Research On Building Contour Extraction Method Based On Edge Enhancement

With the development of remote sensing technology,high-resolution remote sensing images have become the focus of attention because of their richer semantic information.It is of great significance to quickly and accurately extract building information from high-resolution remote sensing images.Remote sensing technology has practical applications in building illegal construction supervision,post-disaster assessment,urban planning,and updating building footprints.In a rapidly developing China,the gradual urbanization of the countryside is an inevitable consequence,which means that the distribution of the population is moving from scattered to concentrated.As buildings are closely related to people’s lives,extracting building footprints can reflect the scope of people’s activities and the direction of urban development,so planning the construction layout of buildings has an important role in improving the quality of life of our people.At present,effective building extraction is still a difficult problem.This paper analyses and explores the previous automatic building extraction methods,and proposes a high performance and fast building extraction algorithm for remote sensing images.The main research contents of this article are as follows:(1)Analyzing the problem of trees obscuring buildings in remote sensing images,increasing the base number of obscured samples by artificially increasing the obscuration rate,and combining data augmentation methods such as random up-and-down flipping,left-right mirroring and multi-scale scaling,a data random augmentation strategy is proposed to increase the base number of training samples and improve the generalization ability of the model.(2)A building extraction model based on edge enhancement is proposed.Using group convolution with a larger receptive field in feature extraction can enable the model to learn a wider range of local information and reduce the amount of computation.At the same time,the feature extraction module combines the dual mechanism of channel and spatial attention to effectively improve the overall accuracy of the model.(3)In order to solve the problem of error in the width of contour line in the existing Sobel edge detection,Laplace edge detection and Canny edge detection algorithms in obtaining building contour labels,an edge detection algorithm based on relative distance is proposed.In view of the lack and irregularity of building edges in building extraction results,an edge supervision assisted segmentation method is proposed.By increasing the edge loss of the building to optimize the edge segmentation effect,and then improve the overall segmentation effect of the model.Use the Wuhan University building data set to verify the effectiveness of the algorithm proposed in this paper,and conduct qualitative and quantitative analysis with the existing better semantic segmentation models UNet,PSPNet,DeeplabV3+,etc.to prove the superiority of the algorithm in this paper.Ultimately,experiments show that the algorithm can significantly improve the integrity of the whole building extraction,and the edge extraction effect can be effectively optimized combined with the edged-assisted segmentation method.