Predicting frying times in recently fried potato fries using imaging spectroscopy and Partial Least Squares Discriminant Analysis (PLSDA)

Description

A new method was developed to classify frying times of recently fried potato fries using imaging spectroscopy and chemometrics. Ninety frozen potatoes, round cut, pre-fried in sunflower oil, were finally fried for 3 and 5 minutes, then cooled for 10 minutes and immediately measured by a spectral camera (Specim FX17, spectral range: 937.33 nm to 1718 nm). All images were converted to .mat format in MatLab using an inhouse developed function. Then, spectra were extracted from the images, from core and crust, using Hypertools version 3 in MatLab. Typical chemometrics analysis was performed in R version 4.2. Samples were distributed in two classes according to the frying time: class 1 (3 minutes) and class 2 (5 minutes). Three classification models were calibrated and validated: a) One model to classify 3 min from 5 min fries using 90 spectra from the cores, b) One model to classify 3 min from 5 min fries using 84 spectra from the crusts, c) One model to classify cores from crusts, using all measured spectra (not shown here). The best models were achieved using PLSDA, results can be seen in Table 1. Core: SNV, then 12 variables selected by CovSel, PLSDA with 8 latent variables (LV). Crust: Detrend, then 14 features selected by CovSel, PLSDA with 12 LV. Both models presented high accuracy of validation, 0.89 and 0.92 for core and crust, respectively. Sensitivity and specificity, were 1.0 and 0.81, respectively, for class 1 in core; and 0.83 and 0.95, respectively, for class 1 in the crust model. Thus, both models presented potential as a fast alternative methods to classify recently fried potato fries using imaging spectroscopy and chemometrics.

Period	5 Jun 2023 → 7 Jun 2023
Event title	SensorFINT Conference 2023- Annual Workshop of the Work Group Chemometrics & Quality Assurance
Event type	Conference/symposium