Master thesis defense by Nicholas Indrehus
Convolutional Neural Networks for Classification of Remote Sensing Data
This work explores the use of convolutional neural networks (CNNs) for pixelwise classification of remote sensing data. In particular, we make use of time series satellite data from the Landsat program, which depicts one of the most popular programs in the field of remote sensing. The experimental evaluation covers two challenging learning scenarios and clearly shows the potential of CNNs in this context. As shown in this thesis, CNNs can be successfully applied for these tasks and yield a significantly better performance for certain scenarios and classes compared to state-of-the-art methods.
Supervisor: Fabian Gieseke
External Examiner: Ira Assent