Multi-Source High-Resolution Remote Sensing Image Fusion Of Urban Buildings Deep Learning Change Detection Research

Buildings are important places for human survival and their spatial distribution dynamics are also related to social development,economic growth and environmental improvement.Therefore,the detection of building changes is extremely important in the fields of land cover planning,sustainable urban development and disaster emergency management.With the continuous development of remote sensing technology,the spatial resolution of remote sensing images has been improving,and its sub-meter spatial resolution makes the detailed information of building edges in the images more abundant,which makes the fine building change detection possible.However,with the higher resolution,the problem of inter-class similarity and intra-class variability in the images becomes more intense,making the existing change detection methods unable to effectively extract the building features of the images.At the same time,due to the limitation of sensor acquisition conditions,it is often difficult to obtain different simultaneous high-resolution remote sensing images with the same spatial resolution in reality,and the traditional resampling methods will lose the detailed spatial information in high-resolution images,causing difficulties to the change detection work.In recent years,deep learning has been widely used in the field of remote sensing due to its powerful feature extraction ability,and it has been studied in depth in the reconstruction of super-resolution remote sensing images and building change detection.In this paper,we use deep learning technology to study the change detection of buildings in multi-source high-resolution remote sensing images,and apply the advanced super-resolution reconstruction network and the change detection model based on twin neural networks to the building change detection.The research work and innovation points of this paper are as follows.(1)To address the problem of spatial information inconsistency and ineffective utilization of multi-source remote sensing images with different spatial resolutions,a super-resolution reconstruction network based on deep learning is proposed to reconstruct the temporal images with lower spatial resolution in multi-source remote sensing images into super-resolution images,so that they have similar image feature information as high-resolution images and improve the interpretability of multi-source high-resolution remote sensing images.In the three multi-source high resolution remote sensing image change detection datasets,the use of super-resolution reconstructed images for simply speechless change detection has a high accuracy improvement.(2)To address the problems of inaccurate location detection,incomplete target extraction and unclear edges in building change detection of high-resolution remote sensing images,this paper proposes a multi-feature fusion method that combines morphological building index(MBI),non-maximum suppression(NMS)Sobel edge detection features and LSD right-angle edge features combined with Harris corner point detection with RGB spectral features of the image respectively,and inputs an improved spatio-temporal attention network(STANet)deep learning network model for building change detection.The results show that all three feature combinations have high improvement on the change detection results of the original image,among which the RGB + Harris-LSD feature combination has the most obvious improvement on the detection accuracy.(3)To address the problems of difficult training and underutilization of temporal phase information in existing change detection methods,this paper improves the feature extraction layer of STANet and fuses it with MBI,NMS-Sobel and Harris-LSD features respectively.The results show that the improved STANet network has higher building change detection accuracy,is suitable for a variety of complex building change detection scenarios,and has stronger model generalization.