Research On Intelligent Building Extraction From Remote Sensing Image Based On Parallel Network

Buildings are one of the most important components of surface information.High-resolution remote sensing image building extraction tasks play an important role in urban planning,smart city construction,urbanization process and other fields.Quickly,accurately and intelligently extracting the characteristic information of buildings from satellite remote sensing images has become a difficult point and hot spot in satellite remote sensing image research.Due to the limitations of traditional machine learning-based building extraction methods in practical applications such as large workload,low efficiency,and low extraction accuracy,in recent years,the technology of extracting buildings based on deep learning algorithms has been used in various algorithms for building extraction.To emerge in China.Aiming at the problems of the existing deep learning extraction methods that the shallow features are not effectively used and the small target information is easily lost,a parallel network is proposed.The network uses the dense connection mechanism to fully extract feature information,and builds a parallel structure to retain spatial information of different feature resolutions,enhance feature information of different depths and scales,and reduce the loss of detailed features;at the same time,it uses the hollow space pyramid module to obtain different receptive field information.Extract deep building features at different scales to improve the accuracy of building extraction.Taking the Yaohai District of Hefei City as the research scope,using the intelligent extraction model of buildings to extract single building information and multiple types of building information from multi-time series remote sensing images in Yaohai District,explore the temporal and spatial distribution changes and reasons of buildings in Yaohai District.The related achievements of this article are as follows:(1)According to the actual situation of GF-2 remote sensing image,establish the high-resolution remote sensing image building interpretation logo and summarize the features of the building interpretation logo,construct single-class building samples and multi-class building samples for model training.(2)In the extraction of buildings from GF-2 remote sensing images,the method in this paper has an overall accuracy of 97.19%,an intersection ratio of 74.33%,and a comprehensive evaluation index of 85.43%.The comparative experiments show that each index is higher than that of traditional methods and others.Deep learning methods,edge detail processing and recognition of small targets have more advantages.The method in this article is also applicable to extracting buildings from GF-2 remote sensing images in different regions.In addition,buildings that deal with multi-source remote sensing images still have The good extraction effect reflects the practicability of the method in this paper.(3)In the multi-class extraction of buildings,the method in this paper has good extraction results for high-rise buildings and factories with obvious characteristics on remote sensing images.The overall accuracy,intersection ratio,and comprehensive evaluation index are85.84%,51.18%,and 54.03%.By comparing with other deep learning methods,it shows that it has better recognition ability for images with complex background and multiple types of buildings of different scales.(4)The distribution area of buildings in Yaohai District is relatively concentrated in the middle and south,and the distribution of buildings in the south is more dense.As the years progressed,the buildings in the central area increased year by year,and extended to the north and east.From 2015 to 2019,the building area has increased from 21.36km~2 to 24.24km~2,and the building area has been increasing in each year.From 2016 to 2017,the area of buildings has increased significantly compared to the rest of the time period,with a growth rate of 9.02%,while the overall building area from 2017 to 2019 has not changed much,with a growth rate of1.74%.(5)In the extraction of multi-class buildings,it is found that the factories are mainly distributed in the industrial area in the middle,while the sparse low buildings in the north and east are distributed in the fields.The south of Yaohai District is close to the main urban area of Hefei City,mainly consisting of urban residential areas and Office buildings,etc.,this is also the reason for the denser buildings in the south.From 2015 to 2019,high-rise buildings increased year by year,and the area of low-rise buildings and contiguous buildings decreased year by year.Through the analysis of specific time and space changes in the Yaohai District Industrial Park,Longgang Comprehensive Economic Development Zone,Qilizhan Street,and Daxing Town,it has also been verified that this trend is due to a series of transformations of old cities and shanty towns.Proposal of the policy.Analyzing the plot ratio of the residential area in Changhuai Street,the main distribution characteristics of the value are:the value of the plot ratio of the residential area is distributed in the interval of 1.0 to 3.2,and the residential areas with higher plot ratio are mainly distributed in the central landscape axis,the core business area and Near the intersection of the main road.