A review of machine learning in hyperspectral imaging for food safety

Abstract

Manual detection methods such as human visual inspection are not quantitative and could lead to inconsistencies in food safety assessments. Conversely, traditional laboratory techniques offer quantitative assessments, but they involve expensive equipment, are time-consuming, and are destructive to the samples. To address these limitations, advances in non-destructive monitoring techniques with the implementation of machine learning (ML) algorithms can be alternative solutions. For instance, hyperspectral imaging technology, which combines spatial and spectral data to acquire a data-rich hypercube, can be integrated with ML models to assess food safety without damaging the samples. Different from the existing review studies on ML models, this review domain focuses more on staple foods and how these ML algorithms can quantify the chemical constituents in staple food sources. This study aims to differentiate the various ML models employed in food safety and discusses the challenges and future directions for effectively quantifying samples like adulterants in foods to ensure food safety. In addition, a bibliometric analysis of ML algorithms was also conducted to understand the research trends in hyperspectral imaging and ML. Besides, this review study also addresses different image-sensing technologies and contributes to research pursuing ML and deep learning for food safety purposes in agriculture.