Emerging Non-destructive Technologies for Indirect Assessment and Prediction of Physical and Physicochemical Quality of Corn Grains as an Alternative to Physical Classification

Corn is one of the main crops in global agribusiness, and its widespread use demands faster and more efficient classification methods for assessing kernel quality. Traditional physical classification, based on visual criteria, has limitations, which reinforces the need to use non-destructive technologies combined with machine learning (ML) models. This study aimed to evaluate and predict the physicochemical quality of corn kernels classified into different classifications and groups, integrating hyperspectral and physicochemical data through optical methods, with a focus on modernizing classification processes. Whole kernels of different cultivars were manually classified into commercial categories (type I, II, III, and out-of-type) and grouped into soft and flint corn. The samples were analyzed using NIRS to determine their physicochemical parameters, and then evaluated by a hyperspectral sensor (Vis–NIR-SWIR, 350–2500 nm). The generated data were integrated to explore relationships between quality and spectral behavior, applying six ML models to predict attributes, with performance evaluated by cross-validation using correlation coefficient and mean absolute error. The proximate composition revealed differences between the groups and classifications, with a higher starch content in type I, while crude protein is more associated with the group. Slight variations were observed in lipids, fiber, and ash. Spectral analysis revealed that the physical structure of the kernel has a direct influence on the reflectance patterns, enabling the differentiation of various groups and classifications. The ML models performed well in predicting predominant variables in their composition, such as moisture and starch content. Thus, the integration between physicochemical analysis (NIRS), hyperspectral data (sensor), and ML has the potential to modernize corn kernel classification, with essential applications in post-harvest quality assessment, allowing the industry to monitor and control attributes quickly and non-destructively.