Buildings Detection in VHR SAR Images Using Fully Convolution Neural Networks

摘要: This paper addresses the highly challenging problem of automatically detecting man-made structures especially buildings in very high resolution (VHR) synthetic aperture radar (SAR) images. In this context, has two major contributions: Firstly, it presents a novel and generic workflow that initially classifies spaceborne TomoSAR point clouds $ - generated by processing VHR SAR image stacks using advanced interferometric techniques known as tomography (TomoSAR) into non-buildings with aid auxiliary information (i.e., either openly available 2-D building footprints or adopting an optical classification scheme) later back project extracted points onto imaging coordinates to produce automatic large-scale benchmark labelled (buildings/non-buildings) datasets. Secondly, these datasets masks) have been utilized construct train state-of-the-art deep Fully Convolution Neural Networks additional Conditional Random Field represented Recurrent Network detect regions single image. Such cascaded formation successfully employed computer vision remote sensing fields for but, our knowledge, not applied The results detection are illustrated validated over TerraSAR-X spotlight covering approximately 39 km$ ^2 almost whole city Berlin mean pixel accuracies around 93.84%