提出一种利用图像的空间结构信息在特征空间中设计大小适宜的超球, 将单形体的顶点分隔在超球外部, 剔除超球内部的数据, 只保留超球外部的少量数据参与端元提取算法。经过分析, 该方法可以大大减少端元提取算法的运算量。通过实验对比, 用相关端元提取算法对简化后的数据进行端元提取的结果精度很高, 与简化前数据的端元提取结果吻合。

This paper proposes a new algorithm which designs a suitable hypersphere in the feature space by utilizing the spatial information of the hyperspectral image in the feature space to separate the simplex vertices outside the hypersphere, then the data inside the hypersphere is excluded and only a small amount of data outside the hypersphere is involved in endmember extraction algorithm. Our analysis indicates that this method can greatly reduce the amount of computation of the endmember extraction algorithm and therefore can improve the operational speed. The final experimental results illustrate that the endmembers extracted from the simplified hyperspectral data by certain algorithms based on convex geometry have high precision and almost are identical to the endmembers extracted from the hyperspectral data before simplification by utilizing same algorithms.