多波段遥感数据的自组织神经网络降维分类研究  

Dimension Reduction of Self-organized Neural Network Classification for Multi-band Satellite Data

作　　者：哈斯巴干[1] 马建文[1] 李启青[1] 戴芹[1] 

机构地区：[1]中国科学院遥感应用研究所

出　　处：《武汉大学学报（信息科学版）》2004年第5期461-465,共5页Geomatics and Information Science of Wuhan University

基　　金：中国科学院知识创新工程基金资助项目 (KZCSX2 3 12 );国家 863计划资助项目 ( 2 0 0 2AA13 3 0 3 0 )

摘　　要：介绍了基于聚类分析的自组织特征映射神经网络分类方法 ,神经网络的输出层结构选用了 3D结构 ,可以更好地保持多波段遥感数据中的内在拓扑结构 ;并选择天津大港地区的ASTER数据中的 9个波段作为试验数据 ,通过对验证点的统计 ,分类精度达到了 94To meet the productive application of satellite data in land cove classification the higher resolution, the more bands are used, the more accurate results can be produced. A clustering in 3D self-organized neural nodes is used. ASTER data is a new kinds of sensors, 3 bands with 15m resolution and 6 bands with 30m resolutions. Dagang region in Tianjing is selected as a case study area. The Wavelet fusion is applied to fused different bands with different resolutions, then the self-organized neural network classification for land use is performed. Finally classification results is compared with that of the maximum likelihood classification with the same training samples. The accuracy of validation result is over 94%.

关 键 词：分类 小波融合 自组织特征映射 神经网络 主成分变换 土地利用 遥感数据 

分 类 号：S127[农业科学—农业基础科学] TP183[自动化与计算机技术—控制理论与控制工程]