期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2006
卷号：XXXVI-4/C42
出版社：Copernicus Publications
摘要：The National Land and Water Information Service (NLWIS), an Agriculture and Agri-Food Canada (AAFC) initiative, is tasked with mapping 'circa 2000' land cover for agricultural areas of Canada with an overall accuracy of greater than 85%. The method for mapping land cover is being developed using a Decision Tree (DT) classifier (see5 software) with specific focus on the identification of 3 agricultural classes: grassland, annual and perennial crops. Class separation is based on rule sets automatically derived from ground information. Using an eCognition segmentation on the Landsat data, a per-object majority filter is performed on the output classification. A unique mosaic procedure has been developed using segmented objects and accuracy results from See5. Preliminary results show the AAFC methodology outperforms both the standard Maximum Likelihood Classifier (MLC) and the object based eCognition method. Implementing multi-date imagery to separate cropland from hay and pasture improves the accuracy of separating other classes, therefore leading to an overall improvement of the classification accuracy. The effect of ancillary data on the classification is dependent upon landscape characteristics. The DT approach, when combined with image segmentation, has proven to meet or overcome our accuracy requirements with minimum analyst intervention
关键词：Agriculture; GIS; Land Cover; Mapping; Classification; Landsat; Segmentation; National
