LAND COVER CLASSIFICATION IN BEIJING AREA BASED ON CBERS-02B CCD DATA

摘要

In this paper, we studied the potential of land cover classification using multi-spectral CBERS-02B CCD data. CEBERS-02B is the third of the series earth resource satellite of China. It was launched on Sept. 19th 2007. CCD is one of the payloads, and operates with a standard 5 bands setting through the visible to near infrared region of the electromagnetic spectrum. Compared with the main counterpart, Landsat TM/ETM+, the CBERS-02B CCD has the advantage of higher spatial resolution at 19.5m. This paper evaluated the ability of land cover classification with CBERS-02B CCD data through taking Beijing as the test area. CCD level2 data were used for this research. Before the classification, the data preprocessing was implemented to improve the geometrical accuracy of CCD. Classification training samples were selected from reference to vegetation atlas and land use map of China, high resolution satellite image and field collecting points. After analyzing the characteristics of samples of each land cover type, this research choose the decision tree classifier to accomplish this assignment. The decision rules were built with the spectrum bands of CCD and the ancillary variables. Finally the author focused on accuracy assessment using confusion matrix and kappa coefficient. Results form the assessment showed class accuracies ranged between 55.18 to 100percent, and the overall accuracy was 86.21percent.