一种改进迭代误差分析端元提取算法

摘要

Iterative error analysis (IEA) algorithm is one of the widely used endmember extraction algorithms.As for the compute-intensive shortcoming, this paper improves it from reducing the number of pixels participated in the iterative process.According to the convex geometry theory, mixed pixel locates on its internal subspace constituted by its endmembers.The projection value of these cells to their endmember orthogonal subspace ( OSP ) is zero, which can be removed during the iterative process for deriving the next endmember, thereby reducing the number of pixels in every iteration step.The experiments with simulated data and real hyperspectral data were done, proving that the improved algorithm has the same precision of the extracted endmember with original algorithm, and with the increasing of the number of extracted endmembers, the number of pixels participated in the iteration decreases grad-ually, which reduces the computation time than the endmember extraction of the original IEA algorithm.%迭代误差分析（ IEA）算法是应用比较广泛的端元提取算法之一，针对IEA端元提取算法计算量大的缺点，从减少参与迭代过程中的像元数目进行改进。根据凸面几何理论，混合像元位于其端元构成子空间内部，这部分像元到其端元正交子空间（OSP）投影值理论上为零，在迭代求下一个端元过程中，可以将这部分无用像元去除，从而减少每步迭代过程像元数目。采用模拟数据和真实高光谱数据进行实验，证明改进算法与原算法提取端元精度相同，随着端元提取个数的增多，参与迭代过程的像元数目逐次减少，比原始的IEA端元提取算法减少了计算时间。