Detection of kiwifruit low-force early bruise by combining surface textural parameters and near-infrared hyperspectral imaging

This study evaluates the feasibility of detecting and monitoring occurrence and progression of low-force mechanical damage in kiwifruits ( Actinidia deliciosa ) using Near-Infrared Hyperspectral Imaging combined with surface texture analysis and multivariate classification. Previous studies have applied HSI for bruise detection but often used unrealistically high impact forces, limiting relevance to real-world handling conditions. ‘Hayward’ kiwifruits were subjected to low intensity-controlled impact forces of very low intensity (23.4 N and 46.8 N) to simulate realistic damage and stored at 25 ± 1 °C for five days. Hyperspectral data were acquired using a short-wave infrared camera (960–2500 nm) at different sampling times. Principal Component Analysis (PCA) was applied to reduce data dimensionality, and surface texture features extracted from PCA score images accounted for fruit shape and surface characteristics. A Partial Least Squares – Discriminant Analysis (PLS-DA) model classified damaged versus healthy tissue, achieving an overall accuracy of 94.7% across hyperspectral images. Changes in damaged pixels allowed an indirect evaluation of progressive tissue degradation. showed a high detection of damaged (invisible to naked eye) and healthy tissue, demonstrating the efficiency of the methodology in preventing fruits affected by realistic damage from entering the fresh market while facilitating their use in alternative applications. • Realistic impact forces were applied to simulate transport and handling conditions. • The method detects low-force mechanical damage before bruises are visible. • PLS-DA achieved 94.7% accuracy in detecting mechanical damage. • The proposed method prevents the sale of compromised fruit, reducing food waste.