Automatická detekce rzi na střechách budov

Abstract

This Master's thesis explores the application of Convolutional Neural Networks and computer vision methods for the automated detection of corrosion on building rooftops. The work includes a theoretical introduction followed by a literature review of the most widely used models, tools, and technologies for object detection in images. The training and testing datasets were constructed from orthophotos acquired by unmanned aerial vehicles. The core of the practical part involves training several convolutional neural network architectures, which were later compared using numerical metrics and visual analysis. Finally, inference of the best-performing model was conducted on test images to verify the model's generalization capabilities.