摘要
由于超光谱图像数据量大,维数高给分类识别处理带来不便,该文提出一种可行有效的波段约简方法.通过FCM聚类将原始波段划分为若干等价波段组,然后根据最大隶属度原则只保留每组中具有代表性的波段,达到维数减小的目的。其中,模糊聚类中相似度的定义是基于超谱相邻波段间的相关性,利用粗糙集理论中的处理属性依赖性的方法合理表达出来。实验表明,这一方法既有效地缩减了高维数据,又尽可能少地损失有用信息,保持了原始波段的分类能力。
A method of hyperspectral band reduction based on Rough Sets (RS) and Fuzzy C-Means (FCM) clustering is proposed, which consists of the following two steps. First, Fuzzy C-Means clustering algorithm is used to classify the original bands into equivalent band groups, which employs the concept of attribute dependency defined in RS to define the distance between a group and the cluster center, viz. the correlatives of adjacent bands. Then the data is reduced by selecting the only one from each group with maximum grade of fuzzy membership. With this approach, great dimension of band is decreased while preserving much wanted information. Simulation results prove the effectiveness of this approach.
出处
《电子与信息学报》
EI
CSCD
北大核心
2004年第4期619-624,共6页
Journal of Electronics & Information Technology
基金
国家自然科学基金(69904004)
教育部跨世纪优秀人才培养计划
高等学校骨干教师资助计划资助课题
关键词
粗糙集
模糊聚类
高维数据
超谱波段约简
超光谱遥感
Hyperspectral remote sensing, Rough Sets(RS), Fuzzy C-Means(FCM) clustering, Band reduction
