The increased availability of high-resolution satellite imagery allows to sense very detailed structures on the surface of our planet and opens up new direc- tions in the analysis of remotely sensed imagery. While deep neural networks have achieved significant advances in semantic segmentation of high-resolution images, most of the existing approaches tend to produce predictions with poor boundaries. In this paper, we address the problem of preserving semantic seg- mentation boundaries in high-resolution satellite imagery by introducing a novel multi-task loss. The loss leverages multiple output representations of the seg- mentation mask and biases the network to focus more on pixels near bound- aries. We evaluate our approach on the large-scale Inria Aerial Image Label- ing Dataset. Our results outperform existing methods with the same architec- ture by about 3% on the Intersection over Union (IoU) metric without additional post-processing steps. Source code and all models are available under https: //github.com/bbischke/MultiTaskBuildingSegmentation.
Bischke Benjamin, Helber Patrick, Folz Joachim, Dengel Andreas, Borth Damian
26 Oct 2019