Suburban Building Detection from Optical Remote Sensing Images Based on a Deformation Adaptability Model

Automatic detection of suburban building areas (SBAs) is one of the research hotspots for remote sensing images (RSIs), which are widely used in the dynamic monitoring of land use, illegal building monitoring, anti-terrorism, etc. However, because the buildings are distributed targets and their appearances are quite different, the current mainstream detection methods have difficulty obtaining good detection results. To improve the detection performance, this paper presents an SBA detection method based on a deformation adaptability model. It can be divided into two main stages. (1) During key structure (KS) extraction, we first obtain building potential area by Mask R-CNN, and then we extract KS from building potential area based on roof parallel structures (RPSs) to achieve the rapid extraction of KSs. (2) In candidate region identification, to make full use of the relationship among the key structures of buildings, we present a deformation adaptability model, which adapts well to distributed targets with different appearances, and it also has strong resistance to intra-class deformations and a strong ability to eliminate false alarms. We test the proposed method on both general and complex datasets, and the experimental results show that our method has a higher detection accuracy and efficiency than typical methods.