Research on the Application of Terahertz Technology in Detecting Additives in Milk Powder

• Published in: Food analytical methods Vol. 18; no. 3; pp. 398 - 415
• Main Authors: Zhang, Hongtao, Gao, Jiahui, Tan, Lian, Zheng, Li, Wang, Longjie
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.03.2025; Springer Nature B.V
• Subjects: Adaptive algorithms; Adaptive sampling; Additives; Algorithms; Analytical Chemistry; Ant colony optimization; Babies; Baby foods; Back propagation; Chemistry; Chemistry and Materials Science; Chemistry/Food Science; Chromatography; Concentration gradient; dried milk; Food additives; Food safety; Food Science; Long short-term memory; Microbiology; Milk; Moisture content; neural networks; Nondestructive testing; Particle swarm optimization; Powder; Preprocessing; Quantitative analysis; Recognition; Regression analysis; Regression models; Scientific imaging; spectral analysis; spectrometers; Spectroscopy; Support vector machines; Terahertz frequencies; Vanillin; Spectroscopy; Milk powder; Detection of additives; Terahertz time-domain spectroscopy
• ISSN: 1936-9751; 1936-976X
• DOI: 10.1007/s12161-024-02720-8

Milk powder is a common food in most families. It is of great significance to accurately detect the quality and safety of milk powder to mitigate food safety problems. This paper presents a method for the determination of vanillin and ethyl vanillin in milk powder based on terahertz (THz) spectroscopy. Samples with varying concentration gradients of these two additives were prepared, and a terahertz time-domain spectrometer was used to collect spectral data from the samples in the 0.2 to 1.5 THz range. Seven spectral preprocessing algorithms were evaluated using the partial least squares (PLS) method, and it was found that the combination of multivariate scattering correction (MSC) and Savitzky-Golay (SG) smoothing preprocessing yielded the best results, significantly improving the accuracy of the test sets for both additives. Subsequently, nine quantitative detection methods were constructed by combining three dimensionality reduction algorithms (ant colony algorithm (ACO), competitive adaptive reweighted sampling (CARS), and successive projection algorithm (SPA)) with three regression models (support vector regression (SVR), long short-term memory (LSTM), and particle swarm optimization-back propagation (PSO-BP)). The results showed that the LSTM regression model, with dimensionality reduction performed by the CARS algorithm, performed best for detecting vanillin in milk powder, achieving a recognition rate of 94.49%. Compared to the other eight methods, this increased the recognition rate by 7.69%. Similarly, the LSTM regression model, combined with the SPA algorithm for dimensionality reduction, performed best for detecting ethyl vanillin in milk powder, reaching a recognition rate of 98.37%. This represented a 6.59% increase in recognition rate over the other eight methods, providing a novel technical approach for non-destructive testing and analysis of milk powder quality and safety.