基于MODIS数据的辽宁省土地利用分类研究

采用最大似然法、马氏距离法、光谱角填图法、支持向量机法、神经网络法和最小距离法6种分类方法，对辽宁省2010年3—12月MODISNDVI数据，用该数据做主成分分析的前3个主成分数据、前5个主成分数据和2010年6—10月MODISNDVI数据等4类数据进行土地利用分类研究。结果表明：6种分类方法中最大似然法、马氏距离法和最小距离法3种方法较适合对MODISNDVI数据进行信息提取，其总体分类精度分别达82．63％、80．29％、79．17％，乔木林类型信息提取精度分别达81．91％、78．54％、80．02％；3种对原始数据进行变换的方法中6—10月数据效果较好，其总体分类精度最高达82．63％，乔木林信息提取的最高精度达78．54％。

Research on Land Use Classification in Liaoning Province Based on MODIS Data

The land-use classification in Liaoning Province was conducted with MODIS NDVI data between March and December in 2010 （by taking the top 3 principal components of MODIS NDVI data, and the top 5 prin- cipal components of MODIS NDVI data of this time duration individually） , and with MODIS NDVI data from June to October in 2010 by means of six kinds of classification methods, i. e. , the Maximum Likelihood Method, the Mahalanobis Distance Method, the Spectral Angle Mapping Method, the Support Vector Machines Method, the Neural Network Method and the Minimum Distance Method. The classification results showed that the Maximum Likelihood Method, the Mahalanobis Distance Method and the Minimum Distance Method were comparatively more suitable for MODIS NDVI data information extraction, whose overall classification accuracy was 82.63%, 80. 29% and 79.17% individually, and whose information extraction precision of arbor forests reached 81.91%, 78.54% and 80.02%, respectively. In terms of growing phases, the best results of vegetation data transformation by these three methods were obtained from June to October, whose overall classification accuracy reached 82. 63% , and whose information extraction precision of arbor forests could reach 78.54%.