Classification of Browning on Intact Table Grape Bunches Using Near-Infrared Spectroscopy Coupled With Partial Least Squares-Discriminant Analysis and Artificial Neural Networks

Andries J Daniels; Carlos Poblete-Echeverría; Hélène H Nieuwoudt; Nicolene Botha; Umezuruike Linus Opara

doi:10.3389/fpls.2021.768046

Classification of Browning on Intact Table Grape Bunches Using Near-Infrared Spectroscopy Coupled With Partial Least Squares-Discriminant Analysis and Artificial Neural Networks

Front Plant Sci. 2021 Oct 29:12:768046. doi: 10.3389/fpls.2021.768046. eCollection 2021.

Authors

Andries J Daniels^{1

2}, Carlos Poblete-Echeverría¹, Hélène H Nieuwoudt³, Nicolene Botha⁴, Umezuruike Linus Opara^{5

6}

Affiliations

¹ Department of Viticulture and Oenology, Faculty of AgriSciences, Stellenbosch University, Stellenbosch, South Africa.
² ARC Infruitec-Nietvoorbij, Stellenbosch, South Africa.
³ South African Grape and Wine Research Institute, University of Stellenbosch, Stellenbosch, South Africa.
⁴ Council for Scientific and Industrial Research, Modelling and Digital Science, Stellenbosch, South Africa.
⁵ SARChI Postharvest Technology Research Laboratory, Africa Institute for Postharvest Technology, Faculty of AgriSciences, Stellenbosch University, Stellenbosch, South Africa.
⁶ UNESCO International Centre for Biotechnology, Nsukka, Nigeria.

Abstract

Table grape browning is a complex physiological disorder that occurs during cold storage. There is a need to investigate novel and innovative ways to manage the problem that hampers the progressive and sustainable growth of table grape industries. Given the complex nature of the browning phenomenon, techniques such as near-infrared (NIR) spectroscopy can be utilized for the non-destructive classification of different browning phenotypes. In this study, NIR coupled with partial least squares discriminant analysis (PLS-DA) and artificial neural networks (ANN) were used to classify bunches as either clear or as having chocolate browning and friction browning based on the spectra obtained from intact 'Regal Seedless' table grape bunches that were cold-stored over different periods. Friction browning appears as circular spots close to the pedicel area that are formed when table grape berries move against each other, and chocolate browning appears as discoloration, which originates mostly from the stylar-end of the berry, although the whole berry may appear brown in severe instances. The evaluation of the models constructed using PLS-DA was done using the classification error rate (CER), specificity, and sensitivity and for the models constructed using ANN, the kappa score was used. The CER for chocolate browning (25%) was better than that of friction browning (46%) for weeks 3 and 4 for both class 0 (absence of browning) and class 1 (presence of browning). Both the specificity and sensitivity of class 0 and class 1 for friction browning were not as good as that of chocolate browning. With ANN, the kappa score was tested to classify table grape bunches as clear or having chocolate browning or friction browning and showed that chocolate browning could be classified with a strong agreement during weeks 3 and 4 and weeks 5 and 6 and that friction browning could be classified with a moderate agreement during weeks 3 and 4. These results open up new possibilities for the development of quality checks of packed table grape bunches before export. This has a significant impact on the table grape industry for it will now be possible to evaluate bunches non-destructively during packaging to determine the possibility of these browning types being present when reaching the export market.

Keywords: artificial neural networks; browning; contactless scanning; intact table grape bunches; near-infrared spectroscopy; partial least squares discriminant analysis.