Machine Learning and Deep Learning for Food Quality Assessment Using Hyperspectral Imaging: Advances, Challenges, and Future Directions

Maintaining food quality and safety standards is essential for reducing postharvest losses, sustaining consumer confidence, and meeting international trade requirements. Hyperspectral imaging (HSI) has emerged as a powerful non-destructive technique for assessing both internal and external quality attributes of horticultural products. This review provides a systematic synthesis of recent advances in HSI-based quality assessment of fruits and nuts using machine learning and deep learning approaches. Applications are categorised into defect and non-defect quality attributes, including moisture content, sugar concentration, ripeness, maturity, firmness, nutritional composition, microbial contamination, structural damage, and chemical indicators associated with toxicity and rancidity, such as aflatoxin, free fatty acids (FFAs), and peroxide value (PV). The review critically evaluates spectral preprocessing techniques, modelling approaches, model evaluation criteria, and the influence of sample size and data variability on predictive performance. In addition, commonly reported wavelengths associated with moisture, sugar concentration, firmness, peroxide value, and internal discolouration are systematically compared to identify recurring and overlapping spectral regions across different horticultural products. To the best of our knowledge, this is the first review to demonstrate that multiple quality attributes across fruits and nuts share common important spectral regions, highlighting opportunities for the development of more efficient, transferable, and generalisable HSI models. Finally, key future research directions are identified, including the development of balanced and standardised datasets, transformer-based architectures for improved generalisation, and red-green-blue (RGB) reconstruction of hyperspectral images to enhance interpretability. The advances discussed in this review have strong potential to improve scalability and facilitate wider industrial adoption of HSI technologies for postharvest quality monitoring and food safety assessment.