Underground pipeline detection technology driven by graph neural network in surveying and mapping engineering

This paper addresses the insufficient accuracy and inadequate spatial relationship representation in urban underground pipeline detection by constructing a graph neural network (GNN) detection model that integrates spatial topological information. Multi-source pipeline data—including coordinates, burial depth, material, and diameter—are organized into a unified structural format, and a spatial topology graph with node-edge representation is established. The model jointly learns adjacency relationships and attribute features through a multi-layer feature propagation mechanism. Experimental results on multi-type pipeline datasets demonstrate overall detection accuracy of 91.8%, with F1-score reaching 90.7%, outperforming CNN-based methods by over 6% points. The model exhibits stable convergence and strong cross-dataset generalization. These findings provide engineering reference for intelligent detection and refined management of urban underground pipelines.