Building extraction from remote sensing images using edge preserving mechanism

Remote sensing images of buildings are due to the diversity of the structure of the building itself and the complexity of its surrounding environment, and due to the impact of factors such as lighting and shadows, there are problems with incomplete structures and unclear edges in the extraction of buildings. Therefore, a building segmentation network incorporating Laplace edge preservation mechanism is proposed. Add a Laplace edge preserving layer after each cross fusion layer in HRNet. For each input branch of HRNet, use a 3x3 convolution kernel to implement the Laplace operator. The output is a feature map processed by the edge preserving Laplace operator. This mechanism can smooth the noise and details in the image, maintain the sharpness of the edges, and improve the accuracy of image segmentation. The model was tested on the open source WHU Building Dataset, Satellite Dataset II (East Asia) building dataset, with an accuracy of 92.09, a recall rate of 91.83, an intersection ratio of 84.95, and an F1 score of 93.43. Compared with the mainstream methods DeeplabV3+, PSPNet, and HRNet, the experimental results show that the proposed method effectively improves the accuracy of building segmentation, while making the boundaries of buildings clearer and more complete. This paper proposes a remote sensing image building extraction model based on Laplace edge preservation mechanism. The experimental results show that the network has a good extraction effect on building extraction from remote sensing images, and has certain reference value in engineering applications.