An Open-Source YOLO & QGIS Workflow for Geospatial Object Detection
TL;DR
This work presents an open - source workflow that bridges the gap between deep learning and GIS, with emphasis on reproducibility, scalability and integration into GIS - based analytical workflows.
Abstract
: The increasing availability of open geospatial data and open - source software has created new opportunities for integrating machine learning into GIS workflows. Object detection models have demonstrated strong performance in computer vision tasks, yet their adoption within geospatial analysis pipelines remains limited. This work presents an open - source workflow that bridges the gap between deep learning and GIS, with emphasis on reproducibility, scalability and integration into GIS - based analytical workflows. The proposed approach is demonstrated on two distinct types of geospatial data. The first dataset consists of digital feature models (DFM) with specialized visualization for archaeological features from Lieskovský et al. (2022), derived from airborne laser scanning (ALS), used for enhancing microtopographic features displayed in Figure 1. The second dataset includes airborne orthophoto mosaics (GKÚ Bratislava, NLC) used for detecting swimming pools in both urban and rur a l en vi r o n m ent s displayed in Fig ure 2 . These datasets represent fundamentally different data characteristics, combining elevation - based and spectral information, and allow evaluation of the workflow’s generalization capability and scalability.