Intelligent Road Extraction And Spatiotemporal Change Monitoring From Domestic High-resolution Remote Sensing Images

Road traffic system is an important part of the city,which changes the composition of the underlying surface of the city.In the process of urbanization,urban road construction can be rapidly developed to improve the road network,so the coverage of the road has increased significantly.On the contrary,the spatial pattern of roads can also reflect the degree of urbanization,which plays an important role in the study of urbanization.The high-speed development of road traffic has not only provided convenience for mankind,but also has a negative impact on the ecological environment of the city,such as the topography and land use.Therefore,monitoring the changing features of urban roads has important practical significance for urban modernization and environmental research.The technology in the field of remote sensing is constantly innovating,the image quality is constantly improving,the speed of image update is also accelerating,and more and more acquisition methods are available.Therefore,the road extraction of high-resolution remote sensing images has become a cutting-edge research direction in the field of remote sensing.In order to solve the problems of low precision and low efficiency in traditional remote sensing image road information extraction methods.This paper proposes a deep learning road intelligent extraction method based on the domestic high-resolution remote sensing image data.The network is mainly based on the Res Net and U-net network frameworks.The coding part is mainly composed of four residual blocks.Extracting target features of different scales enhances the ability to perceive global semantic features;the decoding part has four residual blocks,through upsampling to restore the resolution of the feature map generated by the downsampling operation in the encoder,and then make a jump connection with the low-level features of the same resolution in the encoding part;in the end,the network output will get a segmentation map of the same size as the input image.This study uses the constructed semantic segmentation network model to extract the road network in the multi-time series high-resolution remote sensing image of Yaohai District,Hefei City,analyzes the shape,road network density and accessibility of the extracted results,and studies the time and space of urban roads in Yaohai District Change characteristics,realize road monitoring,and provide certain data support for urban development planning.The specific conclusions of this article are as follows:1.Establish road remote sensing interpretation signs according to road image characteristics,and finally build a domestic high-resolution remote sensing image road sample database.Because of the influence of the full convolutional neural network structure and the complex background,the traditional deep learning road extraction method has problems such as the loss of spatial features and ground detail information,which makes the extraction results have many wrong extraction situations.Based on this,this research proposes two improved road extraction semantic segmentation network models,extracting semantic features,enhancing the local information of the target,and improving the anti-interference ability and practicability of the model method.2.Based on the actual experimental results,this research method shows that the extraction results are basically close to the visual interpretation results,which can meet the accuracy requirements of general remote sensing road information extraction.The extraction results of DGRN and RCA have an average accuracy of 80.29%,88.05%,89.04% and84.54%,91.21%,91.60% for the three evaluation indexes,namely,the intersection ratio(IOU),recall rate(Recall)and comprehensive evaluation index(F1-score),respectively.Compared with the classic deep learning model methods of U-net,D-LinkNet and DeeplabV3+,there are obvious improvements.In addition,the method in this paper shows the universal advantages that conventional remote sensing methods do not possess,indicating that the method in this paper breaks the limitations of conventional remote sensing methods in remote sensing image sensors,spatial resolution,location and other factors to a certain extent.3.In order to verify the practical application of the research method,the road network was extracted from the multi-time series high-resolution remote sensing image of the study area,and the road area and length information were obtained to study the overall change trend of the road.And further calculate the density and accessibility of the road network,and analyze its detailed change characteristics.Through statistics and analysis from two aspects of road area and length in Yaohai District,Hefei City,it is concluded that on the whole,the road pattern in Yaohai District,Hefei has not changed much on the whole year,but the coverage is steadily increasing year by year.From 2015 to 2016,the road area and length in Yaohai District increased by 0.98 square kilometers and 21.81 kilometers,respectively.From2016 to 2017,the road area and length increased by 0.46 square kilometers and 24.69 kilometers,respectively,from 2017 to 2019.In 2015,the road area and length increased by1.82 square kilometers and 71.78 kilometers,respectively.Roads gradually spread to the north and east,especially the roads in the eastern region have undergone great changes;In terms of spatial distribution,the roads in Yaohai District are generally grid-like,mainly distributed in the south of the central part,and they are very densely distributed.This is the same as the experimental results of road network density and accessibility in Yaohai District.Areas with high road network density are mainly distributed in the south-central area,and areas with medium accessibility are also mainly distributed in the central area.