基于拉格朗日的高光谱解混算法研究*

针对混合像元分解误差问题,提出一种基于拉格朗日算法的高光谱解混算法。通过变分增广拉格朗日算法提取出部分端元,由于端元组中存在相似端元影响解混精度,利用基于梯度的光谱信息散度算法进行光谱区分,除去相似端元。通过对得到的端元进行排序,依次增加端元进行光谱解混,将满足条件的端元增加进端元组,最终得到优选端元。该方法不仅有效去除了相似端元的干扰,而且不需要不断搜索端元的组合,根据每个端元对于混合像元的重要性做出相应次数的非限制性最小二乘法计算,得到更精确高光谱端元的子集,该方法对高光谱混合像元解混的效率以及可靠性均有所提高。

Hyperspectral unmixing algorithm based on Lagrangian

For mixed pixel decomposition error presents, an hyperspectral unmixing optimization algorithm based on lagrangian algorithm proposes. Through simplex identification via split augmented Lagrangian algorithm extracts endmembers.Because endmembers subset had similar endmembers and similar endmembers had an impact on the accuracy of spectral unmixing,spectral information divergence based on gradient algorithm is used for spectral discrimination to remove similar endmembers. By sorting the resulting endmember, followed by additional endmembers,endmembers meet the criteria will add into endmember groups and the resulting optimized endmembers will achieves. This method effectively removes interference of similar end, and no longer need to search combinations of endmembers.Each endmembers corresponding to the importance of the number of mixed will use in non-restricted least squares calculation,and more precise subset of hyperspectral endmember wil achieve. Efficiency and reliability of hyperspectral unmixing optimization algorithm will improve.