Urban UAV Images Semantic Segmentation Based on Fully Convolutional Networks with Digital Surface Models

Unmanned aerial vehicles (UAV) have had significant progress in the last decade, applying to many fields for its convenience to explore areas that men cannot reach and the progress of image processing. Still, as basis to further application, semantic image segmentation is one of the most difficult challenges. In this paper, we propose a method for urban UAV images semantic segmentation, utilizing the geographical information, digital surface models (DSM). We introduce an end-to-end, dual stream fully convolutional networks (FCN) based classifier with DSMs to get the segmentation results, which utilizes the proposed fusion decision strategy instead of the pixel-level classification strategy, along with a short-cut scheme. The experiments show that the proposed structure performs better than state-of-the-art networks in multiple metrics.