Building Detection and Segmentation Method in Aerial Photograph Based on Mask R-CNN

In the context of large-scale building segmentation scenarios, where the detection and segmentation accuracy of buildings in aerial images are low and the costs are high, a method based on Mask R-CNN is proposed for automatic and efficient detection and segmentation of buildings in aerial images. To address the issue of poor detection and segmentation performance of trained models in aerial images containing disturbance such as shadows, a non-random masking data augmentation approach is introduced to improve the segmentation accuracy of buildings with shadows and other disturbance. Additionally, experiments are conducted to evaluate the impact of transfer learning methods, data augmentation techniques, and different depths of residual networks as feature extractors on the performance of building detection and segmentation models. The generalization performance of the model is tested on satellite remote sensing data. Experimental results demonstrate that the proposed method achieves automatic detection and segmentation of buildings in aerial images, making it suitable for scenarios where a large number of buildings need to be segmented with high efficiency.