Medium Resolution Image Classification And Application Based On Deep Learning

Due to the development of remote sensing technology and satellite technology in recent years,the way to obtain a large number of remote sensing images is becoming more and more simple.Compared with high-resolution remote sensing images,medium resolution remote sensing images are often easier to obtain and more commonly used,but how to quickly and effectively extract the information contained in the remote sensing data is a challenge for researchers.Remote sensing image classification is the basis of image segmentation and target detection,which is widely used in ecological environmental protection,urban planning and other fields.The classification technology has developed from visual interpretation to computer classification.In recent years,classification methods have gradually developed in the direction of machine learning.Deep learning is a machine learning method of deep neural network based on the idea of human brain learning.In this paper,through the research on the theory of deep learning and convolutional neural network and the introduction of some basic network models,the classification and application of medium-resolution images are studied.Firstly,the classification technology of remote sensing images at home and abroad,the development of machine learning and deep learning,and the related research of land use in recent years are introduced.Then,the relevant principles of the research methods applied in this paper are introduced,and the process of remote sensing image scene classification is explained.This study takes the Xiongan New Area from 2016 to 2022 as the test area,uses the medium-resolution data of the Gaofen-1 WFV field of view as the data source,and the GE image as the user accuracy evaluation basis.The 2016 mid-resolution data of Xiongan New Area are classified by traditional classification methods,machine learning classification methods,and deep learning classification methods.The results are gradually analyzed from the individual classified samples to the whole,and the urban area change of Xiongan New Area from 2016 to 2022 is summarized.The main conclusions are as follows:(1)For medium resolution remote sensing image data,the accuracy of 2DCNN convolutional neural network classification is improved by 8.1% compared with maximum likelihood method and 4.8% compared with machine learning SVM classification method.The accuracy evaluation uses high-resolution GE image as the sample selection of the training area,and then evaluates the classified results more precisely,making the classification accuracy more convincing.Compared with the confusion matrix generated by itself,the accuracy evaluation is higher and more delicate;(2)From 2016 to 2022,different types of land use area in Xiongan New Area changed.It can be concluded that the degree of change was road > building >vegetation > water body,in which the area of road,building and water body increased while the area of vegetation decreased,indicating that the growth area was basically from the transformation of vegetation area.(3)The construction area of Xiongan New Area has increased significantly in recent years,with the proportion of the construction area increasing from 15.56% in2016 to 19.44% in 2022.The change index of roads and building land in built-up areas was observed to be the lowest at 1.21% during 2018-2019,and the largest at 9.33%during 2019-2020.