Dória Neto, Adrião DuarteOliveira, Thiago Theiry de2024-08-212024-08-212024-08-15OLIVEIRA, Thiago Theiry de. GANs (Generative Adversarial Network) na ampliação da base de dados para a detecção de defeitos em linhas de produção da indústria têxtil. 2024. 74 f. Trabalho de Conclusão de Curso (Graduação em Engenharia de Computação), Departamento de Engenharia da Computação, Universidade Federal do Rio Grande do Norte, Natal, 2024.https://repositorio.ufrn.br/handle/123456789/59623In the contemporary context of textile production, the efficient identification and correction of flaws in production lines have become priorities. This study proposes a computer vision and deep learning system using the YOLO (You Only Look Once) algorithm for real-time defect detection. Using the YOLO v8 Small version, the system detects irregularities in the fabric, such as incorrect stitching points. Tools like Roboflow are fundamental for image processing and dataset expansion. To enhance the dataset and improve defect identification, Generative Adversarial Networks (GANs) were employed, generating high-quality synthetic images, diversifying the dataset, and increasing the model’s generalization capacity. The results show remarkable performance with an average accuracy above 97%."(mAP@[0.5]). The model was trained with diversified data. The application performs real-time defect identification in any video simulating fabric passing through a conveyor belt, using the best weights from the YOLO neural network training imported to make the prediction. In summary, the adoption of the YOLO algorithm, along with the use of GANs, marks a significant advancement in flaw detection in textile production lines, simplifying and speeding up inspection, reducing operational costs, and improving the final product qualityAttribution-NonCommercial-NoDerivs 3.0 Brazilhttp://creativecommons.org/licenses/by-nc-nd/3.0/br/Inteligência artificialAprendizagem profundaYOLOGANbachelorThesisCNPQ::ENGENHARIAS