Background: Coffee is a raw material of global interest. Due to its relevance, this work evaluated the performance of calibration models constructed from spectral data obtained using near-infrared spectroscopy (FT-NIR) to determine the pH values and acidity in coffee beans in a practical and non-destructive way. Partial least squares regression was used during the calibration and the cross-validation to optimize the number of latent variables. The predictive capacity of the spectral pre-processing methods was also accessed.
Results: The results obtained showed that the best methods of pre-processing were the first derivative for the pH variable and the standard normal variate for the acidity, which produced models with correlations of 0.78 and 0.92, ratios of prediction to deviation of 2.061 and 2.966 and biases of -0.00011 and -0.152 to test set validation, respectively. The average errors between predicted and experimental values were lower than 7%.
Conclusions: FT-NIR was successfully applied to predict properties related to the quality of coffee. The method was demonstrated to be a fast and non-destructive tool which allows the rapid inline evaluation of samples facilitating industrial and commercial processing. © 2020 Society of Chemical Industry.
Keywords: PLS regression; multivariate calibration; raw coffee; spectroscopy.
© 2020 Society of Chemical Industry.