The aim of this study project is to explore the suitability of convolutional neural networks (CNNs) for analyzing and classifying high-resolution satellite imagery from nano satellites. In addition, participants will get a basic understanding of the key steps for using deep learning in remote sensing image analysis, and how to apply them to a real-world example.
During the first part of the project, participants will get a practical introduction to deep learning in remote sensing using R. You will work on a tutorial that features the workflow with complete code examples (building of a CNN, data preparation, model training, prediction, …) and a minimal data set to get acquainted with the methodology and practical issues involved.
During the second part, you will work in groups on your own projects analyzing high resolution (3-4 m) satellite imagery in a practical use case. For the project work, each group will get access to a GPU-powered instance on amazon web services with the necessary software pre-installed.
Previous experience in deep learning will be useful but is not required. However, you will be required to show the engagement and curiosity necessary to work your way into the topic using the provided (and maybe additional) material.
Contact: christian.knoth@uni-muenster.de |
eG