Research of RGB bands Quick Bird image land cover classification of a sub-watershed in Kunming Dianchi Lake Basin

Land cover classification based on object-oriented method needs all bands including RGB+Pan+Nir[1–3]. But at present, most of processed data(fused data) only include RGB bands in many government departments and organizations. It will be difficult for these data to be used as basic imagery to extract land cover information. For example, NDVI can't be calculated without Nir band. Further more, it will be heavy workload to process original data again in order to obtain Pan+Nir bands in processed final data. This paper developed a method to extract land cover classification information from processed RGB Quick Bird imagery. The methodology was developed in an experimental way. The method only depends on R-G-B bands to extract the land cover classification information. Six classes in 360 square kilometers sub-watershed, including forest, cultivated land, water, building, bare soil and shadow are extracted successfully by using the commercial software Definiens professional. Totally classification accuracy of the sub-watershed reaches 92.32%, and Kappa statistics reaches 0.8914.