Application of deep learning and GIS-based risk assessment for urban gas pipelines in geologically active regions: a case study of Almaty

Purpose. To develop a lightweight, modular method for assessing urban gas-pipeline vulnerability in seismically active areas, using Almaty as a pilot case. Methods. The approach couples deep – learning – based surface defect detection (a customized You Only Look Once model – YOLOv11-M) with GIS risk modeling built on a Digital Elevation Model (DEM), terrain slope, and active fault proximity. Detected defects are georeferenced and fused into a Composite Risk Index (CRI) within QGIS. Results. Validation indicates high recognition accuracy (mean Average Precision, mAP = 95.3%; F1-score = 95.2%) and a strong spatial correlation between geohazard zones and defect density, resulting in risk maps that prioritize segments for inspection and prevention. Significance. The framework is scalable and cost-effective for data-constrained municipalities, supporting urban resilience and Smart City strategies.