Potato quality assessment by monitoring the acrylamide precursors using reflection spectroscopy and machine learning

Acrylamide formation is nowadays one of the major concerns of the potato-processing agriculture industry. We investigate the use of broadband reflection spectroscopy (400-1700 nm), in combination with machine learning, to optically classify raw potatoes inducing different levels of acrylamide after frying, covering concentrations between 200 ppb and 2000 ppb. Using the full spectral range, we obtain a correct classification of a dataset of 200 samples and using 10-fold cross-validation, while applying Linear Discriminant Analysis and Extreme Learning Machine. To reduce the amount of data and increase processing speeds, a sequential feature selection search was performed to identify the critical wavelengths (450 nm, 488 nm, 504 nm, 783 nm, 808 nm, 1310 nm, 1319 nm and 1342 nm) that enable classification performances exceeding 92% when applying Linear Discriminant Analysis. We therefore demonstrate a non-destructive identification of the potatoes unsuited for French fries production, enabling to increase food safety, while limiting food waste.

Automated summary

The study utilizes broadband reflection spectroscopy and machine learning to non-destructively classify raw potatoes, identifying those unsuitable for French fries production, thus enhancing food safety and reducing food waste.