激光与光电子学进展, 2018, 55 (5): 053004, 网络出版: 2018-09-11
基于降维正则化多项式的光谱反射率重建方法 下载: 1081次
Spectral Reflectance Reconstruction Based on Dimension Reduction Regularization Polynomials
光谱学 光谱反射率重建 降维正则化 多项式回归扩展 主成分分析 spectroscopy spectral reflectance reconstruction dimension reduction regularization polynomial regression expansion principal component analysis
摘要
针对光谱反射率重建常用方法中主成分分析法重建后产生病态的问题,提出一种基于降维正则化多项式的光谱反射率重建方法;利用主成分分析法对训练样本的高维光谱数据进行降维,在降维的基础上对样本的通道响应数进行多项式回归扩展来提高光谱反射率重建的精度,同时加入Tikhonov限制条件来避免多项式扩展导致的数据不稳定性和随机噪声产生的病态问题。结果表明:降维正则化多项式光谱反射率重建方法在精度评价中的效果优于主成分分析法和多项式回归扩展法,同时实现了降低光谱数据计算量、优化通道响应、提高反射率重建精度的目的。
Abstract
To solve problems in common algorithms for spectral reflectance reconstruction such as the principal component analysis method producing ill-posed situation after reconstruction, we propose a spectral reflectance reconstruction method based on dimension reduction regularization polynomials. The principal component analysis method is used to conduct dimension reduction for high-dimensional spectral data of training samples. Based on the dimension reduction, the polynomial regression expansion is carried out for channel response numbers of the samples to improve the accuracy of spectral reflectance reconstruction, and Tikhonov restrictions are added to avoid ill-posed situation produced by data instability and random noise due to polynomial expansion. The results show that the precision evaluation effect of the proposed spectral reflectance reconstruction method based on dimension reduction regularization polynomials is better than that of the principal component analysis method and the polynomial regression expansion method. The proposed method can reduce the amount of spectral data, optimize the channel response, and improve the accuracy of reflectance reconstruction.
王可, 王慧琴, 龙艳群, 王伟超, 赵丽娟, 杨蕾. 基于降维正则化多项式的光谱反射率重建方法[J]. 激光与光电子学进展, 2018, 55(5): 053004. Ke Wang, Huiqin Wang, Yanqun Long, Weichao Wang, Lijuan Zhao, Lei Yang. Spectral Reflectance Reconstruction Based on Dimension Reduction Regularization Polynomials[J]. Laser & Optoelectronics Progress, 2018, 55(5): 053004.