Study On Information Extraction Of Main Forest Species In Mountain Land In South Of China Based On Remote Sensing

In this study,we made shunchang county as research area, through constructing classification features such as spectral features, topographic features,vegetation indices and texture features,we have made. a full analysis of the main differences of these classification features between differenct tree species. Selected high spatial resolution remote sensing image ALOS,imployed decision tree classifier and support vector machine classifier to classify trees species in the reseach area.As a result,study out a better method of classification of tree species which exist in complex topographic condition.(1)By analyzing the differences of hyperspectral datas, information in ALOS images, topographic factors, vegetation indices and texture features of various tree species, we concluded that in a complex topographic ondition,relying on a single factor,such as spectral features,vegetation indices,or texture features, can not reach a high-accuracy classification result.Only when we effectively imploy these features, we can have a ideal classificatin result.(2)Topographic factors in complex topographic conditions,plays an important role in the tree species classification.The same tree species in different slope aspects may have very different spectral features,and in the same slope aspect,different trees species have small difference in spectral features.In sunny slope,every tree species have large differences to every tree species in the shady slope.Spectral features of baboo in the gentle slope is different to.spectral features of every tree species in the slanting slopes.(3) The decision tree classification result based on spectral information, vegetation indices and texture features is higher 5.39% in overall accuracy than the decision tree classification based on spectral information, vegetation indices;and is 0.0641 in Kappa coefficient;texture feature of Variance,contrast,homogeinity,and entropy can make Baboo in the research areas improved 5.8% in overall accuraty;texture feature of variance,contrast,homogeinity,dissimilarity and entropy can make Masson pine in the research areas improved 12.11% in overall accuraty;texture feature of variance,contrast and entropy can make economic forest in the research areas improved 10.89% in overall accuraty;texture feature of Variance and entropy can make non-vegetation in the research areas improved 7.39% in overall accuraty. (4) The decision tree classification result based on spectral information, vegetation indices , texture features and topographic features is higher 10.35 % in overall accuracy than that based on spectral information, vegetation indices and texture features;Kappa coefficient is higher 0.1239;Topographic features of elevation,slope and aspect can make overall accuracy of baboo in the research area improved 3.23%,can make overall accuracy of chinese fir improved 22.76%, make overall accuracy of masson pine in the research areas improved 14.85%,make overall accuracy of broadleaf forest in the research areas improved 20.69%and can make overall accuracy of economic forest in the research areas improved 8.17%.(5) By comparing the decision tree classification result and the spport vector machine classification result,we find out that the decision tree classification result based on spectral features, vegetation indices,texture features and topographic features is the highest in overall accuracy and is Kappa coefficient,respectively are 81.90% and 0.7772,as for such Shunchang County,smountain land condition,decision tree classification method based on the spectral features, vegetation indices, texture features and topographic features,is the best method of tree species classification, the classification result is the best.