Robust NIRS models for non-destructive prediction of mango internal quality
Near infrared spectroscopy (NIRS) is increasingly being used with success in fruit supply chains for the non-destructive assessment of internal fruit quality. However, the prediction performance of NIRS was reported to be sensitive to changes in the pre- and post-harvest factors involved in fruit quality variations. This study attempted to establish robust NIRS models to predict mango internal quality, regardless of the conditions encountered by fruits on the tree and after harvest. A database involving mangoes (n = 250) from different production years and orchards and which were harvested at different maturity stages and stored at different temperatures for various periods was used to characterize mango quality using NIRS. The large variations measured in mango total soluble solid (TSS) content, dry matter, flesh color and acidity ensured that the dataset considered in this study covered differences in quality levels that can be encountered for the mango cultivar studied. Variable selection procedures and pre-processing techniques were used to improve the performance of the Partial Least Squares Regression (PLSR) models that accurately predicted the mango quality traits studied, regardless of the origins and storage conditions. The root mean square errors in prediction (RMSEP) were 0.6°Brix, 1.4%, 5.9 meq 100 g FM−1 and 3.16°Brix for the TSS content, dry matter content, titratable acidity and the hue angle of flesh color, respectively. The results presented in this study confirmed that robust NIRS models can be developed to predict mango quality by considering the pre- and post-harvest factors involved in fruit quality. Further studies should assess the use of NIR spectra on fruits to address internal quality variation issues, as well as the ability of NIRS to predict fruit shelf life and eating quality at harvest.
| Summary: | Near infrared spectroscopy (NIRS) is increasingly being used with success in fruit supply chains for the non-destructive assessment of internal fruit quality. However, the prediction performance of NIRS was reported to be sensitive to changes in the pre- and post-harvest factors involved in fruit quality variations. This study attempted to establish robust NIRS models to predict mango internal quality, regardless of the conditions encountered by fruits on the tree and after harvest. A database involving mangoes (n = 250) from different production years and orchards and which were harvested at different maturity stages and stored at different temperatures for various periods was used to characterize mango quality using NIRS. The large variations measured in mango total soluble solid (TSS) content, dry matter, flesh color and acidity ensured that the dataset considered in this study covered differences in quality levels that can be encountered for the mango cultivar studied. Variable selection procedures and pre-processing techniques were used to improve the performance of the Partial Least Squares Regression (PLSR) models that accurately predicted the mango quality traits studied, regardless of the origins and storage conditions. The root mean square errors in prediction (RMSEP) were 0.6°Brix, 1.4%, 5.9 meq 100 g FM−1 and 3.16°Brix for the TSS content, dry matter content, titratable acidity and the hue angle of flesh color, respectively. The results presented in this study confirmed that robust NIRS models can be developed to predict mango quality by considering the pre- and post-harvest factors involved in fruit quality. Further studies should assess the use of NIR spectra on fruits to address internal quality variation issues, as well as the ability of NIRS to predict fruit shelf life and eating quality at harvest. |
|---|