Land Use/Cover Classification Of North Korea Based On The Decision Tree And Mixed

Land cover is used to describe the nature of the earth’s surface, and refers to thebiological or physical coverage types on the Earth’s surface. Land cover provides importantparameters for the global change research, the land use change monitoring, the sustainabledevelopment evaluation and climate model, global and regional carbon cycle and the model ofecological system prediction on the study. Remote sensing technology can catch macro,dynamic, more spectrum section and multi-scale situation of the earth’s surface condition andits dynamic change, and it saves a lot of manpower and material resources. Remote sensinghas become a main way to obtain surface cover information. The study area North Korea is atypical mountain nation which forest vegetation covers more than80%, and the complexterrain factors makes information extraction of North Korea land use/cover becomes adifficult point. As China’s an important neighbor, the ecological environment change of NorthKorea on our border has an important influence on ecological security. This paper based onthe above factors, did classifying research of the North Korea’s land use/coverage status byLandsat-TM images. A typical area which had obvious influence by human activities waschosen and linear spectral mixture model technology was used for quantitative research. Theclassification technologies for different scales were discussed in this paper, and reliable dataguarantee for NorthKorea’s comprehensive system land use/cover analysis、ecological safetyevaluation and food safety issues were provided. The main research results included:1.In the North Korea all land cover classification, the study area was divided into12types by decision tree classification method. Its overall accurac y reached81.5%, showedgood classification accuracy.2.The decision tree classification were compared with maximum likelihood classification,the overall accuracy of maximum likelihood classification was72.7%. The results existedmany wrong or missing classified phenomenon. It declares that in large land coverclassification, the decision tree classification has incomparable advantage over traditionalclassification method.3.In land cover classification of the buffer zone along the YaLu River’s both sides,decomposed mixed-pixels through extracting the five endmember types in the image. Thedecomposition classification accuracy reached90.1%, kappa coefficient was0.8501. And theoverall accuracy of Spectral Angle Mapper classification was only83.7%, it visiblliy couldsee that the accuracy of pxiel-decomposition classification was better than SAM classification.In small range of land classification, using linear spectral mixture model to unmixingmixed-pixel was an effective measure to improve the land use/cover classification accuracy.