Semantic Comprehension Of Remote Sensing Images Based On Deep Learning

In recent years,aerospace technology has developed rapidly.In the field of aerospace,it is an important task for researchers to extract effective information from remote sensing images.Because of the large amount of remote sensing data,the variety and complexity of remote sensing images,traditional methods are more and more difficult to adapt to massive data.Traditional methods rely on the characteristics of artificial design,which is very time-consuming.This paper mainly studies the application of deep learning algorithm in the field of aerospace to process massive remote sensing images and extract effective information.Firstly,the paper implements and improves a deep learning algorithm,and achieves image semantics segmentation for four kinds of target objects in remote sensing images.The paper validates the availability and the accuracy of the network model,and evaluates the segmentation results.The model can classify remote sensing images at the pixel level,which provides a certain basis for the follow-up study of remote sensing images.Secondly,this paper focuses on the analysis of image high-level semantics extraction system.The segmented images are as the processing object.We use binarization,filtering operation,open-close operation,threshold setting and other methods,as well as connected domain algorithm,to count the number of target objects and detect target objects.The forest coverage of each remote sensing image is obtained by calculating the ratio of the total number of pixels with "tree" class to the total number of pixels per image.Different from traditional image interpretation methods,the high-level information extraction system designed in this paper can not only compute the number of buildings in the scene,but also calculate forest coverage.Finally,this paper combines graph database with engineering application.It uses high-level semantic information to model and retrieve effective data information.A real-time remote sensing image processing system based on deep learning is proposed.It solves the problem that massive data can not be processed on-orbit due to the limited bandwidth of the space-ground link.The system has great application value.It lays a foundation for future research on road localization and information retrieval.