Predicción del contenido de ácido ascórbico de aguaymanto (Physalis peruviana L.) usando un espectrómetro NIR portátil y quimiometría
No Thumbnail Available
Date
2024
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Nacional de Trujillo
Abstract
Los productos alimenticios frescos ofrecidos al mercado son evaluados por los consumidores en relación al sabor, textura, apariencia y valor nutricional. En consecuencia, son valorados de acuerdo a ello. El contenido de ácido ascórbico o vitamina C, es parte de los indicadores de calidad en gran variedad de frutas, entre ellas, el aguaymanto. Los métodos tradicionales con los que se cuantifica este nutriente demandan reactivos, son limitados a un número controlado de muestras y requieren su destrucción. Por esta razón, la presente investigación tuvo como objetivo postular un método no destructivo, al evaluar el nivel de predicción del contenido de ácido ascórbico en aguaymanto (Physalis peruviana L.) a partir del uso de un espectrómetro NIR portátil, en el rango de 900 a 1700 nm, junto con quimiometría. Se recolectó aguaymanto de dos procedencias: Bambamarca y Huaraz, y se clasificó cada una, en cuatro estados de madurez: verde, pintón, maduro y sobremaduro. Se adquirió la data espectral escaneando, con un espectrómetro NIR portátil, diversos puntos de la superficie del fruto, previa formación de muestras. Un total de 50 muestras fueron evaluadas por estado de madurez en cada procedencia, resultando en total 400 curvas espectrales, que fueron divididas en 70% para calibración y 30% para validación. Se compararon regresión de mínimos cuadrados parciales (PLSR) y regresión por máquinas de vectores soporte (SVMR), como herramientas de regresión y algoritmo genético (GA), interval partial least square regression (iPLS) y recursive weighted partial least square regression (rPLS), como algoritmos de selección de variables. El algoritmo GA, con ancho de ventana de 15, fue el que permitió obtener los mejores resultados, mejorando los modelos full de PLSR y SVMR. El modelo SVMR con GA obtuvo la mejor predicción: con 90 variables, R2C=0.96, R2P=0.92, RMSEC=1.8, RMSEP=2.52 y RPD=3.73. Los demás modelos consiguieron R2 superiores a 0.85. Los indicadores obtenidos demuestran que los modelos construidos son aptos para predecir el contenido de ácido ascórbico en aguaymanto y colocan al empleo del espectrómetro NIR portátil junto con quimiometría, como una técnica de alto potencial en la cuantificación de este nutriente.
ABSTRACT Fresh food products offered to the market are evaluated by consumers in relation to taste, texture, appearance and nutritional value. Consequently, they are valued for that. The ascorbic acid or vitamin C content, is one of the quality indicators in a wide range of fruits, including cape gooseberry. Traditional methods of measuring this nutrient require reagents, are limited to a controlled samples number and require their destruction. For this reason, the present research aimed to postulate a non-destructive method to evaluate the predictive level of ascorbic acid content in cape gooseberry (Physalis peruviana L.) by using a portable NIR spectrometer in the range of 900 to 1700 nm, together with chemometrics. Cape gooseberry was collected from two locations: Bambamarca and Huaraz, and each was classified into four stages of maturity: unripe, half-ripe, ripe and overripe. Spectral data were acquired by scanning, with a portable NIR spectrometer, various points on the surface of the fruit, after sample formation. A total of 50 samples were evaluated per maturity stage in each origin, resulting in 400 spectral curves, which were divided into 70% for calibration and 30% for validation. Partial least square regression (PLSR) and support vector machine regression (SVMR) were compared as regression tools and genetic algoritm (GA), interval partial least squares regression (iPLS) and recursive weighted partial least squares regression (rPLS) as variable selection algorithms. The GA algorithm, with a window width of 15, provided the best results, improving the full PLSR and SVMR models. The SVMR model with GA obtained the best prediction: with 90 variables, R2C=0.96, R2P=0.92, RMSEC=1.8, RMSEP=2.52 and RPD=3.73. The other models achieved R2 above than 0.85. The indicators obtained demonstrate that the models constructed are suitable for predicting the ascorbic acid content in cape gooseberry, and place the use of the portable NIR spectrometer together with chemometrics as a technique with high potential in the quantification of this nutrient.
ABSTRACT Fresh food products offered to the market are evaluated by consumers in relation to taste, texture, appearance and nutritional value. Consequently, they are valued for that. The ascorbic acid or vitamin C content, is one of the quality indicators in a wide range of fruits, including cape gooseberry. Traditional methods of measuring this nutrient require reagents, are limited to a controlled samples number and require their destruction. For this reason, the present research aimed to postulate a non-destructive method to evaluate the predictive level of ascorbic acid content in cape gooseberry (Physalis peruviana L.) by using a portable NIR spectrometer in the range of 900 to 1700 nm, together with chemometrics. Cape gooseberry was collected from two locations: Bambamarca and Huaraz, and each was classified into four stages of maturity: unripe, half-ripe, ripe and overripe. Spectral data were acquired by scanning, with a portable NIR spectrometer, various points on the surface of the fruit, after sample formation. A total of 50 samples were evaluated per maturity stage in each origin, resulting in 400 spectral curves, which were divided into 70% for calibration and 30% for validation. Partial least square regression (PLSR) and support vector machine regression (SVMR) were compared as regression tools and genetic algoritm (GA), interval partial least squares regression (iPLS) and recursive weighted partial least squares regression (rPLS) as variable selection algorithms. The GA algorithm, with a window width of 15, provided the best results, improving the full PLSR and SVMR models. The SVMR model with GA obtained the best prediction: with 90 variables, R2C=0.96, R2P=0.92, RMSEC=1.8, RMSEP=2.52 and RPD=3.73. The other models achieved R2 above than 0.85. The indicators obtained demonstrate that the models constructed are suitable for predicting the ascorbic acid content in cape gooseberry, and place the use of the portable NIR spectrometer together with chemometrics as a technique with high potential in the quantification of this nutrient.
Description
Keywords
Uchuva, Machine learning, Berries, PLS, SVM, Vitamina C