Research On Road Extraction From Highresolution Remote Sensing Images Based On Deep Learning

With the rapid development of satellites and other aerospace technologies,the cost of acquiring remote sensing data is getting lower and lower.People can obtain the massive amounts of remote sensing data needed through a variety of sensor devices on the remote sensing platform.These data enable humans to observe and explore information on the earth more comprehensively and in real time.With the continuous advancement of remote sensing technology and sensing equipment,the quality of remote sensing data has been greatly improved.The spatial resolution of optical remote sensing images has reached the sub-meter level,and the detailed information of the ground features in the data is becoming clearer.Using high-resolution remote sensing data extracting ground feature information is of great significance to applications such as digital city construction,smart city planning,and traffic management.Road information extraction is an important part of feature information extraction.Due to the complex and diverse road features,traditional road extraction methods based on a single feature design cannot meet the needs of high-precision research.In recent years,deep learning methods have been widely used in the research of high-resolution remote sensing images.Thanks to the powerful feature learning capabilities of deep learning technology,road extraction technology has made considerable progress,and many excellent deep learning algorithms have emerged.However,the road information in remote sensing images is complex and diverse,and the existing technologies and methods still need to be studied and improved.This article is based on high-resolution remote sensing data,combined with deep learning network model to study the extraction technology of road feature information.Based on the research and analysis of the existing deep learning algorithm,an improved algorithm based on AS-Unet is proposed,which has achieved good results.The main research work and innovations of this paper are as follows:1.Summarizes the methods of road extraction from high-resolution remote sensing images and the current research status at home and abroad.And analyzed the basic theoretical knowledge of deep learning fully convolutional network model and Deeplab-V3 network model,which provides a theoretical basis for the improvement of the model.2.Due to the complexity and diversity of road and background features in remote sensing data,the accuracy of road extraction using deep learning methods is limited.Therefore,based on the U-Net network architecture design,a deep semantic segmentation model AS-Unet for road extraction from remote sensing images is implemented.The model is divided into two parts: encoder and decoder.The algorithm first adds a channel attention mechanism to the encoder,to filter the extracted rich lowlevel features,highlight target features,inhibit background noise interference,and improve the accuracy of deep and shallow information fusion.Secondly,considering the single size sensitivity issue of the network to the road targets in the images,a spatial pyramid pooling module is added after the last convolutional layer of the encoding network,to capture road features of different scales.Finally,a spatial attention mechanism is added to the decoder,to further perform learning of location relationship information,and filtering of relevant deep semantic filtering,and to improve the ability of feature map restoration.3.The production process of Gaofen-2 remote sensing image dataset is studied.After the original image is subjected to radiometric calibration,atmospheric correction,orthorectification,fusion and other steps,sample datasets such as training and testing are produced.Collected the Massachusetts road datasets and Deep Globe road datasets,and performed data enhancement operations on the three road datasets,which provided a data basis for the experiment.4.The proposed improved U-Net model of high-resolution remote sensing image road extraction methods were conducted on Massachusetts、Deep Globe and Gaofen-2 road datasets,and the results demonstrate that compared to semantic segmentation networks such as Seg Net and FCN,the AS-Unet network is preferable in terms of the evaluation indexes such as recall,precision,and F1 value.The designed AS-Unet network has satisfactory performance and higher segmentation accuracy,and boasts certain theoretical and practical application value.