摘要
探讨了一种利用多项Logit模型分层提取土地覆盖专题信息的方法。考虑客观存在的异物同谱现象,构建分层分类体系,针对不同层的地物类别选取不同的预测变量构建多项Logit模型,分步骤地提取各地物类专题信息。将此方法应用于美国蒙大拿州中部地区的土地覆盖专题信息提取,结果表明,该方法较常规的使用同一组特征变量构建单一模型一次性地划分所有地物类的方法在总体分类精度上有了明显改善。
Since there exists the objective phenomenon that different classes sharing the same spectral characteristics, a hierarchical categorical mapping approach is developed using generalized linear models. According to the similarities of spectral characteristics, different classes of high similar spectral characteristics are merged into the same one. Then a hierarchical modeling scheme is formed. Different predictor variables are chosen to build different multi- nomial Logit models to extract classes in different layers. Mask technique is employed to ex- tract thematic maps step by step. Ultimately, all thematic maps are incorporated into a whole one. Experimental results show that the hierarchical modeling using generalized linear models is an effective approach to improve land cover mapping quality.
出处
《武汉大学学报（信息科学版）》
EI
CSCD
北大核心
2008年第11期1166-1169,共4页
Geomatics and Information Science of Wuhan University
基金
国家973计划资助项目(2007CB714402)
关键词
多项Logit模型
逐步回归
分层分类
multinomial Logit models
stepwise regression
hierarchical classification
