Research On Spatial Spectrum Feature Extraction And Classification Method Of Hyperspectral Remote Sensing Image Based On Densely Connected Network

With the continuous development of hyperspectral remote sensing technology,the dimensionality and resolution of hyperspectral remote sensing data continue to increase,and the remote sensing data has shown an explosive growth.On the one hand,it brings a wealth of data sources for hyperspectral remote sensing image processing,on the other hand,it also increases the difficulty of hyperspectral remote sensing image processing due to the high dimensionality and redundant information of hyperspectral remote sensing data.Therefore,solving data redundancy and extracting deep features are the key and difficult points of hyperspectral remote sensing image processing.With the continuous development of deep learning,many deep learning methods have been applied to feature extraction and classification of hyperspectral remote sensing images.Deep learning methods can extract the deep spatial and spectral features of hyperspectral remote sensing images.How to effectively extract and make full use of these features is very important.This paper combines dense connection network and spatial spectrum fusion method to extract and classify the features from hyperspectral remote sensing images.The research work is as follows:（1）In the data preprocessing stage,the linear discriminant analysis method is used to reduce the dimension of hyperspectral remote sensing image and remove some redundant band information by taking advantage of the strong correlation between bands of hyperspectral remote sensing image;（2）Feature extraction is divided into two stages.The first stage separately extracts the spectral features and spatial features of the image,and uses the two-dimensional convolutional neural network to extract the spatial features,the main purpose is to reduce the amount of data.Spectral features are extracted by using 3D convolutional network.The second stage is to use the three-dimensional convolutional neural network to fuse the features extracted in the first stage,and then extract the deep features,and finally use softmax to classify the image.In the feature extraction network,the deconvolution layer is added to improve the convergence speed of the network by using the feature reconstruction of deconvolution;（3）On the basis of empty spectrum feature extraction of hyperspectral remote sensing image,a dense connection module is added to fully learn the deep features of the image and reuse the deep features by using the dense connection module.At present,the feature information of the image is fully utilized without increasing the number of network layers.This paper mainly studies the feature extraction and classification methods of hyperspectral remote sensing images from two aspects,the experimental results verify the effectiveness of this method.