Dominik Kossmann, Thorsten Wilhelm and Gernot A. Fink
2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, pages 2616-2619, 2021.
Through the rise of new remote sensing datasets with suffi- cient size for current deep learning models, land cover clas- sification results have improved in recent years. Unfortu- nately, earth exhibits a natural imbalance in different land cover classes, also visible in these datasets. In the domain of zero-shot learning, image attributes have enabled better re- sults in transfer learning by creation of a mid level represen- tation between different classes. This representation can also be constructed for land cover data and used to detect minor- ity classes through shared features. We propose a way for generation of attributes by text mining for one of the biggest land cover datasets. With these attributes we achieve state-of- the-art performance in land cover classification and improve results especially for minority classes. Further, we show that these attributes have great potential in weakly supervised land cover segmentation.