Jiaxiao Cai
Research Subjects
Jiaxiao Cai has a total of 17 co-authors. Here are the top co-authors by number of shared publications:
- Aiming Chen
Department of R & D, Dalian Chem Data Solution Information Technology Co. Ltd., Dalian 116023, Liaoning, China
1 shared publication Bo Kong
1 shared publication- Bo Kong
Technology Center, China Tobacco Hunan Industrial Co., Ltd., Changsha 410014, China
1 shared publication Hongbing Lu
Technology Center of China Tobacco Hunan Industrial Co., Ltd., Changsha, 410007, China
1 shared publication- Hui Jiang
Department of R & D, Dalian Chem Data Solution Information Technology Co. Ltd., Dalian 116023, Liaoning, China
1 shared publication - Kejun Zhong
Technology Center of China Tobacco Hunan Industrial Co., Ltd., Changsha, 410007, China
1 shared publication
Latest Publications
- Open Access
- 21 October 2026
- Published by Elsevier in SSRN Electronic Journal
Enhanced Discrimination of Natural Flavoring Materials Using Near- and Mid-Infrared Spectroscopy and Chemometrics
The compositional complexity of natural flavoring materials presents challenges for rapid and reliable discrimination in food and tobacco quality control. This study evaluated near-infrared (NIR) and mid-infrared (MIR) spectroscopy using 220 samples collected from 14 categories. Partial least squares discriminant analysis (PLS-DA) firstly screened all 91 category pairs, and further identified three challenging comparisons involving coffee extract, cocoa extract, maple concrete, and Maillard reaction products. Five pre-processing strategies, three variable selection methods, and three classification algorithms were respectively compared alongside feature-level NIR-MIR fusion. Standard normal variate combined with the Savitzky-Golay second derivative (SNV+SG2D) performed best for NIR, whereas SNV combined with the first derivative (SNV+SG1D) was preferred for MIR. MIR generally outperformed NIR, particularly for categories showing substantial spectral overlap. Variable importance in projection (VIP) selection improved NIR classification accuracy from 72.73-80.95% to 88.10-96.97%, while MIR accuracy reached 95.24-100%. Selected spectral regions were associated with carbonyl, aromatic, and carbohydrate-related vibrations. PLS-DA achieved the best overall performance, while VIP and interval partial least squares (iPLS) favored dispersed and localized spectral features, respectively. Feature-level fusion increased cross-validated predictive ability (Q2) for all three challenging pairs and improved classification accuracy for maple concrete versus Maillard reaction products from 96.97% to 100%. In four-class classification, MIR achieved 78.05% accuracy compared with 60.98% for NIR. These findings highlight the complementary value of spectroscopy and chemometrics for discriminating natural flavoring materials.
- 21 October 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessEnhanced Discrimination of Natural Flavoring Materials Using Near- and Mid-Infrared Spectroscopy and Chemometrics
The compositional complexity of natural flavoring materials presents challenges for rapid and reliable discrimination in food and tobacco quality control. This study evaluated near-infrared (NIR) and mid-infrared (MIR) spectroscopy using 220 samples collected from 14 categories. Partial least squares discriminant analysis (PLS-DA) firstly screened all 91 category pairs, and further identified three challenging comparisons involving coffee extract, cocoa extract, maple concrete, and Maillard reaction products. Five pre-processing strategies, three variable selection methods, and three classification algorithms were respectively compared alongside feature-level NIR-MIR fusion. Standard normal variate combined with the Savitzky-Golay second derivative (SNV+SG2D) performed best for NIR, whereas SNV combined with the first derivative (SNV+SG1D) was preferred for MIR. MIR generally outperformed NIR, particularly for categories showing substantial spectral overlap. Variable importance in projection (VIP) selection improved NIR classification accuracy from 72.73-80.95% to 88.10-96.97%, while MIR accuracy reached 95.24-100%. Selected spectral regions were associated with carbonyl, aromatic, and carbohydrate-related vibrations. PLS-DA achieved the best overall performance, while VIP and interval partial least squares (iPLS) favored dispersed and localized spectral features, respectively. Feature-level fusion increased cross-validated predictive ability (Q2) for all three challenging pairs and improved classification accuracy for maple concrete versus Maillard reaction products from 96.97% to 100%. In four-class classification, MIR achieved 78.05% accuracy compared with 60.98% for NIR. These findings highlight the complementary value of spectroscopy and chemometrics for discriminating natural flavoring materials.