论文部分内容阅读
遵循美国国家土地覆盖数据库2001分类主题及系统(30 m空间分辨率),研究中等分辨率成像光谱辐射仪MERIS(300 m)土地覆盖产品的发展及评价。4种监督分类器包括马氏距离、最大似然、决策树以及支持向量机被用来发展区域土地覆盖信息。结果表明:(1)支持向量机在土地特征刻画过程中分类性能最优;(2)由支持向量机导出的MERIS土地覆盖产品尽管其识别地面细节的能力不及NLCD2001,但其主要地物类型在空间分布上与NLCD2001比较接近。分析还进一步揭示MERIS数据可成功地区划水体、常绿森林、裸地及栽培作物等地物类型,而对于落叶林及灌木林的刻画则性能相对较差。在MERIS土地覆盖产品中观察到从灌木林向裸地、灌木林向常绿森林及灌木林向草地的误分现象。然而,MERIS土地覆盖产品的生产较NLCD2001要节省人力及成本,中等尺度的MERIS土地覆盖产品对于某些科学应用将具有独特的价值。MERIS土地覆盖产品的发展应该充分应用多种辅助信息以及区域调制的分类策略,以期获得更加可靠的分类结果。
Following the National Classification of Lands Database 2001 classification theme and system (30 m spatial resolution), the development and evaluation of the MERIS (300 m) land cover product for a medium resolution imaging spectroradiometer was studied. Four kinds of supervised classifiers including Mahalanobis distance, maximum likelihood, decision tree and support vector machine are used to develop regional land cover information. The results show that: (1) SVM has the best classification performance in the characterization of land features; (2) MERIS land cover products derived from SVM are less capable of recognizing ground detail than NLCD2001. However, Spatial distribution and NLCD2001 closer. The analysis further reveals that MERIS data can successfully distinguish the types of water bodies, evergreen forests, bare land and cultivated crops while the performance of deciduous forests and shrubbergs is relatively poor. The phenomenon of misclassification from shrubbery to bare land, shrubbery to evergreen and shrubbery to grassland was observed in MERIS land cover products. However, the production of MERIS land cover products saves labor and costs over NLCD 2001, and the mid-range MERIS land cover product will have unique value for some scientific applications. The development of MERIS land cover products should make full use of a variety of auxiliary information and classification strategies of regional modulation in order to obtain more reliable classification results.