Physical Classification of Soybean Grains Based on Physicochemical Characterization Using Near-Infrared Spectroscopy

The study aimed to determine the physical and physicochemical properties of soybean grains using NIR spectroscopy coupled with multivariate data analysis. The experiment was carried out in two stages: first, individual characterization of defects and healthy grains; then, analyses of samples classified into different types (type I, type II, basic standard, and out of type). The centesimal composition of the grains (crude protein, lipids, water content, crude fiber, starch, and ash) was determined by NIR spectroscopy, and the data were analyzed by ANOVA, Scott-Knott test, principal component analysis (PCA), k-means clustering, and Pearson correlation. The results showed significant variations between defects and commercial types in all the variables evaluated (p < 0.05), with an emphasis on germinated grains (higher protein content) and broken grains (higher fiber content). The PCA explained 66.6% of the total variance in the defect sets and 52.2% of the types, allowing the formation of groups defined by the clustering algorithms. Pearson correlations indicated important interactions between the chemical variables, such as the negative correlation between protein and crude fiber (r = −0.73) and between lipids and water content (r = −0.66). It is concluded that the NIR method combined with multivariate modeling allows for the rapid assessment of soybean grain quality in real time, optimizing, reducing waste in, and increasing the efficiency of post-harvest processes.