Determinação dos parâmetros de qualidade da polpa da Spondias mombin utilizando espectroscopia de infravermelho próximo e algoritmo de seleção de variáveis

Abstract

Brazil occupies the third place in the production and commercialization of consumed tropical fruits, behind only China and India. In addition to the known fruits, there is also a wide variety of species that are little-known, even by Brazilians themselves, but which have great potential for commercialization in natura or pulp, for example. In terms of Brazilian regions, the Northeast has tropical fruits that are still little explored, among which the fruit of Spondias mombin L., known as cajá or taperebá in the North, stands out. In addition to the striking flavor of the fruit, cajá contains antioxidant substances, flavonoids, phenolic acids, vitamins A and C, anthocyanins and carotenoids in considerable amounts. However, when the fruit is processed, these substances can be lost, affecting total soluble solids, total acidity and pH, for example. In this sense, it is essential to ensure that the marketed product meets the characteristics that the consumer wants in terms of nutritional and functional properties. However, the recommended analyzes are expensive and time-consuming, which make it difficult to control the quality of the products, given that most of the production is carried out by small companies with local or regional marketing. Therefore, the objective of this work was to evaluate the quality parameters of the pulp of S. mombin by evaluating the titratable total acidity (TTA), pH and Total Soluble Solids (SST), using NIR Spectroscopy associated with Partial Least Squares Regression algorithms: iPLS , iSPA-PLS and FF-iPLSR in raw and preprocessed data. Data were preprocessed using SNV, MSC transformations and Savitzky-Golay derivation (with combinations of first and second derivatives, first and second degree polynomials and windows of 17, 21 and 25 points). NIR spectra were obtained using 36 samples in the spectral range from 910 to 1408 nm. The results were evaluated using the predictive capacity in terms of EJCR, RMSEC, RMSEP, Determination Coefficient for Calibration (R2cal) and Prediction (R2pred) and bias. The samples from the calibration (27) and prediction (9) sets were selected by the SPXy algorithm. In terms of the results obtained, for the determination of pH the three models obtained were with FF-iPLS. This was the only type of variable selection that worked for the determination of this property. Getting as values of RMSEC = 0.1330, R2cal = 0.7195, RMSEP = 0.2727, R2pred = 0.0242 (FF-iPLS raw data); RMSEC = 0.0204, R2cal = 0.1748, RMSEP = 0.1412, R2pred = 0.00862 (FF-iPLS SG21_1_1) and RMSEC =0.1909, R2cal = 0.3298, RMSEP = 0.1617, R2pred = 0, 0666 (SG21_2_2). For SST in terms of °brix, they had two suitable models and both used PLS, with values of RMSEC =0.5032, R2cal = 0.8159, RMSEP = 0.9322, R2pred = 0.9999 (model with SNV) and RMSEC =0.5043, R2cal = 0.8150, RMSEP = 0.9300, R2pred = 0.9999 (model with MSC). For acidity, only one model worked, and this one was also with FF-iPLS, having values of RMSEC =0.0072, R2cal = 0.3945, RMSEP = 0.0087, R2pred = 0.0167 (FF-iPLS with SG21_2_2). In this way, the models built with SNV and MSC pre-processing and with the FF-iPLS algorithm presented themselves as an interesting tool for predicting the physical-chemical properties of theft pulp, especially Spondias mombin.