CAJ | 학술논문

以朝鲜半岛为研究区域,基于2000年和2007年的MODIS数据,应用线性光谱混合模型进行像元组分分解并提取分类特征,对像元组分分解后有错分的地物类型结合纹理进行分析,以决策树模型进行土地利用分类.结果表明,像元组分分解后的决策树分类结果总体精度达78.346 5%,Kappa系数达0.681 3.与像元组分分解前最大似然法和决策树分类结果相比,决策树分类精度优于最大似然法,且像元组分分解方法提高了分类精度.经2000年和2007年像元组分分解后的决策树分类结果对比:朝鲜耕地面积增加;韩国耕地面积减少,居民地面积增加.
이조선반도위연구구역,기우2000년화2007년적MODIS수거,응용선성광보혼합모형진행상원조분분해병제취분류특정,대상원조분분해후유착분적지물류형결합문리진행분석,이결책수모형진행토지이용분류.결과표명,상원조분분해후적결책수분류결과총체정도체78.346 5%,Kappa계수체0.681 3.여상원조분분해전최대사연법화결책수분류결과상비,결책수분류정도우우최대사연법,차상원조분분해방법제고료분류정도.경2000년화2007년상원조분분해후적결책수분류결과대비:조선경지면적증가;한국경지면적감소,거민지면적증가.
In order to do land use classification by decision tree method(DTC), the authors process pixel component unmixing and extract category features using linear spectral mixture model, and analyse the misclassified surface object type after pixel component unmixing which were combined with texture based on 2000 and 2007 MODIS data in Korean Peninsula. The analysis result shows the accuracy of DTC after pixel component unmixing is 78.346 5% and Kappa coefficient is 0.681 3. Compared with the classification results of maximum likelihood and DTC before pixel component unmixing, the classification accuracy of DTC is superior to maximum likelihood and the pixel component unmixing method improves the classification accuracy. According to the DTC classification results after pixel component unmixing of 2000 and 2007, the farmland area increased in northern Korean peninsula, while the farmland area decreased and the residential area increased in southern Korean peninsula.