Srbuildingseg-E2: An Integrated Model for End-to-End Higher-Resolution Building Extraction

Lixian Zhang, Runmin Dong, Shuai Yuan, Haohuan Fu

Abstract: Automatic and accurate extraction of buildings from remote sensing images plays a vital role in many applications. However, existing approaches for building extraction generally apply high-resolution remote sensing images as input to attain high-resolution extraction results, which is time consuming and limited due to its spatiotemporal accessibility and cost. To address this challenge, in this paper, we propose an end-to-end approach, i.e., SRBuildingSeg-E2, to achieve higher-resolution building extraction from relatively low resolution remote sensing images. By integrating super resolution and semantic segmentation techniques, our proposed approach can attain high-resolution representations using low-resolution input. The quantitative assessment results reveal its promising performance in higher-resolution building extraction.