基于烟丝近红外光谱的卷烟品牌识别方法

Cigarette brand recognition based on NIR spectral data of cut filler

为实现对卷烟品牌的快速识别,基于烟丝的近红外光谱数据,结合机器学习技术,以贵州中烟工业有限责任公司生产的10种品牌卷烟为对象,建立了一种卷烟品牌识别模型,并以正确识别率(Recognition Accuracy,RA)为评价指标对模型的各项关键参数进行迭代优化。利用采集的卷烟样品数据对模型进行验证,结果表明:采用连续小波变换(CWT)方法进行光谱数据预处理,概率主成分分析(PPCA)方法进行数据降维,选择Linear作为核函数,基于支持向量机(SVM)方法建立的识别模型,最高RA值达到97.20%,表明利用烟丝光谱数据可以实现对卷烟品牌的准确识别。该技术可为卷烟配方维护提供支持。

In order to recognize cigarette brands rapidly,a pattern recognition model for 10 cigarette brands made by China Tobacco Guizhou Industrial Limited Corporation was established based on the NIR spectral data of cut filler and machine learning method.The parameters in the model were optimized iteratively corresponding to the best recognition accuracy,and the model was verified with the collected data of cigarette samples.The results showed that by applying CWT to spectra processing,PPCA to data dimension reduction,and the SVM of linear kernel function to recognition model establishment,the RA value might reach 97.20%,which indicated that cigarette brands could be recognized accurately via the spectral data of cut filler.This technology provides a support for cigarette blend maintenance.