2026-08-31 –, 609
This workshop builds on "Setting the Scene – GeoAI: An Intro" (FOSS4G 2024). Having established the foundations of machine learning in the previous edition, we now focus more attention to deep learning — one of the most powerful and fast-evolving branches of AI — and its practical applications in geospatial science.
Artificial intelligence continues to disrupt many fields of knowledge, and the Geoscience domain is no exception. This workshop aims to deepen participants' understanding of AI workflows, moving beyond classical machine learning towards neural network-based approaches for geospatial analysis. Participants will gain hands-on experience with open-source tools, working with Jupyter notebooks and Python throughout the session.
The workshop is structured around three main pillars. First, we briefly revisit the machine learning fundamentals— supervised classification, unsupervised clustering, and model evaluation — to ensure all participants share a common baseline. Second, we introduce deep learning concepts relevant to geospatial applications, including convolutional neural networks (CNNs), transfer learning, and semantic segmentation, illustrated through real remote sensing datasets. Third, participants will work hands-on with selected QGIS plugins for spatial data preparation — building a complete pipeline from imagery to model-ready datasets.
There will be opportunities for questions and open discussion throughout the session, encouraging participants to connect the theory to their own research and application domains.
Workshop outline:
1. Recap/ Introduction
a. Concepts recap: machine learning, deep learning
b. ML workflows, supervised classification, clustering
c. Why deep learning is suitable for geospatial tasks
d. Neural networks, CNNs, transfer learning, semantic segmentation
2. Training and inference with QGIS plugins
a. Applying simple classification with ML4QGIS
b. Inference using the dzetsaka plugin
c. Object detection with YOLO within ML4QGIS
3. Data preparation with QGIS plugins
a. Dataset structure for deep learning tasks
b. Preparing a dataset ready for deep learning (patches, labels, splits)
c. Visualizing results
4. Deep learning for Earth Observation tasks.
a. Image (scene) classification
b. Image segmentation (pixel-based)
c. Hands-on: example with pre-trained models
d. Practical considerations
5. Q&A, discussion and wrap-up
a. Open challenges in GeoAI
b. Open resources for further learning and exploration.
Requirements for attendees
• Install Anaconda, QGIS >3.40 and QGIS plugins as indicated in the preparatory email.
• Create a conda environment with the configuration shared via email
• Basic knowledge of Python and Jupyter notebooks is required.
Basic knowledge of Python and Jupyter notebooks is required.
Rosa Aguilar is an Assistant Professor at the University of Twente, where she coordinates the GeoAI module of the UNIGIS Master's programme. She is an active member of the QGIS community and brings a genuine enthusiasm for working with communities in participatory contexts. Her research focuses on developing machine learning models that support evidence-based decisions — and beyond her academic work, she is a dedicated advocate for women in STEM.