Research Of Forestry Remote Sensing Information Extraction Based On Object-oriented Method In Central Shanxi

Forest plays an important role in people’s life. It can regulate the ecological balance, purify the air and so on so, the forest should be developed scientifically, be protected and monitored, to realize the use of forest reasonably and scientifically.The article chooses two typical areas (plain and mountainous area) as the research area, and the eCognition as the software platform.Using the object-oriented technology extracts the forest information of high and middle resolution image, to compare the different methods of classification. At last, this study formed methods and techniques of forest vegetation information extraction method based on object-oriented technologyThe main results of this study were as follows:1、 The parameters of the optimal segmentation scale of forestry information extraction are as follows, it discovers the parameters (50,0.1,0.5) can extract the forest of mountainous high -resolution IKONOS image, the parameters (50,0.1,0.5) can extract the forest of plain high-resolution IKONOS image, parameters(10,0.1,0.5) is fit to extract forest of mountainous middle resolution TM, parameters(2,0.1,0.5) is fit to extract forest of plain middle-resolution TM.2、The total accuracy of four kinds of classification of mountain IKONOS images are as follows, regulation classification based SEaTH algorithm and nearest neighbor (83.6%)> classification based on spectrum and texture by nearest neighbor (80.1%)> regulation classification based SEaTH algorithm (73.2%)> classification based on spectrum by nearest neighbor (72.2%). The total accuracy of regulation classification based SEaTH algorithm and nearest neighbor is the highest, the Kappa coefficient is 0.83, the result is best. And the precision of IKONOS is apparently higher than TM (67.3%)3、Comparing two kinds of nearest neighbor classification shows texture develop the accuracy, especially for coniferous forest, the misclassification error is from 36.9%to 19.8%,the omission error is from 26.9% to 18.0%. The accuracy of regulation classification based SEaTH algorithm and nearest neighbor is higher than the accuracy of classification based on spectrum and texture by nearest neighbor,and the accuracy of the broad-leaved forest has the largest increase. The misclassification error is from 21.4% to 12.9%,the omission error is from 29.0% to 13.0%,and the accuracy of other classes also improve.4、 The accuracy of plain IKONOS images are higher than the accuracy of TM. The overall precision of regulation classification is 85.3%,which is higher than nearest neighbor classification(72.3%).The accuracy of garden and forest have big development.their producer accuracy improve from40.9% and 44.5% to 80.5% and 87.2%, and the misclassification develop from 43.3% and 46.8% to 76.3% and 80.9%. In all, The plain IKONOS images is fit to using the regulation to extract forestry information, whose accuracy is higher than TM, The plain is fit to using regulation of high resolution.