论文部分内容阅读
为了提高高光谱遥感图像混合像元分解的精度,提出基于核方法的高光谱线性解混。采用核化正交子空间投影(orthogonal subspace projection,OSP)算子、最小二乘正交子空间投影(least squares OSP,LSOSP)算子、非负约束最小二乘(nonnegative constrained least-squares,NCLS)算子和全约束最小二乘(fully constrained least-squares,FCLS)算子等方法分别构建核正交子空间投影(Kernel OSP,KOSP)、核最小二乘正交子空间投影(Kernel LSOSP,KLSOSP)、核非负约束最小二乘(Kernel NCLS,KNCLS)和核全约束最小二乘(Kernel FCLS,KFCLS)高光谱图像混合像元解混模型。对CUPRITE矿区AVIRIS数据进行KLSOSP、KNCLS和KFCLS与LSOSP、NCLS和FCLS丰度反演对比实验,结果表明,对于混合像元广泛存在的高光谱遥感图像来说,基于核方法的KLSOSP,KNCLS和KFCLS的解混精度优于LSOSP,NCLS和FCLS;附加约束条件有利于提高丰度反演的精度。
In order to improve the resolution of hyperspectral remote sensing image mixed pixel decomposition, a hyperspectral linear unmixing based on kernel method is proposed. Orthogonal subspace projection (OSP) operator, least squares OSP (LSOSP) operator, nonnegative constrained least-squares (NCLS) ) Operators and fully constrained least-squares (FCLS) operators were used to construct Kernel OSP (KOSP), Kernel LSOSP (Total Least Squares Orthogonal Subspace Projection) KLSOSP, KNCLS and KFCLS hyperspectral image mixed pixel unmixing model. The contrast experiments of KLSOSP, KNCLS, KFCLS and LSOSP, NCLS and FCLS abundance inversion of AVIRIS data in CUPRITE mining area show that KLSOSP, KNCLS and KFCLS based on kernel method are effective for hyperspectral remote sensing images with mixed pixels. The accuracy of unmixing is better than that of LSOSP, NCLS and FCLS. Additional constraints are helpful to improve the accuracy of abundance inversion.