Research On Classification Method Of Remote Sensing Image Based On Convolutional Neural Network

With the rapid development of remote sensing observation technology,remote sensing image data has increased dramatically,and accurate identification of remote sensing images has become more and more important.Remote sensing image classification not only plays an irreplaceable role in detecting strategic targets and precision strike targets in the military,but also has a wide range of applications in production and life such as natural disaster monitoring,land use planning,and weather forecasting.However,the remote sensing image has a large field of view,a large number of scene categories,a large number of objects,a high degree of scene mixing,and a rich amount of information.These factors not only cause great difficulties in remote sensing image classification,but also cause some problems in military target recognition.The features extracted by traditional remote sensing image classification algorithms have certain limitations.Ignoring some object information in the image,it is impossible to accurately describe the remote sensing image.Therefore,in order to more accurately describe remote sensing images,this thesis studies a remote sensing image classification algorithm based on convolutional neural network Convolutio nal Neural Network.The main research work of this thesis is as follows:（1）This article analyzes the remote sensing image features extracted from each layer of the network,and fuses the features of each layer of the network to obtain more accurate remote sensing image features information.This method can realize the end-to-end training of the network.After each convolution layer’s feature is processed by l₂ normalization,it can be fused with other layer’s features through Eltwise layer to obtain the whole image global feature information.Experimental results show that the algorithm studied in this thesis is effective and the accuracy of remote sensing image classification is significantly improved.（2）This thesis studies a CNN remote sensing image recognition method based on object reasoning by analyzing the close relationship between objects and scenes,.VGGNet-16 is used to obtain the prior scene information of the remote sensing image,and the relationship between the objects in the image and the scene is established by counting a large amount of remote sensing image data.The object detection algorithm is used to detect multiple objects in the image.Many types of objects use the Bayesian criterion to infer the types of remote sensing images.Experimental results show that compared with other algorithmic networks,the method studied in this thesis has achieved good results on three public remote sensing data sets,verifying the inference effect of objects in the image on the scene.By analyzing some shortcomings of existing remote sensing image classification algorithms,this thesis studies two methods to improve the accuracy of remote sensing image classification.After a lot of experimental verification,the method studies in this thesis is effective,and improves the accuracy of remote sensing image classification.