Building extraction methods based on deep learning have technical characteristics such as high extraction accuracy and fast processing speed. However, when using high-resolution remote sensing images, traditional methods often struggle to extract occluded buildings. To address this problem, this paper introduces the Multi-Feature Iterative Model (MFIM), a multi-scale feature iterative model that extracts features at different scales and fuses them in a compressed manner to obtain a structure that preserves global features without losing local details. In order to evaluate the performance of the method, we conducted a comparative analysis on the WHU dataset, using U-Net, DeepLab V3+, and HR-Net as benchmark models. The experimental results indicate that the MFIM method achieves IoU, F1, Recall, and Precision scores of 89.85%, 94.74%, 94.86%, and 94.81%, respectively. This method significantly improves the accuracy of building extraction, and its effectiveness in complex scenes makes it valuable for building extraction tasks.
|