Deploying UAV-based detection of bridge structural deterioration with pilgrimage walk optimization-lite for computer vision

Bridges are crucial components of national infrastructure, requiring rigorous maintenance and inspections to ensure their safety and functionality. Inspections are incredibly challenging and costly when accessing elevated bridge decks and substructures, particularly over riverbeds. To address these challenges, this study introduces a novel system integrating advanced computer vision-based deep learning, metaheuristic optimization, and Unmanned Aerial Vehicle (UAV) technology to revolutionize bridge inspections. This system uses UAVs to capture high-resolution images, which are then processed by the You Only Look Once (YOLO) models for instance segmentation. The YOLOv7 model, fine-tuned with the Pilgrimage Walk Optimization (PWO)-Lite algorithm, achieved the highest accuracy, recording a 65.6 % mAP 50 on the testing set. The PWO-Lite algorithm enhanced the YOLOv7's performance, increasing mean average precision by 4.1 % compared to YOLOv7 with augmented images and 13.9 % compared to YOLOv7 using original photos. These improvements significantly boost the model's generalization capability. This integration facilitates precise, automated deterioration quantification and supports accurate, efficient maintenance cost estimation. Explicitly designed for detecting underside deterioration in composite bridges, this system provides bridge management authorities and construction firms with a comprehensive tool. Our findings indicate improvements in predictive maintenance for bridges, offering actionable insights that aid informed decision-making in bridge inspections and maintenance planning. This advancement enhances the sustainability and safety of bridge infrastructure, representing a substantial step forward in infrastructure management and UAV applications. • An unmanned aerial vehicle captures images of deterioration underneath the bridges. • Concrete and steel deterioration is classified and segmented by the proposed model. • A folk culture-inspired metaheuristic algorithm constructs the optimal segmentation model. • A cost estimation system is developed to simulate on-site inspections and automatically generate reports. • The proposed system supports decision-making for bridge management units and construction companies.