Advances in Hyperspectral Image Reconstruction From RGB Images for Food Quality Evaluation.

Abstract

Hyperspectral imaging (HSI) provides detailed spectral and spatial data, making it an essential tool for non-destructive and rapid assessment of quality attributes in agricultural products. Despite its numerous applications in the food industry, widespread adoption has been limited due to the high cost and complexity of traditional hyperspectral imaging systems. To address these challenges, hyperspectral reconstruction from RGB images has emerged as a promising solution. RGB images/data can be obtained from simpler and cheaper sensors such as those found on cellphones. With recent advancements in deep learning methods, it is possible now to reconstruct hyperspectral images from RGB images, bridging the gap between the high spatial resolution of RGB images and the spectral depth of HSI. While promising results have demonstrated the feasibility of this emerging technology, challenges remain in generalizing across diverse food matrices and extending spectral coverage to the full visible and near-infrared range. Also, there is limitation of capability of the reconstructed data in terms of inherent chemometrics which limits accuracy of quantification. There is need also for high computational power for the reconstruction process. Therefore, this review critically assesses current methodologies, identifies and expands existing gaps, and suggests directions for future research to enhance the practical application of this technology in the food sector. Addressing various limitations associated with this method, this technology can offer cost-effective, portable solutions for food quality monitoring and safety assurance, paving the way for affordable, accessible, real-time, non-destructive inspection in both industrial and consumer applications.