Detecção de Patologias em Obras Civis Utilizando Redes Neurais Convolucionais

Abstract

This work proposes a system for automated crack detection in civil structures using the YOLO model, a convolutional neural network known for its real-time accuracy and speed. Transfer learning and fine-tuning techniques were applied to a dataset containing 1,398 annotated images of the “crack” class. The results demonstrated strong performance, achieving a Recall@0.50 of 0.95 and mAP@0.50 of 0.90, indicating high sensitivity and robustness. The solution shows potential for deployment on drones or other devices, offering a scalable and low-cost alternative to traditional visual inspections. However, challenges such as generalization to other pathologies and hardware limitations in the field still need to be addressed. This study represents a significant advancement in the use of artificial intelligence for structural integrity monitoring, contributing to increased safety and durability of buildings.