摘要
针对采用最大体积单体MVS(Maximization Volume Simplex)端元提取算法进行端元初选时存在相似端元光谱问题,提出一种光谱信息散度SID(Spectral Information Divergence)和光谱梯度角SGA(Spectral Gradient Angle)相结合以区分两个相似端元光谱的方法。该方法对经过端元初选之后的端元子集进行端元的二次选择,采用以SID_SG作为最相似端元选择的判据,除去相似端元,降低相似端元对解混精度的影响,利用全约束最小二乘法进行丰度估计。实验结果表明,提出的优化方法与传统方法相比,提高了端元的选择精度,重构影像与原始影像之间的均方根误差RMSE(Root Mean Square Error)也有所降低,分布更加均匀。该方法对高光谱遥感影像进行深度解译具有十分重要的意义。
For the problem of maximisation volume simplex( MVS) endmember extraction algorithm that when applying in primary endmembers selection it would have similar endmembers spectra,we proposed an algorithm which distinguishes two similar endmember spectra by combining the spectral information divergence( SID) and the spectral gradient angle( SGA). This algorithm carries out secondary selection on the endmember subset derived from primary endmembesr selection,and adopts SID_SG rule as the criteria for selecting the most similar endmembers to remove the similar endmembers and to reduce the influence of similar endmembers on unmixing accuracy,and uses the full constraint least square for abundance estimation. Experimental results showed that proposed optimisation algorithm improved the accuracy of endmember selection than the traditional method,the root mean square error( RMSE) between reconstruction images and original image was rather reduced,and the distribution was more evenly as well. The algorithm is of very great significance to the deep interpretation on hyperspectral remote sensing image.
出处
《计算机应用与软件》
CSCD
2016年第7期252-256,共5页
Computer Applications and Software
基金
国家高技术研究发展计划项目(2012AA12A405)
国家自然科学基金项目(61172144)
关键词
光谱解混
端元初选
二次提取
除去端元
解混算法
Spectral unmixing
Primary endmembers selection
Secondary extraction
Removal of endmember
Unmixing algorithm
