EXPLORATORY ANALYSIS OF RICE PROTEIN-ENRICHED CORN MEAL USING HYPERSPECTRAL IMAGING AND PRINCIPAL COMPONENT ANALYSIS

With the current population growth, there is also a significant increase in the demand for food products, which makes food security an increasingly urgent global concern. Several researchers defend the enrichment of corn flour with vegetable proteins, considering the low protein content inherent to the product, which contributes to protein-energy malnutrition in underdeveloped countries where corn is a staple food. This scenario imposes great challenges on inspection agencies and the food industry, especially in the quality control of these products. Therefore, this work aimed to develop a methodology for exploratory analysis of Hyperspectral Imaging data from samples of corn flour enriched with rice protein, using the Principal Component Analysis (PCA) method.

For this, 25 samples were prepared by adding rice protein to corn flour in a concentration ranging from 1.00 to 25.00% and the respective images were acquired with a Hyperspectral Camera. Image backgrounds were removed using the mask, in the MATLAB environment with Hyper-Tools Software, version 3. The data were pre-processed using First Derivative and SNV (Standard Normal Variate) to reduce the effects of scattering and texture differences between the samples. The results of the PCA analysis showed that PC1 accounted for most of the variance strongly correlated with protein content. A continuous gradient of the distribution of the points along the main axis of the samples divided into five groups of five was also observed, corresponding to increments of 5%. This ordering indicates that the model was able to capture the systematic and linear variation associated with the increase in protein, demonstrating good sensitivity of the method, that is, a clear gradual trend of separation among the groups.
The application of the K-Means algorithm to the reduced projections confirmed the gradual behavior identified by the PCA, as observed by: a clustering of the samples into five blocks corresponding to the concentration ranges (1–5%, 6–10%, 11–15%, 16–20%, and 21–25%); a coherence between the increase in concentration and the spatial position of the clusters, which validates the linearity of the model; and a relatively small dispersion within each cluster, indicating good reproducibility among samples of the same range.
From the analyses performed, it is concluded that the variations captured by PCA and confirmed by K-Means reflect the compositional change of the flour. The continuous nature of the separation indicates that the method can also be used for quantitative calibration.