This study investigates the application of hyperspectral imaging (HSI) technology combined with machine learning for the authentication and geographical origin determination of Italian wheat flours and grains, as part of the PNRR project Metrofood-IT. Using a hyperspectral camera system operating in the short-wave infrared range (900–1700 nm), we analysed wheat grain and flour samples from various Italian regions and of different varieties to identify distinctive spectral features that could discriminate their geographical origins. The methodology involved non-destructive hyperspectral imaging of the samples, followed by comprehensive data pre-processing and normalization procedures. Advanced chemometric techniques and machine learning algorithms were employed to analyse the spectral data and extract characteristic features for classification. In particular, the classification task focused on predicting the geographical origin zone of the samples – namely, North, Center, or South of Italy. This approach combines the non-destructive nature of hyperspectral imaging with the analytical power of machine learning, providing a robust method for flour authentication and traceability. Among the tested models, the XGBoost classifier achieved the best overall performance, with an average accuracy of 47.1%. Notably, the Central Italy samples were classified with the highest precision, achieving a class-specific accuracy of 72%, indicating a stronger spectral distinctiveness for this zone. These results demonstrate the potential of hyperspectral imaging as a reliable tool for determining the geographical origin of Italian flours, contributing to the development of rapid and accurate authentication methods for cereal products.