Classification modeling based on surface porosity for the grading of natural cork stoppers for quality wines

tThe natural cork stoppers are commercially graded into quality classes according with the homogeneity of theexternal surface. The underlying criteria for this classification are subjective without quantified criteria and standardsdefined by cork industry or consumers. Image analysis was applied to premium, good and standard quality classes tocharacterize the surface of the cork stoppers and stepwise discriminant analysis (SDA) was used to build predictiveclassification models. The final goal is to analyze the contribution of each porosity feature and propose an algorithmfor cork stoppers quality class classification. This study provides the knowledge based on a large sampling to anaccurate grading of natural cork stoppers.In average all the models presented accuracy in relation to the commercial classification over 68% with a highermismatch in the mid-quality range. Color showed an important discriminating power, increasing the accuracy in10%. The main discriminant features were porosity coefficient and color variables, calculated for the lateral surface. Aquality classification algorithm was presented based on a simplified model with an accuracy of 75%. The classificationbased on color vision systems can ensure improved quality class uniformity and a higher transparency in trade