Kiwifruit under stress: unmasking hidden damage with hyperspectral imaging

The transition to Food Quality 4.0 represents change in how the agri-food industry ensures quality, safety and sustainability [1]. This concept exploits advanced, non-destructive technologies to meet the growing demand for efficient, real-time and data-driven quality assessment methods [2]. In this context, detecting the hidden effects of mechanical stress on fresh produce at an early stage is more than an economic issue, as it enables us to reduce food waste and optimise resource use. Kiwifruit, a globally traded product, is particularly vulnerable to mechanical damage, which often remains invisible to the naked eye yet affects sensory and nutritional attributes [3]. These challenges highlight the need for non-destructive approaches to unmask and monitor such damage throughout the supply chain. Mechanical damage was induced using a controlled setup, where a cylindrical metal weight (34.61 g) was dropped from a height of 1 m onto the base of each fruit. Three experimental conditions were established: undamaged samples served as controls, while other samples were subjected to a single or double weight drop to simulate different levels of mechanical stress. Hyperspectral data were acquired with a short-wave infrared (SWIR) camera (Specim, Spectral Imaging Ltd., Oulu, Finland) at different time points to follow the evolution of mechanical damage over time. The spectral data were processed and analysed with HYPER-Tools 3 software [4]. The analysis of kiwifruits presented specific challenges due to their physical and chemical properties, including the brown colour of the peel, its opacity, and high-water content [5]. These factors can interfere with hyperspectral imaging analysis, necessitating additional preprocessing steps. Techniques such as derivative transformations and scattering removal were applied to improve the quality of the spectral data. Principal Component Analysis (PCA) was then conducted on the complete dataset, consisting of 56 NIR-HSI images. Based on these analyses, relevant wavelengths were identified to guide the development of a simplified multichannel system for mechanical damage detection. Spectral analyses revealed clear patterns of moisture redistribution and tissue alteration, with spectral signatures evolving over time, particularly in the absorption band associated with water (around 1200 nm). These features were enhanced by applying a first derivative transformation prior to PCA. The exploratory analysis revealed that PC1 captured variations related to lighting conditions and fruit shape, whereas PC2 is associated with mechanical damage. To quantify the extent of the damage, a threshold was applied to the PC2 scores, allowing the identification of damaged areas and the calculation of the percentage of affected tissue over time. Damaged tissues were absent in the control samples throughout the storage period, whereas the percentage damage in damaged kiwi samples increased over time. The developed method has the potential not only to prevent damaged fruit from entering the fresh market, thus reducing waste and improving quality control, but also to facilitate alternative uses, contributing to a more sustainable and efficient food system.