Advancing Maize Seed Vigor Assessment Through Multispectral Imaging: Highlighting the Relevance of Ultraviolet Optical Markers

The demand for rapid, objective, and non-destructive methods to evaluate maize seed quality has grown as conventional tests often lack the sensitivity to detect subtle physiological or structural differences. To date, the use of multispectral markers (an emerging non-destructive technique) for diagnosing maize seed vigor remains poorly investigated. This study aimed to develop and validate machine learning models for classifying maize seed lots using multispectral imaging as a reliable optical marker for distinguishing seed vigor. Multispectral images from 19 wavelengths (365–970 nm) were collected for commercial seed lots of four maize hybrids and compared with physiological performance (normal seedlings and vigor), anthocyanin accumulation, and X-ray analyses. X-ray and multispectral imaging revealed clear patterns associated with reduced seed vigor. Random Forest analysis highlighted the 365 nm UV band as the most informative wavelength for discrimination, effectively separating high- and low-vigor lots. Quadratic discriminant analysis (QDA) and support vector machine (SVM) models were trained using the full multispectral range to enhance classification precision. QDA achieved the highest overall accuracy and showed more consistent performance across hybrids. Overall, the findings demonstrate that multispectral imaging, particularly through UV-based markers, provides a rapid, accurate, and non-invasive alternative for seed vigor assessment. This approach improves real-time decision-making in maize seed production systems.