Desenvolvimento de um sistema de classificação de peças em ambiente industrial utilizando visão computacional e a arquitetura YOLO.

Abstract

This work aims to provide an alternative for part detection, replacing conventio-nal sensors and, in some situations, offering a viable solution when traditional sensors are not efficient. To achieve this goal, the study was developed based on the Modular Production System, which simulates an industrial environment on a laboratory scale. The pre-trained neural network model YOLO (You Only Look Once) was used to identify the parts, adjusting it specifically for this task. The implementation of the algorithm was carried out in Python, due to its wide range of machine learning libraries and ease of integration with computer vision tools. The deep convolutional neural network demonstrated an accuracy of 98.9%, recall of 100% and F1-score of 99.3%, indicating a promising performance of the model.