Food and agro-product quality evaluation based on spectroscopy and deep learning: A review

Rapid and non-destructive infrared spectroscopy has been applied to both internal and external quality evaluations of food and agro-products. Various linear and nonlinear chemometric methods have been developed for spectral analysis. The generalizability of previous chemometric methods is hindered by changing noise under various detection conditions and biological variabilities. Recently, deep learning approaches have been developed for spectral noise reduction, feature extraction, and calibration regression modeling. This review discusses the current challenges of conventional chemometric methods and the emerging deep learning approach for spectral analysis. The current state-of-the-art techniques, including unsupervised feature extraction and noise reduction models and supervised multivariate regression approaches, have been addressed in this review. The research on exploring the learning mechanism of the ‘black box’ deep learning model is also discussed. This review focuses on the application of deep learning approaches on quality evaluation of food and agro-products, lessons from current studies, and future perspectives. The deep learning approach combined with spectroscopic sensing techniques has shown great potential for quality evaluation of food and agro-products. Current advances in deep learning-based qualitative analysis include variety identification, geographical origin detection, adulteration recognition, and bruise detection, whereas quantitative analysis includes multiple component content prediction for fruits, grains, and crops. The main advantage of deep learning approach is the decreasing the dependence on human domain knowledge by end-to-end analysis and the improved precision and generalizability.