摘要
为了能够在不需要事先知道地物类别数目的情况下进行光谱的有效聚类,采用均值漂移算法完成聚类这一步骤。根据同一种地物光谱存在变异性的情况,使用了光谱角距离作为均值漂移聚类算法的相似性准则。用模拟高光谱数据和真实高光谱数据进行实验,结果表明基于改进后的均值漂移聚类的端元束提取方法较传统的K均值算法更能有效地形成代表每种地物类别的端元束。
In order to perform effective clustering of the spectral without knowing the number of features in advance,the mean shift algorithm is used to complete the clustering step.According to the variability of the spectral of the same feature,the spectral angular distance is used as the similarity criterion of the mean shift clustering algorithm.Experiments with simulated hyperspectral data and real hyperspectral data show that the endmember bundle extraction method based on improved mean shift clustering is more effective than the traditional K-means algorithm in forming endmember bundle representing each feature category.
作者
陈立伟
邱艳芳
朱海峰
王立国
CHEN Liwei;QIU Yanfang;ZHU Haifeng;WANG Liguo(College of Information and Communication Engineering,Harbin Engineering Universitity,Harbin 150001,China)
出处
《应用科技》
CAS
2020年第1期21-30,共10页
Applied Science and Technology
基金
国家自然科学基金项目(61675051).
关键词
多端元光谱混合分析
端元束提取
聚类算法
均值漂移
K均值聚类
光谱变异
光谱角距离
类分离
multi-endmember spectral mixture analysis
endmember bundle extraction
clustering algorithm
mean shift
K-means clustering
spectral variation
spectral angle distance
class separation