Aplicação do algoritmo KNN para classificação de imagens de cristalização de açúcar
Abstract
The increase in global competitiveness in the sugar and ethanol industry poses challenges to companies in the sector, requiring modernization and precise control of industrial processes in sugar manufacturing. Automation and the use of artificial intelligence allow the integration of analysis and control of process variables. In this context, computer vision has proven to be a tool to assist procedures. With the use of machine learning algorithms, visual classifications can be performed automatically. This work addresses the application of the k-nearest neighbors algorithm to classify supersaturation zones from images of sugar crystals. The automation of this task contributes to quality control and the reduction of human error, promoting advancement in this sector of great economic and social importance. The methodology used image processing techniques, such as histogram equalization and median filter. Edge detection methods and morphological operations were applied to highlight attributes. The extracted features were applied to the k-nearest neighbors algorithm, which was trained and validated with cross-validation and different k values. The model achieved a maximum accuracy of 83.1% with a k value of 5, showing good performance in distinguishing the different crystallization zones. Despite this, there was overlapping of samples in the border regions of the classes, which suggests the need for new approaches to separate the zones.
Collections
The following license files are associated with this item: