基于L1范数线性回归的NIR光谱产地鉴别算法  

NIR Spectral Origin Identification Algorithm Based on L1-norm Linear Regression Classification

作　　者：但松健[1] Dan Songjian(School of Continuing Education,Chongqing No.2 Nonnal University,Chongqing 400067,China)

机构地区：[1]重庆第二师范学院继续教育学院,重庆400067

出　　处：《安徽电子信息职业技术学院学报》2020年第3期1-7,共7页Journal of Anhui Vocational College of Electronics & Information Technology

基　　金：重庆市教委科学技术研究计划项目(KJQN20191620);重庆第二师范学院校级科研项目(KY201711B)。

摘　　要：为了建立更准确、高效的柑橘产地鉴别模型,提出了基于L1范数线性回归(L1-LRC)的产地鉴别算法。该方法利用L1正则化的学习方法进行最小误差重构分类,从而将特征选择和分类器学习过程进行有机的融合,并能够更有效地反应光谱信息中的结构特征。实验结果表明,基于L1-LRC的NIR分析方法仅利用少量的样本就能够达到较高的识别精度,且获得了明显优于其他对比模型的结果,从而为快速高效的NIR产地分析提供了一个新的思路。In order to establish a more accurate and efficient oranges origin identification model,the classification model of geographical origins based on L1-norm linear regression classification(L1-LRC)is proposed.In this method,L1 regularized learning method is used for minimum error reconstruction classification,so that feature selection and classifier learning process are organically integrated,and the structural features in spectral information can be more effectively reflected.The experimental results show that the NIR analysis method based on L1-LRC can achieve high recognition accuracy with only a small number of samples,and the results are significantly better than other comparative models,thus providing a new idea for rapid and efficient NIR origin analysis.

关 键 词：L1范数线性回归 NIR光谱 产地鉴别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]