Hyperspectral Image Classification Based On Generative Adversarial Networks

Hyperspectral image classification is an important branch of remote sensing image processing.Traditional classification methods can not make full use of the spatial-spectrum characteristics of hyperspectral images,which results in low classification accuracy.The hyperspectral image classification methods based on deep learning can mine the hidden information of data,and extract features that are more conducive to image classification.Extracting more abundant features from limited samples and integrating these features to improve the classification accuracy has become a hot research direction.Combining with spectral and spatial features,this paper proposes two hyperspectral image classification methods based on generative adversarial networks.The main contributions of this paper are as follows:In order to solve the problem of insufficient feature utilization in hyperspectral image classification,a hyperspectral image classification method based on generative adversarial networks is proposed.Based on the correlation between the spatial and spectral domains of hyperspectral image,the generative adversarial networks is used to mine deep features,and generate highly discriminable hyperspectral image,which improves the classification accuracy.To overcome the shortcomings of insufficient utilization of spatial information,this paper proposes a feature extraction method based on pixel neighborhood and generative adversarial networks.Firstly,deep convolutional generative adversarial networks are trained by the neighborhood image of pixels,and the trained discriminant network is taken as the feature extractor of the neighborhood image.Finally,the extracted neighborhood features are combined with the spectral characteristics of pixels to improve the classification accuracy.