基于波段优选的改进N-FINDR端元提取算法  

Improved N⁃FINDR endmember extraction algorithm based on band optimization

作　　者：王高飞 朱桂平[1] 张寅[1] WANG Gaofei;ZHU Guiping;ZHANG Yin(College of Astronautics,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China)

机构地区：[1]南京航空航天大学航天学院,江苏南京210016

出　　处：《电子设计工程》2024年第14期6-12,18,共8页Electronic Design Engineering

基　　金：国防科技173计划技术领域基金资助(2021-JCJQ-JJ-0834)。

摘　　要：为了解决传统N-FINDR算法光谱信息利用不足,出现重复提取端元的问题,提出基于波段优选的改进N-FINDR端元提取算法。通过对高光谱数据各波段之间的相关性进行分析,保留相关性较弱的波段,去除相关性强的冗余波段,从而高效利用高光谱图像的光谱信息,对高光谱数据进行波段优选。通过实际高光谱数据和模拟数据进行实验分析,并与现有算法进行对比,实验结果表明,基于波段优选的改进N-FINDR端元提取算法在保证算法效率的同时,端元提取的覆盖率达到100%,算法提取的端元光谱与实际地物光谱具有较高的相似性。Traditional N-FINDR algorithm does not fully use spectral information of hyperspectral images,resulting in the problem of repeated extraction of endmembers.In order to solve this problem,an improved N-FINDR endmember extraction algorithm,which is based on band optimization is proposed.It uses the spectral information efficiently by analyzing the correlation between each band of hyperspectral data,retains the band with weak correlation,and removes the redundant band with strong correlation.Through the experimental analysis of synthetic data and actual hyperspectral data.The experimental results show that the improved N-FINDR terminal element extraction algorithm based on band optimization ensures the efficiency of the algorithm,and the coverage of the end element extraction reaches 100%,and the endmenber spectral extracted by the algorithm has a high similarity with the original spectral.

关 键 词：高光谱图像 端元提取 N-FINDR 波段优选 

分 类 号：TN911.73[电子电信—通信与信息系统]