分区控制和分层提取下的OMIS-I影像土地利用/覆盖分类方法研究

针对宜兴试验区128波段OMIS—Ⅰ机载成像光谱数据，在平原区、山区分区基础上建立多级分层决策树综合分类方法。通过该方法可实现对土地利用／覆盖一级类（林地、园地、水域、耕地）自动分类，精度达88．15％，Kappa指数为0．82，园地漏分率最高（20．23％），耕地漏分率最低（2．24％）；耕地、林地、园地、水域的错分率分别为21．87％、21．98％、0．66％、0。土地利用／覆盖二级分类也可采用分层自动分类，其总体精度达86．16％，Kappa指数为0．81。采用土地利用矢量数据辅助和人工目视解译对部分二级类别进行细分，可获得更加可靠的分类结果。研究表明，分区控制下的多级分层决策树综合分类方法应用于OMIS—Ⅰ高光谱影像土地利用／覆盖分类，具有可操作性和较高的精度。

Classification of Land Use/Cover with OMIS-I Images Based on Segmentation Control and Multi-layer Information Extraction

Based on OMIS-I hyperspectral remote sensing data of 128 bands for Yixing area,a decision tree has been built for both plain and mountainous area.With this method,the first level in land use/cover(forest land,garden plots,water area,cultivated land)can be classified automatically with an accuracy of 88.15% and the Kappa Coefficient is 0.82.20.23% of garden plots and 2.24% of cultivated land are left out in the classification respectively.Error in classification for cultivated land is 21.87%,forest land 21.98%,garden plots 0.66%,water area 0.Classification for land use/cover types in the second level can also be carried out automatically with an accuracy of 86.16% and the Kappa Coefficient is 0.81.Better results can be achieved by interpretation with the aid of vector map of land use.Study in the area shows that the integrated decision tree method based on segmentation control and multi-layer information extraction is operational and results in a good accuracy in classification of land use/cover with OMIS-I hyperspectral data.