Rapid assessment of lychee and mango fruit quality using hyperspectral imaging
Naila Kanwal, Wiebke Kämper, Michael B. Farrar, Mahshid Tootoonchy, Clayton Lynch, Joel Nichols, Helen M. Wallace, Stephen J. Trueman, Shahla Hosseini Bai
Abstract
Rapid quality assessment of fruit is important to ensure consistent-quality production and supply. This study explored hyperspectral imaging (HSI) as a method to predict °Brix, acidity, and mineral nutrient concentrations using skin or flesh images of two lychee and two mango cultivars. Partial least squares regression (PLSR) models were developed using each cultivar. Spectral data across two cultivars were then pooled and these models compared with the models developed using individual cultivars. Artificial neural network (ANN) and support vector machine regression (SVMR) models were also developed for predicting Brix, acidity and Brix/acid ratio. Both the skin and flesh images were useful for developing PLSR models that predicted Brix and the Ca, Cu, Fe and Mn concentrations of lychee and mango flesh, with R 2 from 0.50 to 0.89. Pooled-cultivar datasets were useful for developing PLSR models that predicted Brix of lychee flesh, and Brix, acidity and Brix/acid ratio of mango flesh, with R 2 ≥ 0.60. The prediction accuracies were improved using ANN to estimate Brix and acidity of lychee flesh, while the prediction accuracies were improved using both ANN and SVMR to estimate Brix, acidity and Brix/acid ratio of mango flesh. The results demonstrate that skin images can be used for non-destructive assessment, and that HSI can predict fruit quality even among mixed-cultivar consignments. Advanced machine learning techniques further improve the prediction capacity. HSI provides a rapid method for predicting flesh quality of lychee and mango fruit, facilitating the timely scheduling of harvesting and allowing the grading of fruit into consistent-quality batches.