Combination of hyperspectral imaging and machine learning techniques for non-destructive quality evaluation of on-tree apple fruits

Abstract Hyperspectral imaging (HSI) has emerged as a promising nondestructive technique for fruit quality assessment. This study aimed to predict the soluble solids content (SSC) and firmness of apples ( Malus pumila Mill ) under outdoor field conditions using a push-broom hyperspectral camera (400–1000 nm). Data were collected over two growing seasons (2022-2023), from the immature stage to harvest, under naturally variable solar radiation and cloud cover. Preprocessed reflectance spectra were analyzed with machine learning and deep learning regression models to establish relationships with measured SSC and firmness. Among the tested models, random forest regression yielded the highest prediction accuracies (R 2 = 0.8177 for SSC and R 2 = 0.9002 for firmness), demonstrating the feasibility of applying HSI directly in orchard environments. These results provide a practical framework for advancing nondestructive quality monitoring of apples in real-world agricultural settings.