loading page

Fusion of adaptive edge features and geometric features for building extraction from remote sensing images
  • hongning Qin,
  • Zili Li
hongning Qin
Guangxi Normal University
Author Profile
Zili Li
Guangxi Normal University

Corresponding Author:[email protected]

Author Profile

Abstract

At present, most of the deep learning-based building extraction is based on semantic segmentation,it is strongly influenced by the data scene, as the geometric features of the building are not taken into account. The traditional edge detection methods ignore the different effects of different edge detection operators on images when processing remotely sensed buildings.In order to extract buildings more effectively, we propose a method for extracting buildings named fusion of adaptive edge features and geometric features for building extraction from remote sensing images. Firstly, calculate the final contribution of 5 common operators to edge detection and perform edge enhancement by adaptively determining the weight coefficients; Secondly,the processed image is then texture smoothed using the RTV model; then this is followed by the marking of the connected areas.Specific filters are constructed to filter out non-building noise based on the geometric characteristics of the building; Finally, hollow filled reserved area,a more accurate map of the building results is generated. In this paper, six remote sensing images of buildings situated in different landscapes were selected. The experimental results show that the algorithm has advantages over classical algorithms and deep learning algorithms.