DAMAGE DETECTION IN HISTORICAL STRUCTURES USING FASTER REGION-BASED CONVOLUTIONAL NEURAL NETWORK

Abstract

Historic sites in Thailand, many of which are recognized by UNESCO, hold immense historical and cultural significance. However, these invaluable structures are increasingly at risk of deterioration due to aging, environmental factors, vibrations, and natural disasters such as floods. Cracks and structural damage are common challenges, further aggravated by the inefficiencies and limitations of traditional inspection methods, which are labor-intensive, prone to human error, and often restricted by the inaccessibility of certain areas. To address these challenges, this study proposes an automated damage detection system specifically designed for heritage masonry structures using Faster Region Convolutional Neural Networks (FRCNN). The system efficiently detects and localizes structural damage with high precision, providing a reliable and cost-effective alternative to manual inspections. The system's performance was evaluated using five Faster R-CNN models with different backbone architectures: VGG16, VGG19, ResNet50, ResNet101, and ResNet152. In this comparison, the ResNet152 model achieved the highest accuracy of 81.29% and a recall value of 81.13%, demonstrating its capability to effectively detect cracks in historical structures.