DETECTING FLAT ROOF DEFECTS WITH MACHINE LEARNING AND DEEP LEARNING TECHNIQUES

Abstract

In the context of the ongoing war in Ukraine, ensuring the safety and longevity of buildings and infrastructure is paramount. Traditional inspection methods for detecting structural damages—such as cracks, spalling, or corrosion—are labor-intensive, time-consuming, and prone to human error. This study addresses these challenges by leveraging deep learning techniques, particularly for flat roofs. Models including CNNs, U-Net, YOLO, and autoencoders enable efficient defect detection from both visual and thermal data, even in hazardous or hard-to-reach areas. UAVs facilitate rapid image collection, thereby reducing costs and risks associated with manual inspections. Our findings indicate that AI-driven methods can significantly improve inspection accuracy, accelerate maintenance, and ensure early detection of critical damage, crucial for infrastructure safety in conflict-affected zones. Ultimately, integrating deep learning into structural health monitoring offers a robust and automated approach to safeguarding buildings and optimizing maintenance efforts.