Non-destructive determination of total sugar content in tobacco filament based on calibration transfer with parameter free adjustment

The advantages of near-infrared spectroscopy (NIR) for non-destructive determination have an important impact on quality assessment in the tobacco industry. However, spectral measurements are very sensitive to external conditions, such as moisture content and sample morphology. The accuracy and stability of analytical models are often affected by variations in sample spectra. In this study, an efficient non-supervised parameter-free calibration enhancement (NS-PFCE) method was proposed for the calibration transfer from dried tobacco powder to undried tobacco samples in order to correct the effects of moisture and material scattering on spectral absorption. While comparing the transfer effects of piecewise direct standardization (PDS), spectral space transformation (SST), and slope/bias correction (S/B) methods. The results showed that the best prediction (RMSEP = 0.9800) was obtained with the NS-PFCE method, with a decrease of 14.19 %, 2.26 %, and 27.19 % in RMSEP compared to the prediction results calibrated transferred by PDS, SST and S/B, respectively. In addition, to further optimize the calibration effect and improve the transfer efficiency, competitive adaptive reweighted sampling (CARS), variable combination population analysis employing the iteratively retaining informative variables (VCPA-IRIV), and variable combination population analysis genetic algorithm (VCPA-GA) were employed to select the important wavelengths to simplify the model. The results showed that the VCPA-IRIV-PLSR model combined with NS-PFCE method obtained the optimal prediction results of undried filament samples (RMSEP = 0.7252), which were 50.08 %, 6.74 % and 33.10 % lower than those RMSEP values by the other three methods under the same treatment conditions, respectively. Therefore, it can be concluded that NS-PFCE method combined with VCPA-IRIV-PLSR model can effectively compensate for the spectral differences caused by moisture and sample morphology, and achieve an excellent quantitative analysis of moist samples in non-homogeneous form. The method helps to achieve rapid non-destructive testing at-line in the tobacco industry.

Automated summary

The study proposes an efficient non-supervised parameter-free calibration enhancement method for quality assessment in the tobacco industry. It corrects the effects of moisture and material scattering on spectral absorption in the calibration transfer from dried tobacco powder to undried tobacco samples. Compared to other calibration methods, the proposed method achieves the best prediction results. Additionally, competitive adaptive reweighted sampling, variable combination population analysis, and variable combination population analysis genetic algorithm are employed to select important wavelengths and optimize the calibration effect and transfer efficiency.