正则化方法的统一  

The Unified Framework of the Regularization Techniques

作　　者：郭淑妹[1] 郭杰[1] 张宁[1] 

机构地区：[1]信息工程大学理学院,郑州市科学大道62号450001

出　　处：《大地测量与地球动力学》2015年第1期129-131,共3页Journal of Geodesy and Geodynamics

基　　金：国家自然科学基金(41174005)

摘　　要：对最小二乘估计作线性变换,使得新估计是真值的最优拟合,并且包含模型参数的先验误差协方差阵。从滤波因子的角度对正则化方法进行统一,提供了常见的Tikhonov正则化方法、截断奇异值法、广义岭回归方法等的滤波因子与对应的误差协方差阵的特征值。In this paper,the estimator is based on a linear transformation of the LS estimation.It provides an optimal solution to the true and a constrained covariance matrix.The regularization techniques are unified with the respect to filtering factors.Filtering factors and corresponding eigenvalue of error covariance matrix of some known regularization estimators as Tikhonov-Philips regularization,truncated singular value decomposition,generalized ridge regression and others are provided.

关 键 词：最小二乘估计 协方差矩阵 正则化 滤波因子 

分 类 号：P207[天文地球—测绘科学与技术]