Cart-based Land Use/cover Classification of Remote Sensing Images
基于分类回归树分析的遥感影像土地利用／覆被分类研究

Nowadays,investigations on land use / land cover change detections constitute a main objective for the global research.As a part of rapid development in technology,remote sensing has become an important tool to acquire the information of the land use/cover.Therefore,how best the extraction of timely and accurate information from these remotely sensed images is an impending problem.Recently,the knowledge-based interpretation of these images has become an effective and efficient approach to realize the automatic interpretation,which can integrate the spectral and other associated information based on experts' knowledge and experience to improve the accuracy.However,it is a bottleneck problem to obtain the knowledge for its wide application.A case study on the land use/cover classification of Jiangning study area in Jiangsu Province is discussed in the present article.At first,the data are preprocessed,then the relevant sixteen variables including geographical coordinate,grey value of four bands,textural statistics,DEM,slope and aspect are selected and extracted.The defined training sample areas are picked up by stratified random sampling techniques based on geographical coordinates.Thirdly,classification rules are discovered from these samples through Classification and Regression Tree(CART) Analysis,which integrates spectral,textural and the spatial distribution characters.Fourthly,the interpretation was performed by a judgment based on these rules.Finally,the traditional supervised as well as logic channel classifications are also performed to check the classification accuracies.The results have suggested that the accuracy of classification based on the CART is higher than others',which can obtain a lot of reasonable rules most quickly and effectively.So,it was felt that it is a good way to promote the wide application of knowledge-based interpretation of remote sensing images.