Land Cover Classification and Information Extraction of Multi-temporal High Resolution Images;

摘要

This paper presents a technique of land cover classification of high-spatial resolution image which consists of two Quick Bird data with obvious differences to implement classification and improve classification accuracy, one of the data is covered by snow. The study area locates at the sub-watershed of Kunming Dianchi Lake watershed. There is an obvious border between two images after data fusion, which cannot be cancelled by data processing. Hence, the classification of land cover by use of a same process set in Definiens developer 7.0 becomes difficult or causes errors, that is so called edge effect. Our research develops a practical process to extract the classification information from these special regions to form thematic map for environmental and socioeconomic applications. Six classes, including forest, cultivated land, architecture, water, bare soil and shadow are classified successfully. In the class thematic map, the border effect was cancelled completely. The methodology was developed in an experimental way by suitable selection of processes, parameters and methods. Overall accuracy of classification reaches 9232%, with Kappa statistics 0.8914.