Análise e classificação da qualidade do linter e do óleo de algodão utilizando técnicas de visão computacional

Abstract

The classification methods applied to industrial processes are in constant enhancement and one of its factors is the technological alignment in industrial plant. However, many technologies are not economically practicable for factory floor implantation. As an example the methods utilized on feathers and cotton short fibers (linter) analysis which use applied technology and are only available in a few Brazilian analysis laboratories that comply with national quality standards, leading to higher production costs and longer periods to make the product available on market. Linter is a cotton’s byproduct, raw material of wide applicability, used in the manufacture of currency paper, hydrophilic cotton, printed circuit boards and surgical fabrics. Its classification importance comes from the appreciation of product quality in the internal and external market, which present different standards, leading to a raw material production with different quality characteristics. Another byproduct that requires laboratory examination during its manufacture is the cotton oil, characterized by intense coloring in its raw state, and, when it is used for alimentary purposes, goes through a lightening procedure and acidity reduction, which can be carried out by the refining or semi-refining processes. This study analyzes the semi-refining process, in which are analyzed the soaps concentrations, oil’s quality and stability, verifying at the end of the process the carotenoids absence and determining if it is suitable for consumption. Considering the significance of the linter categorization and the cotton oil semi-refining technique, this study proposes the semi-automatic analysis methodologies development for both classification methods, providing readiness in the delivery of products to the consumer market, obtaining reliable data and alignment between production and quality sectors. Thus, two methodologies are proposed using digital image processing and pattern recognition. In the linter classification, texture descriptors were applied to obtain characteristics and different classifiers were compared. In the classification of the semi-refining stages, different color models were used to extract characteristics, which were grouped by classifiers. With this study, promising results were obtained both for the classification of cotton oil semi-refining process and linter.