Hyperspectral imaging and machine learning to identify epicuticular wax loss in Masena Blueberries for post-harvest freshness

Abstract

Blueberries harvested using mechanical harvesting techniques, like over-the-row (OTR) harvesters, are not suitable for the fresh market due to high damage rates. Assisted harvesting (AH) techniques using hand-held shakers offer a new approach which shows promise in increasing harvesting rates without compromising fruit quality. Epicuticular wax is one of the fruit qualities needing to be preserved during harvesting. This paper investigates the effectiveness of hyperspectral imaging techniques and machine learning to construct a model to identify epicuticular wax rapidly and non-destructively. The best performing model produced was a linear SVM, and had an F1-score of 98.6%. Additionally, this model was used to show that blueberries harvested using hand-held shakers retain more epicuticular wax than traditional hand-harvesting (HH) techniques, a result that shows promise of potentially increased fruit quality using automated methods.