Front-face Excitation-Emission Matrix Fluorescence Spectroscopy and Interpretable Deep Learning Unveil Storage Years of Ningxia Wolfberry

In an effort to combat the deceptive practice of selling aged Ningxia wolfberry as fresh produce, researchers at Hunan University in Changsha, PR China, have developed a novel method for rapidly identifying the storage year of this valuable fruit. Their study, published in Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, introduces the combination of front-face excitation-emission matrix (FF-EEM) fluorescence spectroscopy and interpretable deep learning as a powerful tool for this purpose (1).

Fructus lychii goji wolfberry | Image Credit: © Le Do - stock.adobe.com

FF-EEM fluorescence spectroscopy is a technique that involves illuminating a sample with various wavelengths of light and measuring the corresponding emitted fluorescence. It provides a comprehensive spectral fingerprint of the sample, capturing its unique fluorescence characteristics. By analyzing the excitation and emission spectra using deep learning, valuable chemical information about the sample can be extracted. Deep learning is a branch of machine learning that uses artificial neural networks to learn and make predictions from complex data.

Ningxia wolfberry, known for its antioxidant properties and health benefits, fetches premium prices in the market. However, unscrupulous traders may attempt to pass off old wolfberries as freshly harvested ones, leading to potential consumer deception and financial losses. The proposed method aims to address this challenge by providing a lossless, fast, and accurate means of determining the storage year of Ningxia wolfberry.

The researchers employed the alternating trilinear decomposition (ATLD) algorithm to extract chemically significant information from the three-way data array obtained from Ningxia wolfberry samples. This preprocessing step allowed for the isolation of key features related to the storage years of the fruit. Furthermore, the team introduced a convolutional neural network (CNN) model called EEMnet, specifically designed for the identification of storage years. EEMnet leverages the subtle spectral differences among wolfberry samples to achieve robust classification.

Remarkably, the EEMnet model exhibited a correct classification rate exceeding 98% for the training set, test set, and prediction set, showcasing its effectiveness in distinguishing wolfberry samples of different storage years. To enhance transparency and interpretability, the researchers conducted a series of analyses to unravel the inner workings of the deep learning model, offering insights into the decision-making process.

The results of this study demonstrate that the combined approach of FF-EEM fluorescence spectroscopy and EEMnet holds great potential for rapid and accurate identification of the storage year of Ningxia wolfberry. By providing a green and reliable solution, this method offers a valuable contribution to the identification of storage years for Chinese medicinal materials.

As the demand for high-quality agricultural products continues to rise, the development of innovative techniques like FF-EEM fluorescence spectroscopy coupled with interpretable deep learning paves the way for improved food authentication, ensuring consumer confidence and safeguarding the integrity of the market. The findings from this research could potentially be applied to other valuable food products, enabling reliable traceability and quality control across various industries.

Reference

(1) Yan, X.-Q.; Wu, H.-L.; Wang, B.; Wang, T.; Chen, Y.; Chen, A.-Q.; Huang, K.; Chang, Y.-Y.; Yang, J.; Yu, R.-Q. Front-face excitation-emission matrix fluorescence spectroscopy combined with interpretable deep learning for the rapid identification of the storage year of Ningxia wolfberry. Spectrochimica Acta Part A: Mol. Biomol. Spectrosc. 2023, 295, 122617. DOI: 10.1016/j.saa.2023.122617