Cargando...
Transformations for non-destructive evaluation of brix in mango by reflectance spectroscopy and machine learning
Author(s)
Ernesto Paiva-Peredo
Universidad Tecnológica del Perú
Diego Gonzales-Rodriguez
Universidad Tecnológica del Perú
W. T. Herrera
Universidad Tecnológica del Perú
Universidad Tecnológica del Perú
Universidad Tecnológica del Perú
Christian Ovalle
Universidad Tecnológica del Perú
Date Issued
14 de noviembre de 2023
Type
Article
Volume
14
Issue
1
Series/Report No.
International Journal of Power Electronics and Drive Systems/International Journal of Electrical and Computer Engineering
Start Page
532
End Page
532
ISSN
2722-256X
2722-2578
Abstract
Mango is a very popular climacteric fruit in America and Europe. Among the internal properties of the mango, total soluble solids (TSS) are an adequate indicator to estimate the quality of mango, however, the measurement of this indicator requires destructive tests. Several research have addressed similar issues; they have made use of pre-processing transformations without making it clear which of them is statistically better. Here, we created a new spectral database to build machine learning (ML) models. We analyzed a total of 18 principal component regression (PCR) models and 18 partial least squared regression (PLSR) models, where 4 types of transformations, 3 different feature extractors, and 3 different pre-processing techniques are combined. The research proposes a double cross validation (CV) both to determine the optimal number of components and to obtain the final metrics. The best model had a root mean square error (RMSE) of 1.1382 °Brix and a RMSE on the transformed scale of 0.5140. The best model used 4 components, used y2 transformation, reflectance R as the independent variable and MSC as a pre-processing technique.
Keywords