|
| Model Update of Total Sugar Content in Tobacco Leaves of Different Years by Near-infrared Spectroscopy Combined with Parameter-free Calibration Enhancement |
| |
| View Full Text Download reader |
| DOI： |
| KeyWord:near infrared spectroscopy model update tobacco semi-supervised parameter-free calibration enhancement（SS－PFCE） |
| Author | Institution |
| GENG Ying-rui，SHEN Huan-chao，NI Hong-fei，WANG Hui，WU Ji-zhong，ZHANG Li-li，LI Yong-sheng，HE Wen-miao，CHEN Yong，LIU Xue-song |
1. College of Pharmaceutical Sciences，Zhejiang University，Hangzhou ，China; 2. Innovation Institute for Artificial Intelligence in Medicine of Zhejiang University，Hangzhou ，China; 3. Technology Center， China Tobacco Zhejiang Industrial Co.，Ltd.，Hangzhou ，China |
|
| Hits: 791 |
| Download times: 842 |
| Abstract: |
| The near infrared spectroscopy technology has been widely applied to the quantitative analysis in tobacco industry because of its advantages of rapidness and non-destructiveness．However，the accuracy and robustness of the original model may deteriorate when predicting samples with new variations．When samples are obtained from different harvest years and different environmental conditions，new variations will be introduced．Therefore，there is a need to maintain the predictive performance of the original model when it works on a new batch．In this study，a method called semi-supervised parameter-free calibration enhancement（SS－PFCE） was used to update the original model．The regression coefficient of original model was corrected by constrained optimization．The original model for total sugar determination was firstly developed with the tobacco samples of 2016，which showed a predicted correlation coefficient（Rp） of 0.997 8 and a root mean square error of prediction（RMSEP） of 0.310 8．After updating the model by the SS－PFCE method，the total sugar contents in the samples of 2017，2018 and 2020 years were predicted，respectively．The Rp values of the three test sets were improved by 0.13%，1.32% and 4.29%，and the RMSEP were decreased by 15.26%，58.69% and 36.53%，respectively，compared with the non-updated model．Furthermore，the updated model by the SS－PFCE approach offered a better predictive performance than re-modeling method，while significantly reduced modeling costs．The results of this study showed that the SS－PFCE method could maintain the prediction accuracy for tobacco samples of different years efficiently，and it is of great practical application value in industrial production． |
| Close |