论文部分内容阅读
针对非线性混合的高光谱图像目标检测问题,在核信号空间正交投影法(KSSP)的基础上,提出了一种光谱和空间信息结合的组合核信号空间正交投影方法(CKSSP)。分别基于边缘序和像元距离为序尺度函数的导出序将灰度形态变换扩展到多值图像空间中的形态变换,利用多结构元素组合的扩展数学形态学方法提取高光谱图像的空间信息。根据核函数定义,结合光谱信息和空间信息构造出组合核函数并加以证明,通过组合核信号空间正交投影实现目标检测。该方法在充分利用光谱信息的同时,合理利用了空间信息。仿真数据实验结果表明CKSSP的均方根误差比KSSP小0.03,真实高光谱图像数据实验和ROC曲线均表明CKSSP目标检测结果优于KSSP。
Aiming at the problem of hyperspectral image target detection with nonlinear mixing, an orthogonal spectral space (CKSSP) method combining spectral and spatial information is proposed based on the orthogonality of nuclear signal space (KSSP). Based on the order of edge order and pixel distance, the morphological transformation of the gray morphological transformation is extended to the multi-valued image space based on the derivation order of the ordinal scale function. The spatial information of the hyperspectral image is extracted by using the extended mathematical morphology combining the multi-structure elements. According to the definition of kernel function, combined with spectral information and spatial information to construct a combined kernel function and to prove it, by combining the orthogonal projection of nuclear signal space to achieve target detection. This method makes full use of the spectral information and makes rational use of the spatial information. Experimental results of simulation data show that the root mean square error of CKSSP is smaller than that of KSSP by 0.03. Experiments with real hyperspectral image data and ROC curves show that the CKSSP target detection result is superior to KSSP.