Proposal of automated computational method to support Virginia tobacco classification

Abstract

This article proposes an automatic method for classification of cured tobacco leaves. Typically this process is performed manually, allowing the occurrence of human errors. In addition, the existence of an automated comparative procedure, helping to perform the classification, can make this process faster and more transparent. In order to implement the method, non-invasive to the agricultural product, 250 samples of Virginia tobacco digital images in the RGB and HSV color models were analyzed. The validation of the method was carried out using partial least squares (PLS) and artificial neural network (ANN), presenting a qualitative and quantitative analysis of both tools. It has been verified that the PLS can be applied to this method, as it has a shorter computational time, better suiting a real-time process. It can be verified that the ANN obtained better prediction results. Both methods employed had better results when adopting the RGB color model, reaching coefficient of determinations of 68 and 96% for the PLS and ANN methods, respectively.