结合邻域SAC-NED测度的高光谱图像分类方法

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

A spectral similarity measure SAC-NED was proposed in hyperspectral imagery classification by integra-ting 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 im-agery classification had better accuracy and stability.