Comparative analysis of methods and model prediction performance evaluation for continuous online non-invasive quality assessment during drying of apples from two cultivars

Under a Creative Commons license
open access

Abstract

Quality attributes such as moisture content, colour parameters and shrinkage of apples change undesirably during the drying process. Drying is a highly dynamic process, thus, an effective optimisation in terms of product quality and process performance requires continuous non-invasive measurement of the parameters in question. In this study, visual to near infra-red hyperspectral imaging was used in combination with the respective gold standard methods to estimate moisture ratio, CIELab chromaticity, browning index, shrinkage, and rehydration ratio of apple slices during the hot air-drying process. Two varieties (cv. Elstar and Golden delicious) of apples at three slice thicknesses (2, 3, and 4 mm) were dried at 60 °C and 70 °C. Prediction models for the space-averaged spectral reflectance curves were built using the partial least square regression method and including both varieties. The performance of moisture ratio prediction was excellent (adj R2 = 0.94, RMSEP = 0.076) and the Variable Importance in the Projection value cut off above 0.8 at 970 nm and L* at 685 nm. Similarly, partial least square regression modelling showed a good prediction for a*, b* value, BI, shrinkage and acceptable prediction for L* and RR. The model performance was robust to the system settings irrespective of slice thickness, drying temperature and apple variety. Additionally, method comparisons using Bland-Altman, Bablok, and Deming regression were performed. The results confirm that the compared destructive (laboratory gold standard) and non-destructive hyperspectral methods can be interchangeably used within the limit of agreement (±1.96 standard deviations) and precision for determination of the MR, CIELAB chromaticity and BI, shrinkage, and rehydration ratio. Therefore, these results confirm that hyperspectral imaging system can be used in online monitoring of the apples during the drying process, and thus, in the optimisation of product and process performance quality attributes.

Abbreviations

adj R2
Adjusted coefficient of determination
AOAC
Association of Analytical Communities
BI
Browning index
CI
Confidence interval
CIELab
Chromacity
F& V
Fruit and vegetable
HSI
Hyperspectral imaging
LoA
Limits of agreement
LV
Latent variables
MC
Moisture content
MR
Moisture ratio
NIPALS
Nonlinear Iterative Partial Least Squares
NIR
Near infra-red
PCA
Principal component Analysis
PLS
Partial least square
PRESS
Predicted residual error sum of square
RMSE
Root mean square error
RR
Rehydration ratio
S
Shrinkage
SD
Standard deviation
VIP
Variable Importance in projection
Vis/NIR
Visible to near infra-red
WHO
World Health Organization

Keywords

PLS
Method comparison
Variable Importance in projection (VIP)
Vis/NIR hyperspectral imaging system
Dried apple slices
View Abstract