基于形态学与正交子空间投影的端元提取方法

针对现有端元提取方法从数据光谱或空间或特征信息的单一方面出发进行混合像元分解、不同类型端元难以区分等不足,提出一种扩展形态学与正交子空间投影结合的端元自动提取方法.利用扩展膨胀和腐蚀操作,通过计算形态离心率指数进行高光谱数据的端元数据集计算;利用光谱角匹配方法提取不同类型的端元,通过向端元正交子空间投影消除已经提取端元的影响;并通过航空高光谱数据进行算法性能验证.实验结果表明:提出方法能够实现在无任何先验信息情况下图像端元的自动提取,并且能够有效地区分相似光谱端元.

Endmember extraction method based on morphology and orthogonal subspace projection

An automatic endmember extraction method based on the extended morphology and the orthogonal subspace projection was proposed to solve the problems, including identification of different endmembers with similar spectra and the endmembers extraction used by the spectral or spatial information only. Extended dilation and erosion operators were used to calculate the morphological eccentricity index (MEI), and the MEI was used to extract the endmember set. Spectral angle matching (SAM) was applied to distinguish different endmembers, and then the endmember was projected to its orthogonal subspace to eliminate the effect of endmembers that had been extracted. The airborne hyperspectral data was used to evaluate the performance of this method. The results illuminate that image endmembers are extracted automatically without any prior information and endmembers with similar spectra are distinguished effectively.