典型高光谱图像端元提取算法在黄河口湿地应用评价研究

Evaluation of the prime hyperspectral endmember extraction algorithm in Yellow River Estuarine wetland

端元提取是混合像元分解的基础,也是高光谱遥感的研究热点。对于特定区域的高光谱图像应该使用哪种端元提取算法,需要对各种端元提取算法进行客观地评价。作者针对黄河口湿地CHRIS高光谱图像,使用了重建图像与原图像的均方根误差、有效端元数量两个指数对PPI、N-FINDR、VCA、OSP、IEA和SISAL六种典型的端元提取算法进行了评价。结果表明,SISAL算法重建误差最小,仅有其他算法误差的10%-28%;OSP算法识别了具有物理意义的6种有效端元,多于其他算法识别的地物类型,而SISAL算法识别的端元缺乏物理意义。

Endmember extraction is the foundation of mixed pixel decomposition and also the focus of hyperspectral remote sensing research.It is necessary to objectively evaluate all kinds of endmember extraction algorithms to determine which algorithm should be used in the hyperspectral image in a specific region. In this paper, we evaluated six kinds of endmember extraction algorithms（PPI, N-FINDR, VCA, OSP, IEA and SISAL） based on two indexes（the mean square root error between the reconstructed image and the original image, and the valid endmember number） for the CHRIS hyperpectral images of Yellow River Estuarine wetland. The results showed that the reconstruction error of SISAL algorithm is the minimal, which is only about 10%-28% of that of other algorithms. The OSP algorithm recognized six kinds of valid endmembers with physical meaning, which is more than other algorithms. In contract, the endmembers extracted by SISAL algorithm lacked of physical meanings.