Research On Building Segmentation Method Of Satellite Remote Sensing Images Based On Context Informatio

With the rapid development of remote sensing earth observation technology,the quantity of high-resolution satellite remote sensing image data shows exponential explosive growth,and this trend becomes more obvious as time goes by.Buildings are an important part of human life,and building segmentation based on high-resolution satellite remote sensing images has important application value in the fields of urban development,smart city construction,land measurement and map mapping.The traditional high-resolution remote sensing image building segmentation has problems such as poor generalization and difficulty in handling complex scenes.Although the segmentation of high-resolution remote sensing images based on deep learning can have high generalization,there are still cases of multi-scale target loss and unclear edges.Accordingly,this paper proposes a deep learning semantic segmentation network based on spatial context and object context for feature resolution from contextual semantic information,and the main work is as follows.(1)To address the problems of too little perceptual field and insufficient multi-scale information in deep learning networks,this paper proposes a one-sided two-branch network based on spatial contextual information.First,we introduce the Res2 Netplus convolutional unit to optimize the Xception backbone network to improve the perceptual field of the network and make the network more accurate in recognizing multi-scale information;then,we embed CBAM and feature fusion modules in the middle layer of feature processing to suppress nonimportant features and enhance multi-scale information.Finally,feature upsampling is carried out by a two-branch decoder fusing the coded semantic information to achieve crosscomplementation of the underlying semantic and decoding information.The above method is able to capture multi-scale information and reduce the occurrence of missed building detection cases.(2 To address the problem of unclear edges of building segmentation,a class feature enhancement network based on object context is proposed on the basis of the previous work.First,the coding and decoding structure of Res Net50 as the backbone network is designed to capture inter-pixel contextual semantic information,and a dual-attention mechanism is proposed in the jumping layer to achieve gated screening based on codetail edge semantic information.Then,feature transformation methods are embedded in the intermediate layer to enable the model to extract more nonlinear features.Finally,a class feature-based branching decoder is designed to effectively enhance the extraction of building class features and achieve the accurate segmentation of building edges.