Data-Driven Feature Learning with Discriminative Models for Satellite Time Series

Marc Rußwurm

Abstract: Machine learning enables us to learn relationships from annotated datasets. This is useful in fields where finding an explicit mathematical model is difficult. The classification of agricultural field parcels from satellite image time series is such a field where feature learning can be used to accurately predict the crop type from available satellite data. This dissertation investigated what deep learning models and learning schemes are most effective when many and few labels are available.