基于最大化N维立体光谱角的高光谱端元提取

Hyperspectral Endmember Extraction Based on Maximum N-dimensional Solid Spectral Angle

提出了一种基于最大化N维立体光谱角(Maximum N-dimensional Solid Spectral Angle,MNSSA)的端元提取方法.该方法通过计算N个光谱向量在高维欧几里得空间的光谱夹角,定量衡量该N个光谱向量的独立性.在线性混合模型假设下,端元光谱向量的欧几里得空间夹角大于混合像素构成的夹角.MNSSA法不受待提取端元数目及波段数目的限制,对光谱向量幅值变化不敏感,能够克服阴影及光照因素对端元幅值的影响.使用模拟数据及AVIRIS(Airborne Visible/Infrared Imaging Spectrometer)获取的真实高光谱数据对MNSSA端元提取法及现有基于几何的端元提取法进行了对比评价.仿真结果表明,MNSSA法能够克服阴影影响因子对端元幅值的影响,端元提取准确率优于现有端元提取法,且具有良好的抗噪声性能,能显著降低高光谱数据的重构误差.

An endmember extraction method was proposed based on the maximum N dimensional solid spectral angle theory, which was termed as MNSSA （Maximum N-dimensional Solid Spectral Angle） in this paper. By using the ability of calculating the solid spectral angle which constructed by N spectral vectors in the high-dimensional Euclid space for the method, the independences of the N spectral vectors were measured. In the assumption of linear unmixing model, the solid spectral angle constitutes the maximum value for N endmember spectral vectors. The MNSSA method is not restricted by the numbers of endmembers or the numbers of bands. More importantly, the MNSSA method is not sensitive with the amplitude variations as well as robustness with effects from shadow strength and illumination intensity. Experiment results on synthetic and real hyperspectral data collected by AVIRIS （Airborne Visible/ Infrared Imaging Spectrometer） indicate that the method of MNSSA is better than the current mainstream used endmember extraction methods. The influences on the amplitude of the endmembers caused by the shadow factor are overcomed by the method of MNSSA. The method is with good anti- noise performance. Furthermore, the reconstruction errors for real hyperspectral data were reduced remarkably.