Research On Remote Sensing Image Depth Estimation And Information Management System

Remote sensing can help people understand the world more intuitively.Through remote sensing technology,researchers can monitor objects on the earth from the air,which is of great significance to resource exploration,ground feature monitoring,and navigation.Deep learning is a new method for information extraction from remote sensing data,and it is also a tool for processing large-scale remote sensing data.So far,the combination of remote sensing technology and deep learning has achieved remarkable results,improving the accuracy of classification tasks,semantic segmentation tasks,target detection tasks,change detection tasks and many other tasks.In this paper,the depth estimation task which has attracted much attention in the field of vision is transferred to the field of remote sensing,and its practical application is analyzed and studied,and some new explorations are made.On this basis,two kinds of remote sensing image data enhancement ideas at the scene semantic level are proposed,which are a zeroparallax method based on depth information and semantic information and a low-parallax method that simulates the 3D-ken-burns effect,and achieved good results.Finally,an management system for remote sensing images is constructed by synthesizing some of the existing and commonly used remote sensing technologies.The main research contents are as follows:1.A monocular depth estimation algorithm for remote sensing images based on perceptual enhancement network is proposed,which can estimate the depth information(that is,the elevation information)in a single optical remote sensing image,which is an important and important part in photogrammetry and remote sensing.Challenging mission.Aiming at the problems of a large number of complex noises,target details,and information loss in remote sensing images,a perceptual enhancement loss is proposed,which enables the network to obtain supervision from semantic information,thereby making the depth estimation result smoother.2.Based on the depth estimation algorithm,a data augmentation method for remote sensing images is proposed,which can perform scene-level data enhancement on remote sensing data.This method utilizes the depth information of remote sensing images to realize the distortion of the image perspective,thereby simulating more effective samples.Compared with the commonly used data enhancement methods,data enhancement based on depth estimation utilizes the information of the third dimension of the sample,and has achieved good results in classification and segmentation tasks in multiple remote sensing scenes.3.A remote sensing information management system is constructed.The system can set up a variety of data sources to support the addition,deletion,modification and search of remote sensing images,image search and direct extraction of common features.The system also integrates intelligent interpretation strategies common in most remote sensing fields,supports the classification,detection and segmentation of remote sensing images,and optimizes the performance of large-scale image usage scenarios to achieve a balance between accuracy and efficiency.