DEM构建的加权最小二乘支持向量机算法  

Robust Algorithm of Weighted Least Squares Support Vector Machine and Its Application in DEM

作　　者：李明飞[1] 陈传法[1] 戴洪磊[1] 李翼龙[1] 李宇航[1] 

机构地区：[1]山东科技大学测绘科学与工程学院,山东青岛266590

出　　处：《地理空间信息》2015年第6期82-83,13,共2页Geospatial Information

基　　金：山东省泰山学者建设工程专项经费资助项目

摘　　要：为了减少采样点中的粗差对DEM构建精度的影响,将最小二乘支持向量机(LSSVM)运用到DEM中,发展了一种DEM构建的新算法,即加权最小二乘支持向量机(LSSVM-W)。LSSVM-W以LSSVM为基础,依据LSSVM计算的拉格朗日系数阵α求出各误差变量的权重系数,再求出更好的拉格朗日系数阵α,这样迭代多次,最终获得满足拟合精度的系数阵α。文中以数值模拟曲面为研究对象,分析并比较了当采样误差为不同污染率的正态分布时LSSVM与LSSVM-W模拟结果的精度。模拟结果表明,LSSVM-W的抗差性优于LSSVM,是一种较好的DEM构建方法。A robust algorithm of weighted least squares support vector machine(LSSVM-W) was developed to reduce the effect of outliers on DEM construction. LSSVM-W used the result(parameters array α) of the LSSVM as the initial values to compute the weights of sampling points. Then, a iteratively LSSVM-W was formed to compute a more perfect parameters array. A synthetic surface was employed to comparatively analyze the estimation accuracies of LSSVM and LSSVM-W,where the sampling points were contaminated by the standard normal distribution, contaminated normal distribution with the contaminating proportion of 5%, 10%, 20% and 30%. The tests indicate that the accuracy of LSSVM-W is better than LSSVM. LSSVM-W is a good way to build DEM construction.

关 键 词：LSSVM 粗差 抗差 DEM LSSVM-W 

分 类 号：P207[天文地球—测绘科学与技术]