Nondestructive detection of pungent and numbing compounds in spicy hotpot seasoning with hyperspectral imaging and machine learning

Di Zhang; Xu Chen; Zitao Lin; Minmin Lu; Wenhao Yang; Xiaoxia Sun; Maurizio Battino; Jiyong Shi; Xiaode Huang; Bolin Shi; Xiaobo Zou

doi:10.1016/j.foodchem.2024.142593

Nondestructive detection of pungent and numbing compounds in spicy hotpot seasoning with hyperspectral imaging and machine learning

Food Chem. 2024 Dec 22:469:142593. doi: 10.1016/j.foodchem.2024.142593. Online ahead of print.

Authors

Di Zhang¹, Xu Chen², Zitao Lin³, Minmin Lu³, Wenhao Yang³, Xiaoxia Sun⁴, Maurizio Battino⁵, Jiyong Shi³, Xiaode Huang⁴, Bolin Shi⁶, Xiaobo Zou³

Affiliations

¹ School of Food and Biological Engineering, Jiangsu University, Zhenjiang 212013, China. Electronic address: [email protected].
² School of Food and Biological Engineering, Jiangsu University, Zhenjiang 212013, China; Faculty of Agricultural Engineering, Jiangsu University, Zhenjiang 212013, China.
³ School of Food and Biological Engineering, Jiangsu University, Zhenjiang 212013, China.
⁴ China CO-OP Nanjing Institute for Comprehensive Utilization of Wild Plants, Nanjing 211111, China.
⁵ International Research Center for Food Nutrition and Safety, Jiangsu University, Zhenjiang 212013, China; Department of Clinical Sciences, Faculty of Medicine, Polytechnic University of Marche, Ancona, Italy.
⁶ Food and Agriculture Standardization Institute, China National Institute of Standardization, Beijing 102200, China.

PMID: 39729663
DOI: 10.1016/j.foodchem.2024.142593

Abstract

The levels of capsaicin (CAP) and hydroxy-α-sanshool (α-SOH) are crucial for evaluating the spiciness and numbing sensation in spicy hotpot seasoning. Although liquid chromatography can accurately measure these compounds, the method is invasive. This study aimed to utilize hyperspectral imaging (HSI) combined with machine learning for the nondestructive detection of CAP and α-SOH in hotpot seasoning. Spectral reflectance within the range of 370-1030 nm was used to develop regression models to predict CAP and α-SOH content. The results indicated that the PSO-BPNN model was optimal for predicting CAP (R² = 0.9942) and α-SOH (R² = 0.9939). Feature selection algorithms and tallow model experiments identified characteristic wavelengths for CAP (740-800 nm and 850-940 nm) and α-SOH (450-550 nm, 650-700 nm, 740-800 nm, and 850-940 nm). These findings demonstrated the potential of HSI for rapid, precise, and nondestructive assessment of CAP and α-SOH levels in hotpot seasoning.

Keywords: CAP; HSI; Hotpot seasoning; Machine learning; Nondestructive detection; α-SOH.