基于离散粒子群算法的凸多模态高光谱图像端元提取研究

Piecewise Convex Mulutiple-model Hyperspectral Imagery ENDmember-extraction based on Discrete Particle Swarm Optimization

为了提高以凸几何模型为基础的标准线性解混模型的计算精度,利用离散粒子群算法((Discrete Particle Swarm Optimization,D-PSO)来求解分段凸多模态高光谱图像端元提取。D-PSO为智能随机搜索算法,适合求解非凸函数的全局最优解,减少由凸分段数不确定性引起的解混误差。选取了16个地物作为端元,利用D-PSO方法求得其解混丰度反演结果,实验结果表明:端元位置与地面真值图上的端元位置互相匹配,D-PSO方法能够更有效地检测出高光谱图像中的目标,是一种行之有效的解混方法。

Piecewise COnvex Multiple-Model ENDmember（PCOMMEND）spectral unmixing can well solve unmixing of the nonconvex hyperspectral data,which improves the calculation accuracy of the standard linear mixed model based on the convex geometry model.the number of piecewise convex is not sure in the practical application,which limits the calculation ccuracy of unmixing and the wrong endmembers will sometimes extracted,in view of the situation,the Discrete Particle Swarm Optimization（D-PSO）is proposed to unmix the piecewise convex mulutiple-model hyperspectral imagery,D-PSO is the intelligent algorithm of random search,and is able to find the global optimal solution of convex function,which reduce the unmixing error caused by the uncertainty number of the convex section,experiments on the simulative data and real data has indicate D-PSO improves the accuracy of the extracting endmember and estimating the proportion.