Design And Development Of Remote Sensing Image Intelligent Interpretation Platform Based On On Deep Learning

With the rapid development of aerospace industry and micro satellite remote sensing image technology,the data scale of remote sensing image is increasing,and the image quality is also getting higher and higher.The time cost and labor cost of traditional image processing technology in processing high-quality large-scale remote sensing images are too high,whereas in recent years,deep learning has been the latest field of computer vision.With the wide attention and rapid development of society,semantic segmentation technology based on deep learning,target recognition technology,scene classification technology have become the focus of computing.It is a hot research direction in the field of machine vision and deep learning.This paper compares the different properties of highresolution image and ordinary image,selects different deep learning models for different tasks,and applies them to the analysis platform of high-resolution satellite remote sensing image,which improves the efficiency of remote sensing image interpretation through platform automation.On this basis,this paper designs and develops an intelligent remote sensing interpretation platform which can automatically complete different parsing tasks according to the requirements.The main contents of this paper are as follows:(1)Establish corresponding remote sensing image data sets for different remote sensing image interpretation tasks.In semantic segmentation and scene tasks,for high-resolution remote sensing images and manually labeled graphical vector files,through the development of corresponding matching algorithms to turn it into a truth map.In the target recognition task,for high-resolution remote sensing images and manually labeled targets,text files of target size and location information are generated.Finally,a large-scale remote sensing image data set that can be used for various model training is constructed.(2)This paper analyzes and compares the classic target recognition network Fast-RCNN,Faster-RCNN,R2CNN;classic semantic segmentation task network,FCN,U-Net,DeeplabV3+;classic scene classification network,DFL-CNN,GoogleNet.The paper makes network selection according to the needs of remote sensing image interpretation,and carries out the corresponding training through different data sets to make the accuracy of the model reach the level of practical application and improve the interpretation speed of remote sensing images.(3)This paper designs and implements an intelligent remote sensing interpretation platform based on semantic segmentation,target recognition,and scene classification.The TensorFlow framework is used for model training and solidification,and the remote sensing interpretation platform is realized through front-end and back-end frameworks such as Django and Vue.Target recognition for aircraft,ships and other targets,semantic segmentation for waters and buildings,and scene classification for farmland,woodland,buildings,and waters.The platform provides functions such as data upload,data preview,multi-task automatic interpretation,result preview and download.You can also view historical tasks and download the results of historical tasks.Finally,this article carried out a system test and interface display on the intelligent interpretation platform.