Land Cover Classification With SVM Applied To MODIS Imagery

Land cover classification is the basis of surveying, planning and dynamic monitoring land resource, and is the effective means of studying environment effect, ecosystem security and global changing, especially acquires land cover data of regional area exactly and real-time is a pressure mission in the global change study. Because MODIS imagery have characteristic of pantoscopic, acquired in-phase information easily, data is continuous, cheap, unlimited to area, so this paper takes Qiqihaer city as an example and applies MODIS imagery to regional land cover classification.Traditional classification methods are based on the principal of Experiential Risk Minimization, they can achieve the best result only when the number of samples approaches infinity. Unfortunately, the number of samples is actually limited and the data dimension is high. Taking into account the good generalization of support vector machine in small samples, nonlinearity and high dimension space, and according to multi-spectral and multi-temporal features of MODIS data, this dissertation deeply studies support vector machine and its application for regional land cover classification. The main works of this thesis are given as follows:This paper discusses flow and method of MODIS data land cover classification using SVM technology; analyzes spectral characters of typical land cover types in research area, according to analysis and then take features selection and extraction; analyzes what about effect on class accuracy when different features combination of index information and temporal information in experiment; at last compares SVM and traditional methods, analyzing experiment results, it contains train rate and test rate of all methods, and connection of accuracy, data dimension and samples number.This study demonstrated that classification accuracy can bring different effect with different combination of features, SVM method gains higher accuracy, compared with traditional methods it has some advantages, for example, study rate, adapt ability, unlimited to feature space dimension and generalization ability. So it demonstrated SVM has maneuverabi- lity and precision guarantee in the regional land cover classification.