高光谱影像端元提取算法的进展分析与比较

对当前国际经典和前沿的6种代表性的端元提取算法进行比较研究,包括SPP-N-FINDR、VCA、SPICE、PCOMMEND、MVSA和MVC-NMF,通过理论和实验两种方式对这些算法进行综合性对比和分析,总结其优势和存在的问题。通过模拟和真实数据实验得出:SPP-N-FINDR算法的抗噪声能力不如其他5种算法;VCA和MVSA的稳定性较好;MVC-NMF和SPICE无需知道端元数目,且能直接得出丰度矩阵,自动化程度较高;PCOMMEND在真实高光谱图像中提取端元的结果最好,能直接得出丰度矩阵,但若端元数量为素数时精度会下降。研究成果将为今后围绕这些算法的相关研究提供必要的理论支持和参考。

The Analysis and Comparison of Hyperspectral Endmember Extraction Algorithms

This paper summarizes six popular and cutting\|edge algorithms,including SPP\|N\|FINDR,VCA ,SPICE,PCOMMEND,MVSA and MVC\|NMF.A comprehensive comparison and analysis concludes the advantages and disadvantages of each of the six algorithms.From the experimental results show that this paper concludes that SPP\|N\|FINDR algorithm lacks the ability to resist noise when compared to the other five algorithms;VCA and MVSA are more stable than the other five algorithms; MVC\|NMF and SPICE can autonomously determine the number of endmembers and simultaneously can also obtain abundance matrix; The outcomes of PCOMMEND by true hyperspectral image which is best and gain abundance matrix,but the accuracy of this algorithm declines when the number of endmembers is prime.In the future,embracing these algorithms to process relational study,the research will offer theoretical support and consult.