Chemometrics, VIS-NIR-SWIR spectroscopy, and deep learning algorithms to classify and predict qualitative attributes of rice in grain processing and storage units.

The study aimed to investigate the relationship between physical quality classifications based on grain defects and the physicochemical attributes of milled rice using non-destructive techniques combined with machine learning algorithms as a method for determining grain quality in storage and processing units. Samples of white, black, red, and parboiled rice were analyzed using hyperspectral spectroscopy (350-2500 nm) and subjected to traditional and deep machine learning models, including Linear Regression, Support Vector Machine, Random Forest, Gradient Boosting, Convolutional Neural Networks, and Recurrent Neural Networks. Spectral and physicochemical data were explored using multivariate analysis. The findings indicated that Support Vector Machine, Random Forest, and Gradient Boosting models exhibited superior performance. Hyperspectral spectroscopy proved effective in differentiating rice types, enabling the selection of relevant bands for optimized sensors in the analysis of the rice. The use of non-destructive technologies integrated with machine learning is promising for industrial applications in rice quality control.