analyzed. Features, methods and parameters in the process of building decision tree classification module suitable to test area are digged out based on the above experiments. In the end, decision tree classification experiments results and contrastive precision accuracy are obtained. The study demonstrates the feasibility, maneuverability and higher precision assurance of decision tree classification in land cover classification using hyperspectral remote sensing imagery in the end of the paper.