Building Extraction from High-Resolution Remote Sensing Image Based on Improved ResUNet Network

Existing methods include traditional methods and deep learning-based methods, but there are problems such as inefficiency and poor generalization ability. In this paper, a new method named IARUNet is proposed, which integrates Improved Atrous Spatial Pyramid Pooling (IASPP) module and ResUNet. The IASPP module replaces the convolution module in ResUNet network, which improves the ability of building extraction. Notably, experimental results from the Massachusetts dataset demonstrate that IARUNet achieved an average intersection over union (MIoU) of 84.31%, a recall rate of 89.42%, a precision of 90.04%, and an F1 score of 89.94%, which is significantly improved compared with other existing models. In addition, the IASPP module enhances the retention of context information, further improving the extraction effect.