遗传算法及其在遥感线性、非线性模型反演中的应用效果分析

分别以2类常见的遥感模型--线性光谱混合模型和GOMS模型为代表研究了遗传算法在遥感线性和非线性模型反演中的应用效果,并与逐步二次规划法等确定性搜索算法进行了比较.结果表明,遗传算法在线性遥感模型反演中的优势并不明显.而在非线性遥感模型反演中优于确定性搜索算法.这主要是因为对于线性模型来说,一般定义的代价函数为凸函数,大多优化算法可以收敛于全局最优解,相比较而言,遗传算法因其搜索效率低,故优势不明显.而在非线性模型反演中,代价函数的形式比较复杂,可能是非凸的,因此遗传算法的优势得以发挥,这也说明与一般确定性搜索算法相比,遗传算法具有更好的全局收敛性.

EFFECTS OF GA ON THE INVERSION OF LINEAR AND NONLINEAR REMOTE SENSING MODELS

Iteration is an important way to invert complicate remote sensing models. The global convergence of the algorithm used in this method should be considered. GA is a randomly searching algorithm, which is currently considered to be relatively mature and have good global convergence ability. In this paper, GA is applied to invert two common remote sensing models, Linear Spectral Mixing model and GOMS model, to analyze the effect of GA on the inversion of linear and nonlinear remote sensing models. The results show that GA is not better than commonly used algorithms for the inversion of linear remote sensing models, but is more effective for the nonlinear models. This is mainly because that for the former, the commonly defined merit functions are convex and the maximum point can be found easily, while for the later are not. In the latter case, the superiority of GA can be exerted. This result further attests that GA has better global convergence than the other optimum algorithms.