Research On Improvement Of Remote Sensing Images Building Segmentation Algorithm Based On Feature Fusion

The information contained in remote sensing images is an important source of data in many fields.Segmenting buildings from many targets in remote sensing images is of great significance for intelligent city construction,national defense and security,and other aspects.With the rapid development of deep learning,great progress has been made in building segmentation of remote sensing images based on deep learning methods.However,the scale of buildings in remote sensing images is variable,the edges are complex,and there are many background information interferences,making the segmentation of buildings in remote sensing images based on deep learning still full of challenges.Therefore,in this paper,a building segmentation algorithm based on deep learning technology and from the perspective of feature fusion is proposed to solve the problems of low edge accuracy of building segmentation in remote sensing images,small target building segmentation adhesion,and mis-segmentation and missing segmentation.The main work and innovations of this paper are as follows:(1)Aiming at the problems of low accuracy of building edge segmentation in remote sensing images and small target building segmentation adhesion,a multi-scale feature fusion network EAMFNet based on edge attention is proposed in this paper.Specifically,an edge detection branch based on HED algorithm is added to the EAMFNet model to extract the edge feature information of buildings.Moreover,location information and spatial details information are fused by skip connections in the HED algorithm.At the same time,the high-level features rich in strong semantic information will be superimposed on the low-level features in the edge detection branch by the gated attention module AG in EAMFNet,thus highlighting the building area and improving the extraction effect of edge feature information.In addition,the atrous spatial pyramid pooling module ASPP in EAMFNet will be used to acquire larger multi-scale features of the receptive field.Finally,feature fusion module FFM is used to fuse edge feature information and Semantic information.The experimental results show that with the help of the proposed network EAMFNet,the segmentation accuracy of building edges is not only improved,but also the segmentation adhesion problem of small target buildings is alleviated,and the segmentation accuracy of buildings in remote sensing images is improved as a whole.(2)Due to the variable scale of buildings in remote sensing images,it often leads to issues such as mis-segmentation and missing segmentation of buildings.To address this issue,a multipath feature fusion network MPFFNet is proposed based on UNet in this paper.First,in this paper,the RDI module with multi-scale feature extraction capability is embedded in UNet to take into account buildings of different scales.Secondly,continuous atrous convolution and multipath residual structure are used to design the MPRCA module,which is placed at the tail of the coding part,and is used to expand the receptive field of the feature map and obtain rich multi-scale context information.The experimental results show that the requirements of different scale buildings can be satisfied by the proposed algorithm MPFFNet,and the missing segmentation and mis-segmentation segmentation of buildings can be effectively alleviated.(3)To facilitate the use of various segmentation algorithms to segment remote sensing image buildings,the automatic segmentation system of remote sensing image buildings is designed in this paper.Specifically,the construction of the system interface and the writing of related business logic,as well as the organization and coordination of various algorithm work are implemented by Qt Desinger and Py Qt5 technologies.The test results show that the system is simple and practical,and the buildings in remote sensing images can be segmented effectively.