Research Of Hyperspectral Image Classification Method Based On Deep Learning Algorithm

Hyperspectral image remote sensing technology can simultaneously acquire spatial and spectral information of the imaging region by combining spectroscopy and imaging technology,the formed hyperspectral image(HSI)has the characteristics of image-spectrum merging and high resolution.However,too high spectral dimension of images and limited number of labeled samples incurred great pressure to hyperspectral image classification(HSIC).In recent years,deep learning has been widely used in HSIC because of its powerful feature extraction and target classification ability.In view of this,the HSIC method based on deep learning method is studied in this paper.Aiming at the problem of image information loss caused by continuous pooling operation in the traditional convolutional neural network structure,a HSIC method based on dual channel dilated convolution neural network is proposed.The method combines the dilated convolution with the convolutional neural network,and extracted the spatial and spectral features of the image by two-channel method,after fused them by a weighted fusion method,the support vector machine is used for classification.The method can expand the receptive field of the filter while maintaining the image resolution,effectively avoid the loss of image information,and improve the classification accuracy.To solve the problem of limited number of labeled samples in HSI,this paper proposed a HSIC method based on unsupervised feature extraction.After the original image is dimensionally reduced by PCA,the first principal component of the image is selected as the label of the training sample,the unsupervised training is performed on the dilated convolutional neural network,the spatial feature and spectral feature of the image are extracted respectively by combining with the stack autoencoder,after fused them by a weighted fusion method,the support vector machine is used for classification.In order to evaluate the classification performance of the proposed classification method,experiments were conducted on two public HSI data sets and compares them with some existing classification methods.Experimental results show that the proposed method has better classification performance and has certain performance advantages in the case of limited number of samples.