Multi-task deep learning utilizing VOCs detected by HS-GC-IMS for geographical origin tracing and adulteration detection of turmeric powder

Xiao Han, Tong Wu, Shixin Cen, Jie Yong, Xinrang Tian, Xinyi Li, Rui Sun, Minghui Liu, Heshui Yu, Zheng Li,
Multi-task deep learning utilizing VOCs detected by HS-GC-IMS for geographical origin tracing and adulteration detection of turmeric powder,
Food Control,
Volume 178,
2025,
111522,
ISSN 0956-7135,
https://doi.org/10.1016/j.foodcont.2025.111522.
(https://www.sciencedirect.com/science/article/pii/S0956713525003913)
Abstract: The widespread use of turmeric powder in global cuisines, coupled with its high market demand, has led to its susceptibility to economically motivated adulteration. To address this issue, this study developed a detection method based on Headspace-Gas Chromatography-Ion Mobility Spectrometry (HS-GC-IMS). A multi-task learning (MTL) model utilizing deep learning algorithms was constructed to perform geographical origin tracing and adulteration levels prediction simultaneously. The performance of this approach was evaluated against traditional chemometric methods and single-task learning (STL) models based on machine learning algorithms. The comparison results indicated that chemometric methods struggled to differentiate turmeric origins and adulteration levels accurately. While the STL model achieved high accuracy in geographical origin tracing, it exhibited limitations in forecasting adulteration levels. In contrast, the MTL model not only precisely classified turmeric origins but also effectively forecasted adulteration levels, achieving a coefficient of determination of 0.9651, a root mean square error of 0.5640, and a mean absolute error of 0.4296. These findings highlight the superiority of the MTL model in addressing both classification and regression tasks. In conclusion, this study successfully established an MTL-based HS-GC-IMS method, offering a rapid, accurate, and reliable solution for detecting turmeric powder adulteration and ensuring food safety.
Keywords: Turmeric powder; HS-GC-IMS; Geographical origin tracing; Adulteration; Deep learning; Multi-task learning