结合邻域SAC-NED测度的高光谱图像分类方法  

A Hyperspectral Imagery Classification Method Using SAC-NED Measure with Neighborhood Information

作　　者：张斌[1,2] 朱述龙[1] 孟伟灿[1] 曹彬才[1] 宋新阳 

机构地区：[1]信息工程大学 [2]96633部队 [3]61243部队

出　　处：《测绘科学技术学报》2015年第6期600-604,共5页Journal of Geomatics Science and Technology

摘　　要：提出了一种光谱相似性测度用于高光谱图像分类方法。通过将光谱向量进行归一化处理,将计算得到的欧氏距离与光谱角余弦的值域归化到相同区间,得到光谱角余弦与欧氏距离联合测度值(SAC-NED)。在对图像像元进行分类时,以距离加权的方式将邻域像元参与中心像元SAC-NED值的计算,将像元分到SAC-NED值最大的类别。通过与其他5种常用相似性测度方法的实验结果对比表明:该算法能够提升高光谱图像分类的准确性和稳定性。A spectral similarity measure SAC-NED was proposed in hyperspectral imagery classification by integrating the normalized Euclidean distance（ NED） and the spectral angle cosine（ SAC） into the same range with the normalized spectral vector. By using neighborhood＇ s SAC-NED value with its weight, the pixel is classified to its corresponding class with the maximum SAC-NED value. The performance of the proposed measure was compared with five traditional spectral similarity measure approaches. The result showed that the improved algorithm for imagery classification had better accuracy and stability.

关 键 词：邻域信息 欧氏距离 光谱角余弦 高光谱图像 图像分类 

分 类 号：P237[天文地球—摄影测量与遥感]