Fully Convolution Networks for Buildings Segmentation in High Spatial Resolution Remote Sensing Images

EasyChair Preprint 877

Abstract

Segmentation with high precision of buildings in high resolution remote sensing images is an important issue for the construction of digital city and classification the military targets. The main technology of this paper is based on FCN-VGG (Fully Convolution Networks-Visual Geometry Group) to implement the building segmentation in remote sensing image, in which the features are extracted by multiple convolution and the transposed convolution is employed to recover the original image size. The valid experiment results show that the method proposed could overcome disturbance produced by roads, trees, pedestrians, the cross-bedding of buildings effectively. Meanwhile, the multi- experiment demonstrates the generalization of the buildings segmentation in multi-resolutions.

Keyphrases: buildings segmentation, fully convolution network, remote sensing image

@booklet{EasyChair:877,
  author    = {Linxiang Li},
  title     = {Fully Convolution Networks for Buildings Segmentation in High Spatial Resolution Remote Sensing Images},
  howpublished = {EasyChair Preprint 877},
  year      = {EasyChair, 2019}}