简化模糊H-ARTMAP神经网络在高光谱遥感分类中的应用  

Application of simplified fuzzy H-ARTMAP network to hyperspectral remote sensing classification

作　　者：张伟[1] 杜培军[1] 尹作霞[1] 

机构地区：[1]中国矿业大学地理信息与遥感科学系,江苏徐州221008

出　　处：《测绘科学》2008年第5期113-115,共3页Science of Surveying and Mapping

基　　金：国家自然科学基金资助项目(40401038);地理空间信息工程国家测绘局重点实验室开放基金资助项目

摘　　要：人工神经网络作为一种不需估计类别分布参数的遥感影像分类方法,能够克服分类中的不确定性,提高分类精度。模糊ARTMAP人工神经网络具有稳定、泛化性能好、支持增量式学习等特点,通过对简化模糊AR-TMAP神经网络和H-ARTMAP神经网络的分析和集成,构造了一种用于高光谱遥感影像分类的简化模糊H-ART-MAP网络。实验证明该方法在分类效率、运算时间和分类精度等方面都优于最大似然分类、BP神经网络、最小距离分类、光谱角制图模型等分类方法。Artificial Neural Networks have been widely used in remote sensing classification and gained higher accuracy than traditional statistical classifiers since they don＇t need to estimate probability distribution and parameters. Simplified Fuzzy H - ARTMAP Network, that combines the advantages of Fuzzy ARTMAP and H - ARTMAP Network, is used to construct a classifier for hyperspectral remote sensing imagery. Experiments demonstrate that Simplified Fuzzy H -ARTMAP Network outperforms those traditional methods, e. g. MLC, BP Network, Minimum Distance, Mahalanobis Distance, Spectral Angle Mapping, Binary Encoding classification, and it has higher efficiency than BPNN due to less processing time and computation burden.

关 键 词：简化模糊ARTMAP神经网络 H—ARTMAP神经网络 超球体 高光谱遥感 分类 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]