Workshop 1. Earth Observation Foundation Models with Prithvi
Presented by IBM Research
This workshop introduces how to utilize the Prithvi-EO-2.0 foundation model and the Terratorch open source library for fine tuning and inference. This session is ideal for those familiar with the basic concepts of machine learning and Earth observation data.
Participants will come away from the workshop understanding how to set up and use the TerraTorch environment, load and preprocess satellite imagery, fine-tune a foundation model for different AI tasks, and evaluate the model performance.
Workshop Agenda
Part 1: Introduction to Prithvi-EO-2.0 and TerraTorch
· Overview of Prithvi-EO-2.0.
· Setting up your environment.
· Basic usage of TerraTorch.
Part 2: Hands-on Session
Participants will be provided with an environment to run their code and notebooks tailored to explore the following tasks:
· Image Classification: Perform land cover mapping using satellite imagery.
· Segmentation: Extract and delineate geographical features with precision.
· Regression: Predict environmental variables from satellite data.
Workshop 2. An introduction to Machine Learning for EO
Presented by NEODAAS
Target audience. EO users with limited experience in machine learning and AI approaches, and a basic knowledge of Python
Purpose. This course is designed to introduce users from a broad range of backgrounds to machine learning approaches for EO data and give them practical experience running them in python with Jupyter notebooks.
Length. 1/2 day
Learning objectives. By the end of the course participants should:
- Understand what is machine learning and the different types of ML approaches (classification, regression, and segmentation)
- Be able to load, process, and visualise optical satellite data using existing Jupyter Notebooks (Python).
- Trial and compare the running of applications on the Central Processing Unit (CPU) and the Graphics Processing Unit (GPU), and why it may become important to your future work to have access to a GPU cluster for Machine Learning and Artificial Intelligence.
Description. The course will start with an introductory talk, outlining some fundamental Machine Learning theory, and then going into more detail on the specific application of ML for EO, with a focus on practical advice and considerations that need to be made. Then we will move onto a practical session using a series of Jupyter Notebooks which will take participants through several applications of machine learning at their own pace, from simple random forest models to U-Net for semantic segmentation. This course is hosted on the NEODAAS MAGEO (Massive Graphics Processing Unit Cluster for Earth Observation), which can be accessed through a browser. Participants will need to bring their own laptop for this session, register for Github (github.com) if they do not already have an account, and provide their Github Username ahead of the training course.