Study Of Hyperspectral Remote Sensing Image Classification Algorithm Based On Deep Belief Networks

Hyperspectral remote sensing image classification is one of the key technologies in the field of hyperspectral remote sensing.It is of great significance to urban planning and management,land use and so on.This thesis mainly studys hyperspectral remote sensing image classification based on deep belief networks.Furthermore,this thesis improves and innovates the dimensionality reduction method of hyperspectral remote sensing image and the training method of deep belief networks.In this thesis,two hyperspectral remote sensing image datasets,Salinas and PaviaU,are chosen as the experimental datasets of this thesis.First of all,this thesis studies two methods of principal component analysis and kernel principal component analysis.And combining with the advantages of low dimensional dimensionality of principal component analysis and the nonlinear relationship dealled with kernel principal component analysis well,this thesis proposes a fusion dimension reduction method with principal component analysis and kernel principal component analysis.Experimental results show that the proposed method has the better dimension reduction effect.Then,the deep belief networks method(DBN)is deeply studied,including the basic principle of the Restricted Boltzmann Machine,the network structure and training method of the DBN.In this thesis,an improved DBN training method is proposed to solve the problem that DBN method has a gradient disappearance in the pre-training phase,which leads to the decrease of the recognition precision.In this method,the original unsupervised learning process of the pre-training phase is changed to the first unsupervised learning and then supervised learning carring out to train the entire network.The experimental results show that the improved DBN method can effectively solve the problem of gradient disappearance in the pre-training process,so that the recognition accuracy increases with the increase of the number of hidden layers in the network.Finally,this thesis designs an improved DBN network.Then this thesis use the DBN network designed in this thesis to classify and test the two hyperspectral remote sensing image datasets of Salinas and PaviaU,and get the results of hyperspectral remote sensing image classification.The classification results are compared with BP neural network method,Support Vector Machine method and classical DBN method.The results show that the proposed method has the highest classification accuracy.