Model for classifying the maturity of Biloxi blueberries using UVVIS-NIR hyperspectral images and chemometric modeling.

Abstract

The main characteristic to define quality conditions and marketing destinations for blueberries is their ripeness, which is directly related to the internal conditions of the fruit. This creates the need for a technique that allows for rapid and individual measurement of this product. The objective of this research was to use hyperspectral images to predict the ripeness of Viloxy blueberries using chemometric analysis. For this purpose, an HSI-VNIR system was used, which works by reflectance and consists of a hyperspectral camera with a range of 400-1100 nm. Nine prediction models based on machine learning systems were developed, with the SVM model showing the best average performance, reaching an accuracy of 0.97 and an F1_macro ≈ 0.97, followed by XGBoost (accuracy ≈ 0.96, F1_macro ≈ 0.96) and Random Forest (accuracy ≈ 0.95, F1_macro ≈ 0.95). These values demonstrate that the non-destructive technique of hyperspectral imaging can be implemented for the classification and prediction of the ripeness of blueberries, thereby generating a useful tool for deciding on the marketing destinations of these fruits.