A Hybrid Single-Shot Vision-AI and Geospatial Framework for Multimodal Infrastructure Degradation Mapping in Urban Networks

Abstract

Conventional transportation infrastructure inspection is often labor-intensive, time-consuming, and limited to individual asset types, reducing the efficiency of large-scale infrastructure management. This study proposes a hybrid Vision-AI and geospatial framework for automated multimodal infrastructure deterioration detection and digital asset monitoring. The framework integrates a YOLO-based object detection model, edge computing, geospatial synchronization, and a Digital Twin Web-GIS platform to detect, localize, and visualize infrastructure defects in real time. Experimental evaluation under multiple environmental conditions demonstrated stable detection performance, achieving 95.1% Precision, 94.3% Recall, 94.7% F1-score, and 96.4% mAP@0.5 under daylight conditions while maintaining reliable accuracy under shadowed, wet, and low-light environments. Edge deployment on the NVIDIA Jetson Orin Nano also provided low-latency inference suitable for real-time field applications. The novelty of this research lies in the integration of multimodal Vision-AI defect detection, edge computing, geospatial synchronization, and Digital Twin visualization within a unified framework capable of simultaneously monitoring heterogeneous transportation infrastructure assets. This study contributes by addressing the research gap between AI-based inspection and spatial asset management while providing an operational framework for automated maintenance decision support. The results demonstrate that integrating artificial intelligence with geospatial technologies can improve the accuracy, efficiency, and scalability of intelligent transportation infrastructure inspection and management.