摘要
光谱角填图算法作为一种常用的光谱相似性度量方法,不能正确区分光谱间的正负相关是其内在的缺陷,因此,导致基于光谱角的高光谱遥感图像分类算法误分率较高。详细分析了光谱角的数学公式,提出了逐波段修正负相关影响的光谱角填图算法。算法首先判断每个波段间是否存在负相关,然后对负相关产生的光谱角加以修正,使最终的全波段光谱角可以正确反映光谱间的相似关系。算法由IDL7.0实现,在模拟数据上的实验表明,修正了负相关的光谱角可以将光谱角填图算法无法区分的光谱正确分离出来;在实际高光谱遥感图像上进行目标探测的实验表明,修正后的光谱角可以提升探测效果,有效地压制误分率。
As the most common distance metrics, the spectral angle mapper （SAM） algorithm can not distinguish between positive and negative correlations correctly so that high misclassification rate occurred in hyperspectral image classification algorithms based on spectral angle. A SAM modified negative correlation （MNC-SAM） band by band is the formula of spectral angle, which determines proposed in this paper through exact decomposition and modifies spectral angles affected by negative correlations to reflect similarity between spectra correctly using all bands spectral angles. The algorithm is implemented in IDL7.0 and the results from experiments on simulation data show that the spectral angle modified negative correlations can discriminate spectra correctly which can not be classified by SAM while the targets detection results on real hyperspectral remote sensing image prove that the modified spectral angle can improve detection effect and suppress misclassification rate significantly.
出处
《火力与指挥控制》
CSCD
北大核心
2013年第2期22-25,30,共5页
Fire Control & Command Control
基金
国家自然科学基金资助项目(60802084)
关键词
高光谱图像
相似性度量
光谱角填图
负相关
二分法
hyperspectral image, distance metrics, SAM, negative correlation, bisection method
