The Research Of Bamboo Information Extraction Based On SVM Classification And Multi-Source Remote Sensing Data

With the advances and mature in remote sensing technology,more and more being used in environmental monitoring,planning,disaster assessment,military and other fields.In this paper,focuses on SVM classification theory and methods,first time applied such a classification to the information extraction research of bamboo;For the Shunchang County,Fujian Province as an example,applied a variety of classification methods,combined with multi-source remote sensing to extract the bamboo information of Shunchang County,and the extraction accuracy was evaluated and compared,the results showed that multi-source remote sensing-based SVM classification method achieved high accuracy.First of all,this paper adopt the TM data in 2007 and in May 2008 ALOS data,and do a large number of pre-processing work,including images correction,data integration and evaluation and so on. And do the classfication research based on the resolution merge data,the terrain plans,and resources distribution maps.Secondly,this paper using a variety of classification methods to extract bamboo information, including non-supervised classification,maximum likelihood classification,sub-pixel classification and the classification based on spectral characteristics,and a variety of classification accuracy results were evaluated and compared.The results showed that the total accuracy of SVM classification is 81.01%, the classification of bamboo has reached 82.52%,the indicators of accuracy are better than other classification methods in all aspects.Finally,this paper give the suggest to the course of this study for the problems and how to further enhance the accuracy of bamboo classification.In this paper,we done a lot of work for the SVM classification model selection,kernel function selection and parameter determination of C,taking into account the time efficiency and accuracy factors,we choose the LOO model,RBF kernel function and C parameters for the 13 cases for the bamboo to finalize the application of information extraction,classified research on bamboo.In this study,the SVM classification method is a more advanced classification methods,it is built on the experience of a theoretical risk minimization method which can overcome the traditional classification of "skills" over-reliance on the more appropriate classification of bamboo.This article holds that the remote sensing data for the pre-treatment affected the quality of classification accuracy is still one of the key factors.