Study On Extracting Land-use Information Based On QuickBird Image

The study selected 4km~2 area of GuCheng town in Pixian city as the research area, based on QuickBird image, and extracted the contrast texture in different windows, involved in SVM classification with multi-spectral data supporting by the remote sensing software of ENVI. Based on the separability, the study fined the best texture window of different object, using the different windows for the classification of multi-window image texture. By the support of Definiens software, the study extracted the land-use information, using Object-Oriented image classification, injecting texture into image multi-scale segmentation, building the fuzzy rule, using fuzzy classification and the nearest neighbor classification, compared the result with different ways. Based on the classification image, calculated to classification matrix of confuse, accuracy assessment parameter, and according to classification accuracy assessment parameter carry out accuracy analysis, accuracy assessment. The study provided a rapid extraction of technical support for the second national land survey, land-use dynamic monitoring and updating, remote sensing of crop yield estimation of the accuracy of thematic information. The study results are as follows:The study calculated the best separability window of different objects through J-M distance. The best separability of arable land is 3×3 window, water is 5×5 window, garden plot and woodland is 13×13 window; bare arable land is 15×15 window, residential area is 17×17 window.The image information extracted that the accuracy by spectrum information combination with different window texture higher than only using spectrum, and the accuracy of 11×11 window texture is the highest.The accuracy of the method that three textures windows participate in classification is 79.42%, and the Kappa is 0.7370, higher than two windows texture participates in. Compared with the classification results of single texture participate in, the classification effect had been obviously improved. Multiwindow texture can describe the texture feature object preferable, and can effective solve the problem of object distributing.Object-oriented classification method can make flexible using features, improve the accuracy of classification. The accuracy of method Object-Oriented is 87.22%, and the Kappa is 0.8363.Compared with pixel-oriented classification, OA increased 7.80%, and the Kappa increased 0.0993. It reduced the spectral characteristics of the impact of a stronger that texture information participating in multi-resolution segmentation, what makes a more reasonable segmentation results. The study comprehensive used fuzzy classifier and the nearest neighbor classifier, giving full play to their advantage that make classification results are more reasonable.