Land Cover Classification Based On MODIS Data In Yunnan Province

Application of remote sensing technology for land cover classification is a basic research work in the field of remote sensing applications, particularly in accurate and real-time access to large-scale regional land cover data. It is an important task of the current global change research faced. Due to the large imaging area is accessible in synchronization information, and low cost, compared with AVHRR data, MODIS date is advantage of multispectral, multi-temporal, higher resolution and other advantages, these have become an important land cover classification data sources. It is necessary to use MODIS data to carry out the classification method of land cover.Taking the northwest of Yunnan as experimental areas, used the MODIS images in January, April, July and October of 2013 as data sources, extracted the feature classification and selected the relevant index information by the experimental evaluation of the classification accuracy and performance of the maximum likelihood, BP neural network, decision tree classifier and SVM classifier four, and ultimately determine the best method, and then use it for land cover classification of Yunnan Province and to obtain the final classification results. The main conclusions are as followings: 1. Based on the characteristics of multichannel MODIS data, Extracting classification characteristics from MODIS data, experimental studies have found that the optimal classification feature in classification experiment,the accuracy of 79.86% by the combination of I increased to 84.90% of the combination of II, increasing by nearly 5 percentage.2.The classification of the different characteristics of impact on the classification accuracy of each are not identical, joining some characteristics of classification may not be able to improve the accuracy of some categories, such as surface temperature during the day Tday, instead, the accuracy of construction land down 67.44 percent from 72.26 percent, so that it is important for the classification to select and extract the optimal classification feature.3.Shangri-la as experimental research area, respectively by maximum likelihood, BP neural network, decision tree classification and support vector machine(SVM) four methods to classify the experiment, and through the experiment to respectively analyze index information and phase information of the impact of different combinations of classification accuracy, but also from the performance analysis and evaluation of four kinds of classifier, since different combination of classification accuracy, the maximum likelihood method, BP neural network, decision tree classification and support vector machine(SVM) classification accuracy were 82.74%, 83.38%, 85.57% and 82.74% respectively, therefore, it is appropriately to use support vector machine(SVM) method to classify the whole Yunnan province, in order to get higher accuracy of classification of land cover types.4.The optimal classification method by support vector machine(SVM), use the year four phase when the surface reflectance data classification, at the same time also to join the optimal classification indexes is EVI, NDWI, Tday and NDBI, a total of 32 band, land cover classification in Yunnan province, using the data from the part two adjustable field sample data, part of the national geographic conditions survey the field survey data and through the classification of TM data interpretation chart data verifies the accuracy of the selected sample points and the resulting confusion matrix overall classification accuracy and kappa coefficient were 70.78% and 0.6264 respectively, for classification in terms of quality, a good one for the column, therefore, this article choose the method of support vector machine(SVM) classification experiments, in Yunnan province is feasible, can be promoted.5. Six land use land area of the province of Southwest Yunnan, North Southeastern,Yunnan region, eastern Yunnan region, western region, western North, cover types are analyzed and discussed, and through land-use problems in specific regions propose appropriate the use of the direction and measures for land use survey in Yunnan and updated survey, real-time dynamic monitoring, land resources planning and management, and promote the use of MODIS data to provide technical methods and regional case.