CAJ | 학술논문

多酚类物质是烟叶香气产生的重要前体物质。采用近红外光谱结合偏最小二乘算法建立了烟叶中总多酚含量的回归模型。采用建立的模型对检验集进行预测，预测集决定系数R2为0.8671，模型误差S E P为1.4287。结果表明近红外光谱分析技术可以成功应用于烟叶中总多酚含量的检测。此外，为消除烟叶近红外光谱中无效波长变量，采用无消息变量消除算法对所建近红外模型进行优化。结果表明采用该算法后，剩余变量数得到减少，模型维数显著降低，预测性能有所提高。
다분류물질시연협향기산생적중요전체물질。채용근홍외광보결합편최소이승산법건립료연협중총다분함량적회귀모형。채용건립적모형대검험집진행예측，예측집결정계수R2위0.8671，모형오차S E P위1.4287。결과표명근홍외광보분석기술가이성공응용우연협중총다분함량적검측。차외，위소제연협근홍외광보중무효파장변량，채용무소식변량소제산법대소건근홍외모형진행우화。결과표명채용해산법후，잉여변량수득도감소，모형유수현저강저，예측성능유소제고。
Polyphenols are important aroma precursors in tobacco. In this paper, near-infrared spectroscopy combined with partial least squares algorithm was used for establishing regression model of total polyphenol content in tobacco. For test set, model R2 is 0.8671, SEP is 1.4287. Results show that near-infrared spectroscopy technology can be successfully applied to the total polyphenol content detection in tobacco. In order to eliminate interference variables in tobacco near infrared spectrum, Uninformative Variables Elimination algorithm was used. Experimental results show that after the algorithm used, remaining number of variables is reduced, modeling dimension is significantly reduced, and prediction performance is improved.