Enhancing wheat flour origin traceability by using laser-induced breakdown spectroscopy and Raman spectroscopy

With the increasing demand for the multi-dimensional composition detection and rapid traceability of wheat flour, this study presents the application of a combination of laser-induced breakdown spectroscopy (LIBS) and Raman Spectroscopy for traceability of wheat flour from five provinces of China, i...

Full description

Saved in:
Bibliographic Details
Main Authors: Minghui Gu, Chao Liu, Hansong Huang, Xin Zhang, Jiguo Li, Qingbin Jiao, Liang Xu, Mingyu Yang, Xin Tan
Format: Article
Language:English
Published: Elsevier 2025-07-01
Series:Results in Chemistry
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2211715625004291
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:With the increasing demand for the multi-dimensional composition detection and rapid traceability of wheat flour, this study presents the application of a combination of laser-induced breakdown spectroscopy (LIBS) and Raman Spectroscopy for traceability of wheat flour from five provinces of China, including Henan, Shanxi, Anhui, Shandong, and Hebei. A 2D Convolutional Neural Network (2D CNN) based on feature selection was proposed for origin classification. First, a hybrid feature selection method, termed ANOVA-Sine Cosine Algorithm (AVSCA), was developed by integrating analysis of variance (ANOVA) with the Sine Cosine Algorithm (SCA) to extract spectral fingerprint information of wheat flour. Next, we established a 2D CNN model by transforming the one-dimensional spectral data into a square matrix, and trained it with LIBS-Raman data under low-level, mid-level and high-level fusion strategies. Then, the model’s metrics were obtained using 10-fold cross-validation optimized by Euclidean distance. Finally, to enhance the practical applicability of the model, model parameter transfer techniques were introduced, with a small amount of wheat flour spectral data from five other provinces used as the training set to fine-tune the established 2D CNN model. The results showed that the proposed method can enhance the ability of origin traceability, and high-level fusion strategies perform the best, achieving an average accuracy of 98%. The transfer model achieved a prediction accuracy of 97% on the remaining data, demonstrating the effectiveness of transfer learning.
ISSN:2211-7156