Research On Building Extraction Based On Deep Learning

With the continuous progress of remote sensing technology,the problem of building extraction from remote sensing images has been paid more and more attention.The rapid development of computer vision in digital image processing has brought new opportunities for building extraction from remote sensing images,and data-driven deep learning methods have become the mainstream method.It is of great significance to study deep learning algorithms for higher-precision extraction of buildings in remote sensing images.This paper mainly uses the deep learning algorithm to classify and extract the buildings in the remote sensing images of the rural areas around the Great Wall in the suburbs of Beijing obtained by drones.Aiming at the shortcomings of the current mainstream deep learning instance segmentation algorithm Mask R-CNN in feature fusion and poor prediction of instance mask boundaries,two improved schemes are proposed to realize the automatic extraction of buildings in remote sensing images of rural areas around the Great Wall in the suburbs of Beijing.The main work innovations are as follows:1.A feature pyramid network based on attention augmentation guidance is proposed,which enhances the features with self-attention mechanism and contextual information and guide feature learning in adjacent layers of the feature pyramid network with using channel attention mechanism.The experimental results show that the improved algorithm model can obtain higher detection accuracy.2.An edge optimization algorithm for instance segmentation based on wavelet transform is proposed.It uses wavelet transform to perform feature extraction on real masks to obtain the low-frequency and high-frequency information,which is used for network-supervised main feature and edge feature learning of masks.The experimental results show that the improved algorithm model is more accurate in the segmentation edge prediction and obtains higher prediction accuracy.3.The improved algorithm model is used to extract buildings from remote sensing images.First,a dataset for village building extraction is constructed using UAV remote sensing images.And then we trained and evaluated on the dataset by the improved algorithm model to extract buildings from the entire village.The experimental results show that the improved algorithm model can accurately extract the buildings of the whole village.