Integrating pretrained Deep Learning models into GIS workflows for object detection in urban environments

The integration of Deep Learning (DL) techniques into geospatial analysis has significantly enhanced the capacity to automate object detection and feature extraction from highresolution imagery. This study presents a practical approach to incorporating pretrained deep learning models within Geographic Information System (GIS) workflows, specifically using ArcGIS Pro and models available through the ArcGIS Living Atlas. The research focuses on the application of four pretrained models designed for detecting vehicles, counting crowds, extracting building footprints, and identifying trees in urban settings. Each model was integrated into a GIS-based workflow and applied to high-resolution imagery of an urban area to evaluate its usability, processing efficiency, and detection accuracy. The results demonstrate that accurate object-level analyses can be achieved without model training or customization. Precision (0.85) and IoU (0.72) were achieved in sample-based evaluations. This study contributes a validated and reproducible GeoAI methodology applicable to planning, monitoring, and decisionmaking in complex urban environments.