Nutrient Content Prediction and Geographical Origin Identification of Bananas by Combining Hyperspectral Imaging with Chemometrics

Abstract

The nutritional quality of bananas and their geographical origin authenticity are very important for trade. There is an urgent need for rapid, non-destructive testing to improve the origin and quality assurance for importers, distributors, and consumers. In this study, 99 banana samples from a range of producing countries were collected. Hyperspectral data were combined with chemometric methods to construct quantitative and qualitative models for bananas, predicting soluble solids content (SSC), potassium content (K), and country of origin. A second derivative analysis combined with competitive adaptive weighted sampling (CARS) and random frog jumping (RF) was selected as the best pre-treatment method for the prediction of SSC and K content, respectively. Partial least squares (PLS) models achieved R2p values of 0.8012 and 0.8606 for SSC and K content, respectively. Chinese domestic and imported bananas were classified with a prediction accuracy of 95.83% using partial least squares-discriminant analysis (PLS-DA) and an RF method that screened the spectral variables after a second pretreatment. These results showed that hyperspectral imaging technology could be effectively used to non-destructively predict the nutrient contents of bananas and identify their geographical origin. In the future, this technology can be applied to determine the nutritional quality composition and geographical origin of bananas from other countries.

Keywords

• banana
• hyperspectral imaging
• soluble solids content
• potassium
• origin discrimination
• model