Research On Hyperspectral Image Classification Method Based On Deep Learning

With the development of image technology,hyperspectral image has been paid more and more attention in the field of remote sensing.At present,hyperspectral images have been widely used in agricultural development,military detectives,geological surveys and weather forecasting.Hyperspectral image classification is the basis of hyperspectral image data processing,Because hyperspectral image is the image data with high dimension and large amount of data,which contains a lot of spatial information and spectral information,there is a high correlation between the data,resulting in a large number o f redundant data,so it brings a huge challenge to the subsequent accurate classification.With the rapid development of deep learning,many deep learning methods are applied to hyperspectral image classification.Deep learning method can extract the deep abstract spatial and spectral features of hyperspectral images,so how to extract and classify these features effectively has become the focus of current research.In this paper,we mainly studied the classification of hyperspectral images by combining descending and local Gabor convolutional neural networks.The specific research contents are as follows:(1)We propose a new classification method based on spatial spectral joint features to solve the problem of underutilization of hyperspectral image features.First,principal component analysis and linear discriminant analysis were used to reduce the dimensionality of hyperspectral image data.Secondly,the Gabor kernel is introduced to design a convolutional layer based on the Gabor kernel,the Local Gabor Convolutional Layer,then two new Local Gabor Convolutional Neural networks LGCNN-1 and LGCNN-2 are designed based on the LGC layer to extract the deep features of hyperspectral images.Finally,the Softmax classifier is used for classification.The experiment shows that the proposed method has better classification effect.(2)In order to make full use of the rich spatial and spectral features of hyperspectral images,a hyperspectral image classification method based on the multi-channel local Gabor convolutional neural network was proposed based on the research content(1).In this method,four local Gabor convolutional neural networks were simultaneously constructed to extract four representative deep spatial-spectral features of hyperspectral images,that is,features of 0,?/4,?/2 ? 3?/4directions,Then the features of these four directions are combined and finally classified by Softmax classifier.Experimental results show that the proposed method makes full use of the multi-direction features of hyperspectral images and impro ves the classification performance greatly.