Front-face excitation-emission matrix fluorescence spectroscopy combined with interpretable deep learning for the rapid identification of the storage year of Ningxia wolfberry

Ningxia wolfberry stored for many years may be disguised as fresh wolfberry by unscrupulous traders and sold for huge profits. In this work, the front-face excitation-emission matrix (FF-EEM) fluorescence spectroscopy coupled with interpretable deep learning was proposed to identify the storage year of Ningxia wolfberry in a lossless, fast and accurate way. Alternating trilinear decomposition (ATLD) algorithm was used to decompose the three-way data array obtained by Ningxia wolfberry samples, extracting the chemically meaningful information. Meanwhile, a convolutional neural network (CNN) model for the identification of the storage year of Ningxia wolfberry, called EEMnet, was proposed. The model successfully classified wolfberry samples from different storage years by extracting the subtle feature differences of the spectra, and the correct classification rate of the training set, test set and prediction set was more than 98%. In addition, a series of interpretability analyses were implemented to break the "black box" of the deep learning model. These results indicated that the method based on FF-EEM fluorescence spectroscopy combined with EEMnet could quickly and accurately identify the year of Ningxia wolfberry in a green way, providing a new idea for the identification of the storage years of Chinese medicinal materials.