The maintenance of civil infrastructure has become a major challenge for modern engineers due to its continuous growth, especially buildings and bridges which are susceptible to structural damage caused by factors like aging, design flaws, and natural disasters, which can seriously jeopardize their safety, health, and structural soundness. The conventional inspection methods are not only costly and timeconsuming, but also pose safety hazards, rendering them ineffective. The paper introduces an automated method for detecting and assessing the severity of damages in buildings and bridges, which aims to address and mitigate the limitations of traditional techniques. A collection of 5000 images, ranging in size from 416x416 to 640x640 pixels, were gathered from damaged sites and annotated in polygon annotation format, with damages categorized into three classes: spalling, corrosion, and crack. To expand the dataset and enhance the precision of the deep learning models, data augmentation techniques from the data Albumentation library were utilized for image processing. Several object-based instance segmentation deeplearning models, including Yolo V5, V7, V8 Instance Segmentation, and Mask- RCNN, were trained on 80% of the dataset to obtain the coordinates of the damaged area's outline and generate masks for the detected damages. The area of these masks and the percentage of damage severity in an image were computed. The trained models achieved an accuracy range of approximately 60% to 70%, indicating the potential effectiveness of instance segmentation deep learning models. The proposed approach offers a quick and efficient method for determining the severity of structural damages, resulting in improved safety and decreased maintenance costs for buildings and bridges. To enhance the accuracy and precision of the model, future research could include a larger dataset with additional types of structural failures.
Furthermore, the automated approach can be integrated into mobile monitoring and inspection devices, such as drones, as well as street camera systems to identify damaged buildings following natural disasters. An alarm can be triggered based on a specific threshold value, and appropriate actions can be taken to reduce the response time for emergency management and improve the structures' resilience to natural disasters.