Building Extraction from High-Resolution Remote Sensing Images Using Improved HRNet Method

The accurate extraction of buildings using high-resolution remote sensing images is crucial for urban development. Deep learning has become an important method for building extraction, but there are still major difficulties in the extraction of building edges and small structures. To improve the accuracy of building extraction, this study applies the rotation-variable-size-window attention mechanism (RVSA), and introduces the Segfix mechanism in the post-processing stage to improve the boundary accuracy of building extraction, based on optimized High-Resolution Network (HRNet) and fused OCRNet algorithm. The experiment using the WHU aerial building dataset proved that the new method has better accuracy and robustness, and the building extraction accuracy of the new method is 92.457%, which is 2.585% higher than the traditional HRNet.