Road Extraction Of High Resolution Remote Sensing Image In Changsha-Zhuzhou-Xiangtan Urban Agglomeration Based On Fully Convolution Network

With the rapid development of aerospace remote sensing technology,the spatial resolution of optical remote sensing images is getting higher and higher,even reaching the sub-meter level,which means that the details of the objects can be clearly seen from the high-resolution remote sensing images.However,the detailed information have advantages and disadvantages for the feature extraction.On the one hand,the detailed feature information can enhance the distinguishing of the features,facilitate the identification of different features,and improve the accuracy of feature extraction.On the other hand,complex ground information will increase the information that affects the extraction of features,which will interfere with the extraction of features,especially the special road information extraction.Different from other features,the road features are complex and of various types,with various possible features and backgrounds,as well as shadows and trees.These make the traditional road extraction method using single feature unable to adapt to the road extraction of highresolution remote sensing images with complex details.In recent years,artificial intelligence technology has developed rapidly,and the core technology behind artificial intelligence,deep learning,has been widely used in various industries with its powerful feature extraction capabilities.In the past two years,deep learning technology has also been applied to the road extraction of high-resolution remote sensing images,and has achieved good results.With the continuous development of deep learning technology,remote sensing image road extraction technology will also make breakthrough progress.In this paper,deep learning is used to study the road extraction of high-resolution remote sensing images.The advanced full-convolution network model is applied to road extraction.A new road extraction deep learning model is proposed.The details are as follows.1?The method of road extraction for high-resolution remote sensing imagery and its research status at home and abroad were summarized.The basic structure and theoretical knowledge of deep learning full convolutional network were analyzed.The GF-2 training sample data set were also produced in Changsha-Zhuzhou-Xiangtan urban agglomeration,and the DeepGlobe road extraction dataset were collected.These provided a theoretical basis and data foundation for the improvement and application of the model.2?By analyzing the complex characteristics of roads in high-resolution remote sensing images,a road extraction improvement model DC-Net based on full convolution network with ASPP structure was proposed.3?Through the analysis of the correlation between the road extraction effect and the number of training data sets,data enhancement,and the full convolutional network structure.In the meanwhile,we also analyzed the road extraction model training acceleration method.These works provided guidance and support for the selection of training data in Changsha-Zhuzhou-Xiangtan urban agglomeration for road extraction models,verification of model structure,and training of models.4?A road extraction method based on improved full convolutional network model for high resolution remote sensing imagery in Changsha-Zhuzhou-Xiangtan urban agglomeration was proposed.The proposed method achieved an excellent level on the road extraction accuracy,road extraction connectivity and ground object occlusion by comparing with several typical full convolutional network models in the road extraction effect.