Course Overview: Geospatial Deep Learning for Satellite Image Analysis
In this course, we dive into geospatial deep learning, focusing on the powerful combination of satellite imagery and pytorch, specifically torchgeo for advanced image segmentation tasks. You'll learn how to obtain cloud-free satellite images from Google Earth Engine, create labels in QGIS, and apply deep learning techniques for practical applications such as cropland mapping and change detection.
By the end of the session, participants will have clear knowledge to leverage geospatial data and deep learning tools in addressing critical environmental and agricultural challenges such as encroachment into protected areas and biodiversity loss.
Course Contents
Introduction to geospatial AI
torchgeo for image segmentation
Applications of torchgeo in segementing images
Key steps in working with torchgeo for image segmentation
Raster/Image and Vector/Mask/Label data preparation for torchgeo image segmentation
Intersection and Union of datasets
The concept of BoundingBox for patch and mini_batch extraction
Selection of GeoSamplers
The role of Collate function
Data loading
Trainer selection
Inclusion of indices (Transformations)
Augmentations
Training model
Evaluating the model
Predicting over the whole region of interest
Prerequisites
Participants are expected to have basic programming skills in Python. Skills in raster image processing would also be handy, but not explicitly necessary. All participants should have computers connected to the internet as everything will be done in Google Colab, though one can reproduce them locally. It is also important to have Google Meet set up apriori. Having a Google Earth Engine account would be beneficial for preprocessing and extraction of satellite images. Having QGIS will also be necessary for label data creation, though one can do this in GEE as well.
Dates and Schedule
The sessions will be held on Xth and Yth Month Year.
From 15:00 - 19:00 CET.
Total course hours: 8
Language
English
Places
The total number of participants in the course will be 30 based on registration order. All participants who shall participate and submit their projects within 2 weeks of the course completion will receive a certificate of participation.
Fees and Discounts
Early bird (Until DATE):
€ xxx
(€ xxx for participants from UN LDCs)
Regular (From DATE until the 30th participant):
€ xxx
(€ xxx for participants from UN LDCs)
Prices include VAT
Upon registration, you will receive a confirmation e-mail of your acceptance on the course and invoice. Payment is not required during registration but should be completed before the commencement of the course. All fees collected will be used to enhance programming literacy in Kenya.
Discount
Apart from the mentioned offers, participants from Kenya will receive an additional 20% discount.