Evaluation of β-carotene content in apricots during the drying process using visual and near-infrared spectroscopy

Non-destructive applications for the detection of food quality, especially internal properties, are highly relevant for process control in the food industry. In this respect, visible and near-infrared spectroscopy (VIS and NIR) were evaluated and compared for their ability to predict β-carotene content in apricots. Two regions called VIS-NIR from 400–1000 nm and called NIR from 900–1700 nm regions were analyzed for prediction ability. In the paper two types of regression models that present the β-carotene content are used: partial least square regression (PLSR) and support vector machine regression (SVMR) developed in both spectrum ranges, with good results for coefficient of determination (R2) and standard errors of cross-validation (RMSECV). The best model performance for calibration and validation is obtained using SNV->SG pre-treatment with PLS and SVM regressions in the VIS-NIR range. The performance measurements are as follows: R2c=0.91 and RMSEC=20.1 for calibration and R2v=0.84 and RMSEV=27.1 for validation. On the other hand Support Vector Machine Regression (SVMR) has the following results: R2c=0.92 and RMSEC=19.2 for calibration and R2v=0.84 and RMSEV=26.7 for validation. When using second derivative Savitzky-Golay (SG’’) both models showed a poor performance. On the basis of the developed models the polynomial linear regression and equations are developed to evaluate the β-carotene content during the apricots drying process. The results show that the both ranges VIS-NIR and NIR can be used for non-destructive and reliable determination of β-carotene content in apricots using different kinds of predictive models. In this study the best model is obtained in the VIS-NIR range. The conclusion is that there is a work to be done for improving the prediction. The variance of the drying process model is still large. Despite the closeness between the reference values and the predicted ones, the error in estimating the concentration is still not satisfactory.