A Study On Object-oriented Remote Sensing Image Classification In Land Use/Land Cover

the development of the remote sensing technology makes us obtain very abundant information of nature, the method on remote sensing image process improved these years, especially with the import of object-oriented technology makes remote sensing image classification breakthrough the single traditional classification method based on spectral of pixels. Otherwise, many use of remote sensing technology in culture classification is applied to the research of land use/land cover, and receive a lot success. In this paper, some key problems for extracting information of land use/land cover from TM in the south of the Gaoligong Mountain are discussed. Through several key steps, such as image pre-processing, terrain correction, image enhancement, feature extract, multi-segmentation, image classification and other effective method to get Land use/Land cover information in the mountainous area by using remote sensing data is successfully conducted in the Gaoligong Mountain. Then, analysis the 20 years change of land use/land cover based on the RS image in 1987, 1997&2006, offers the basic data for general environment research of this study area.The main content and conclusion of this research is as follows:(1) This paper mainly use the idea of object-oriented according multi-scale segmentation, calculate the spectral attribute of objects, analyzed the connections between objects, and choose K Nearest Neighbor method to classify the image, then appraise the accuracy based on field sample data.(2) The research on traditional classification method based on spectrum of pixels shows that it is sensitivity with noise. In high resolution image, the polygon is more fragmentation. The single pixel classification often brings on pepper and salt phenomenon. Also, the choice of training sample can make great influence to the result.(3) The results of accuracy analysis indicate that the method of object-oriented classification is better than traditional method based on spectral of pixels. the overall accuracy is 88.6%, increase 14.9% than the traditional method based on spectral of pixels, the Kappa coefficient is 0.82, increase 0.22.and the confusion between every culture reduced in evidence. So, it is feasible that bring in necessary feature extract knowledge in object-orient classify. it demonstrates that the object-oriented method can satisfy the demands of land use classification, and can be used as a new technique in land resource survey.(4) The results indicate that forest land is the main land type of the study area, and most of it is distribution in the nature reserve, the farm land is the main land type outside nature reserve. According to the land use/land cover changes analysis, the result indicates that the forest land in the nature reserve is protected better than outside. It is show that the existing protection measure of Gaoligong mountain nature reserve is efficient.The conclusion of this research indicates that remote sensing is suitable for land use/land cover investigation of mountainous nature reserve and its neighborhood. The data comes from research can be used to evaluate whether the protection measure was efficient or not, and the results come from research is the steadily foundation for making protection measure, resources development pattern of the nature reserve.