摘要
正交子空间投影(OSP)方法广泛用于目标与背景的分离之中,对于高光谱影像,OSP可用于目标提取和混合像元分解,但缺点是需要端元的先验知识。针对这一问题,基于OSP的原理提出一种自动提取端元的方法,该方法不需端元的先验知识并且不需对原始数据进行降维。实验使用由OMISⅠ和PHI获取的两组高光谱数据进行端元提取,并采用带全约束条件的最小二乘法进行混合像元分解,结果精度令人满意,证明算法进行端元自动提取的可行性。
Orthogonal subspace projection(OSP) is employed widely to separate the targets and background, extract targets and mixed pixels unmixing for hyperspectral imageries. The disadvantage of OSP it needs full prior knowledge of endmembers. A method to extract endmembers from hyperspectral imageries automatically according to the principle of OSP is put forward. This method needn't prior knowledge as well as reduction of dimensions. The endmember extracting experiments is carried on by two sets of hyperspectral data, collected by OMIS I and PHI respectively. The results of the experiment on the process mixed pixel unmixing through the least squares with full constraints are satisfied. The experiments indicates that the arithmetic is feasible to automatic extract of endmembers from hyperspectra imageries.
出处
《测绘工程》
CSCD
2008年第6期37-40,共4页
Engineering of Surveying and Mapping
关键词
高光谱影像
正交子空间投影
最小二乘
像元分解
hyperspectral imagery
orthogonal subspace projection(OSP)
least squares
pixel unmixing