The final project for the course 'Intelligent Data Analysis & Machine Learning II' at the University of Potsdam. The given task was to apply learned methods from the course on a self-chosen data set from kaggle. I chose a data set of Sentinel-2 satellite imagery of 10 classes with a total of 27k labeled images. The goal was to predict the right class labels for the given input images.
For prediction of the classes, I used a simple and an advanced convolutional neural network (CNN) and the ResNet50 with transfer learning. As a result the pre-trained ResNet50 outperforms all other models by far.
EuroSAT: A Novel Dataset and Deep Learning Benchmark for Land Use and Land Cover Classification (Vol. 12, Number 7, pp. 2217–2226). Zenodo. https://doi.org/10.5281/zenodo.7711810