| With the technological development of remote sensing image sensors and emerging platform drones,the availability and popularity of urban high-resolution images have been greatly improved.The feature information contained in massive data has become urban planning,urban environmental change monitoring and urban The focus of research in three-dimensional modeling and other fields.Buildings and green spaces are an important part of urban feature information.The traditional methods of extracting and updating this information are generally based on artificial features and unsupervised machine learning algorithms.They are greatly disturbed by external conditions and have not been able to get rid of the low-level The dependence of features,therefore,how to accurately,quickly and automatically extract ground feature information from high-resolution images has become one of the current key research directions in this field.In recent years,with the development of deep learning technology,it has achieved good results in the field of computer vision.Therefore,on the basis of previous research,this article introduces deep learning technology to build an automated extraction algorithm that focuses on urban buildings and green spaces and other features;and divides the extraction process and methods according to urban features to complete the extraction Design and implementation of high-resolution urban image feature information interactive interface.The main research work of this paper is as follows:(1)Aiming at the problems that the current semantic segmentation network only uses part of the convolutional layer to produce the final output,the edge accuracy of the building targets in the high-resolution image is low,and the quality of the predicted image is not high.The scale feature extraction network BuildingNet,and the introduction of the hollow space pyramid pooling module,makes the network have a good feature extraction performance for the overall image.Secondly,an improved Lovasz loss function is designed to train the proposed network,which effectively improves the image quality of the extracted results.Finally,through comparative experiments and ablation experiments,it is verified that the network has higher performance in the accuracy and adaptability of building identification and extraction,and the process is highly automated,which can extract large-scale image building information.(2)Based on the idea of generating a confrontation network to perform image conversion tasks,an improved conditional generation confrontation network UAV-GAN is designed and applied in the field of image semantic segmentation,so as to achieve the goal of converting data images into segmented images.At the same time,use the drone tilt photography technology to obtain and produce high-resolution urban area models,and use tilt photography professional software to obtain complete orthophotos,and produce urban feature information data sets.The experiment is carried out in the data set produced,and the visual qualitative and quantitative results of the buildings and urban green spaces(including vegetation)in the data set images are obtained.(3)A high-resolution urban vehicle data set is established,and two semantic segmentation networks,BuildingNet and UAV-GAN,are applied to the recognition of urban small-scale targets,and the task of urban vehicle recognition based on the semantic segmentation model is realized.(4)Based on the algorithm model proposed above,a GUI interactive interface for extracting the feature information of high-resolution urban images is designed.This interactive interface integrates the semantic segmentation model proposed in this article,and integrates basic functional modules such as high-resolution image selection and opening,model invocation,segmentation and extraction,and result storage.At the same time,it adds the function of identifying the pixel position of the image segmentation result,which is accurate and convenient.Automatic extraction of urban feature objects. |
PDF Full Text Request