71820

MEASURING COLOR OF ORANGE FRUIT BY HYPERSPECTRAL IMAGING AND MULTIVARIATE REGRESSION

Favorite this paper

Color is considered as one of the most important quality attributes for fresh fruit industry. This attribute is traditionally evaluated by human inspection or measured using different instruments, which can be subject to human errors and be time-consuming. In this work, a method of measuring color of orange fruit based on Hyperspectral Imaging (HSI) and multivariate regression was developed. For this, a total of 74 oranges were randomly separated into calibration and validation groups of 58 and 16 samples respectively. Then reference color of samples was measured in L*a*b* color system with a CIELab colorimeter and the fruits were scanned by a HSI system of 400 -1000 nm in reflectance mode to obtain their spectral signatures. Partial Least Squares Regression (PLSR) was used to build models that correlate the spectral signatures with their L*, a*, and b* components. The performance of models in the calibration and validation was evaluated based on determination coefficients (R2) and Standard Error (SE). The PLSR models for predicting L*, a* and b* components were reasonably good in the calibration with R2 values of 88.1%, 94.1% and 87.5%, and SE of 1.3, 1.1, and 2.0, respectively; and also in the validation with R2 values of 85.1%, 93.1% and 84.9%, and SE of 1.4, 1.3, and 2.3, respectively. Analysis of regression coefficients of PLSR models allowed an effective selection of most influent wavelengths enabling the building of simplified models by PLSR or Multiple Linear Regression. The simplified models presented similar performance to the full PLSR models in terms of R2 and SE. The results demonstrated that HSI can be used to measure the color of orange fruit in a rapid, objective and contact-free way. This method could be integrated to others HSI methods of measurement of quality attributes for a complete classification of the sample.