Desenvolvimento de um sistema para leitura de códigos de etiquetas de caseína em queijos minas artesanal utilizando técnicas de visão computacional e inteligência artificial

This work explores the use of computer vision techniques for the detection and reading of characters present on casein tags applied to Queijo Minas Artesanal, using images as input. Casein tags are essential elements for quality control and traceability within the production chain. The study proposed the development of a system composed of machine learning models capable of automatically locating the tags and recognizing the characters printed on them. Concepts of deep learning and convolutional neural networks were applied to a dataset of images of Canastra cheese. The detection model, based on the YOLOv8 architecture, achieved an average mAP@0.5 of 91.9% in the detection tasks. The OCR system, based on PaddleOCR, obtained an average accuracy of 81.8% and a mean Character Error Rate (CER) of 18.2%, demonstrating the technical feasibility of automating the registration process. The results indicate that the proposed solution contributes to the efficiency and reliability of the certification process of artisanal production in the Serra da Canastra region.