Optical fingerprinting of cereal grains for predicting wheat flour extraction rates using machine learning models

The global demand for high-quality food exposes a significant bottleneck in cereal processing: inadequate grain screening before milling, leading to inconsistent flour quality. Seedsight is a disruptive screening platform that combines high-resolution optical analysis with machine learning to characterize wheat grain mixtures. The extraction of biomolecular and biophysical data from milled wheat is achieved by means of dynamic light scattering and a near-infrared laser. Predictive algorithms are then utilized to estimate industrial performance indicators, including extraction yield. The integration of optical-signal descriptors with conventional process metrics resulted in promising results in pre-milling prediction accuracy (R2 = 0.7035, Cross-Validation, R2 = 0.6832, Test), attaining industrial standards with a root mean squared error of 0.0039 (test set), representing an improvement over current standard practice. This approach demonstrates the potential of the Seedsight platform in reducing the environmental footprint while leading to more consistent, efficient and sustainable flour production.