中文摘要：

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曰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作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。

英文摘要：

For mixed pixel decomposition error problem , this paper proposed an hyperspectral unmixing optimization algorithmbased on Lagrangian algo rithm . Through simplex identification via split augmented Lagrangian algorithmed , it extracted endmembers.Because endmem bers subset had similar endmem bers and simila r endmem bers had an im pa ct on the accuracy ofspectral u mixing , it used spectral info rm atio n divergence based on gradient algo rithm for spectral discrimination to removes im ila r endmem bers. By sorting the resulting endmember, follow e d by addition alendm embers, endmembers met the crite riaw ould add in to endmember groups and the resulting optim ized endmem bers w ould achieves. T h is m ethod effectively removesinterference o f s im ila r en dm em ber, and no longer needs to search com binations o f endmem bers. Each endmembers corresponding to the im portance o f the num ber o f m ixed w ill use in non-re stricted least squares c a lc u la tio n , and more precise subset o f h yperspectralendm em ber w ill achieve. Efficiency and reliability o f hyperspectral u n m ixin g op tim ization algo rithm w ill im prove.