Application Of Multi-dimensional Classification Technology In Remote Sensing Image

With the development of remote sensing technology,more and more high spatial resolution images are applied to all walks of life.The emergence and rapid development of UAVs make the acquisition of remote sensing images more and more convenient.Remote sensing image classification,as an indispensable process in the application of remote sensing in practice,has become a hot issue.Compared with general images,remote sensing images contain spectral information,and rich spectral information provides great help for classification tasks.In recent years,with the progress of science and technology,deep learning has developed rapidly and has been well applied to remote sensing image classification tasks.In this thesis,aiming at some problems existing in the classification task of remote sensing images,the application of deep learning in remote sensing image classification task is further studied based on previous studies.The main contents are as follows :(1)Aiming at the problem of insufficient feature extraction and high computational cost in remote sensing image classification,this thesis proposes a remote sensing image classification method based on parallel 3D-2D-1D CNN.The model uses one-dimensional convolution,twodimensional convolution and three-dimensional convolution to extract the spectral information,spatial information and spatial-spectral information of remote sensing images,and fully excavates the characteristics of the combination of space and spectrum of remote sensing images.Then,the feature maps of three dimensions are fused to obtain the feature maps that fully contain the spatial-spectral characteristics of remote sensing images.Finally,the remote sensing images are classified by the classifier.The model solves the problem of insufficient feature extraction of one-dimensional convolution and two-dimensional convolution and greatly alleviates the problem of the high computational cost of three-dimensional convolution.Comparative experiments were conducted on Indian Pines,Pavia Center and Pavia University datasets to compare the proposed model with the four traditional models.The proposed model obtained the optimal results,and the overall classification accuracy reached 99.210 %,99.755 %and 99.770 %,respectively.The experimental results fully demonstrate the superiority of the proposed model.(2)In the deep learning network model,with the increase of network depth,it is easy to produce gradient disappearance and gradient explosion,which will greatly affect the classification effect of the model.The parallel 3D-2D-1D CNN classification model proposed in this thesis is also prone to precision saturation with the increase of iterations.To solve the above problems,this thesis further improves the algorithm model and proposes a remote sensing image classification model integrating residual network and parallel 3D-2D-1D CNN.The residual network is extended to one-dimensional convolution and three-dimensional convolution,making full use of the advantages of the residual network.Through comparative experiments,the classification accuracy of parallel 3D-2D-1D R-CNN is improved compared with parallel 3D-2D-1D CNN on Indian Pines,Pavia Center and Pavia University datasets,which fully proves the advantages of the model.