基于Relief-F学习算法的烟叶近红外光谱特征贡献度分析

The Contribution Analysis of Tobacco Near-infrared Spectral Features Based on Relief-F Learning Algorithm

近红外光谱分析技术已被广泛应用于烟叶质量检测,检测的内容包括烟叶常规六项化学成分(总烟碱、总糖、还原糖、总氮、钾、氯)及淀粉含量等。近红外光谱是一种介于可见光和中红外之间的电磁辐射波,经过数字化处理后可以表示为近红外光谱向量,向量中每一维特征与烟叶化学成分定量分析的相关性(贡献度)是不同的,利用机器学习的分析方法对烟叶的近红外光谱特征贡献度进行综合分析,找出与烟叶品质最相关的光谱特征子集,为改进烟叶品质近红外光谱分析算法,提高烟叶品质检测准确率及执行效率,拓宽近红外光谱在烟叶品质方面的应用范围打下基础。

The analytical technology of Near-infrared spectrum has been widely used in the quality detection of tobacco leaves.The detection includes six conventional chemical components(Total Nicotines,Total Sugar,Reducing Sugar,Total Nitrogen,Potassium,Chlorine)and Starch.Near Infrared(NIR)is an electromagnetic wave between visible light(Vis)and mid-infrared(MIR)which can be expressed as a vector by digitizing.The correlation(contribution)of each dimension in the vector is different.This paper analyzes the contribution of the near-infrared spectral features of tobacco leaves used the method of machine learning,and finds out the subset of spectral features which has higher relevance to the quality of tobacco leaves.The results can be used to improve the accuracy and efficiency of the the near-infrared spectrum analysis algorithm of tobacco quality detection,and make a foundation for broadening the application of near-infrared spectrum in tobacco quality detection.