AI-Based Debris Detection for Resilient Bridge Monitoring in Flood-Prone Areas

Bridges are highly vulnerable to extreme hydraulic events, where debris accumulation acts both as a direct load and as a driver of scour at the foundations. Scour may remain hidden, but debris is a visible and actionable early-warning indicator. This paper presents an AI-based vision approach for automated debris detection. A prototype object detection model was trained on a hybrid dataset of synthetic and real images. Using the YOLOv5 framework, a mid-sized model with higher accuracy was developed. The study demonstrates how synthetic images can accelerate development and prototyping in scenarios where real-world examples are rare, hazardous, or difficult to capture. Early results indicate that, even with limited data, the system can reliably detect debris in real time and classify risk in a prototype dashboard, demonstrating potential as an early-warning tool for bridge monitoring.