基于数据场的端元提取算法研究

Research on Endmember Extraction Algorithm based on Data Field

传统端元提取算法一般需要人工指定端元数目,易导致多选或漏选端元。利用数据场自然拓扑聚类、可视化的特性,提出了基于数据场的端元提取方法。首先对图像进行分区处理,然后应用数据场思想计算各区域数据点的势能,并分别选择一定数量的特征点,将所有特征点集合成特征图像,再计算特征图像的数据场;最后根据数据场形成的拓扑聚类结构,可视化地提取端元,获得最佳端元的数目和位置。利用Cuprite矿区的AVIRIS数据进行端元提取实验,结果表明:该方法是合理有效的,能够应用于高光谱图像的端元提取中。

Endmember＇s number in traditional endmember extraction algorithms always need to be specified manually,which will result in multiple or miss chosen of endmembers. To solve this problem, this paper propose a new endmember extraction algorithm based on some characteristics of data field,such as natural topological clustering and visualization. Firstly, partial original images into several zones calculate potential of all pixels in each zone based on the theory of data field. Then select a certain number of feature points in each zone and combine all the feature points into a feature image, and then calculate data field of the feature image. Finally,according to the structure of topological clustering in data field, endmembers can be extrac- ted visually. At the same time, the best endmembers position can be gained. Using the data of AVIRIS ob- tain in the Cuprite mining area,endmembers of high quality are extracted. The results show that the meth- od is reasonable and effective. It demonstrates that the proposed algorithm can be applied to endmember ex- traction of hyperspectral images.