基于RM S 误差分析的高光谱图像自动端元提取算法

提出了一种基于RM S ( root mean square) 误差分析的自动端元提取算法。对图像每做一次线性解混合, 就得到一幅以均方根RMS误差表示的残余误差图像, 从中选出误差较大的像素作为新的端元开始下一次解混合, 通过多次迭代, 直到得到了要求数目的端元。该算法克服了以往端元提取方法监督特性的局限, 减少了对先验信息的依赖, 同时保留了图像中的异常。利用仿真和实验数据验证了该算法的有效性。

Endmember Extraction Algorithm Based on RMS Error Analysis in Hyperspectral Imagery

The linear mixed mixture is a commonly accepted model for hyperspectral data processing. As the important parameter of the linear mixed mixture, endmember represents a certain ground component whose spectral is changeless comparatively. Over the past years, several techniques have been developed for accomplishing the determination of the endmembers in the literatures. Aimed at the disadvantages of these existing methods, this paper presents an automatic endmember extraction algorithm based on RMS error analysis. The new algorithm overcomes the insufficiency of other methods and needs less prior information. Not only the mean spectrum of the imagery is extracted, but also the anormaly is reserved. This algorithm also depends on the linear mixed mixture, it is executed directly on the data with no previous dimensionality reduction. The mean spectrum of the data is chosen to start the process, and then, a series of constrained unmixing operations is performed, each time selecting as endmembers the pixels that minimize the remaining RMS error in the unmixed image. This process is continued until the required number of endmembers is found or the unmixing error is below some threshold. The results of simulated and real hyperspectral data show the validity of the algorithm by a detailed comparative analysis with PPI.