Abstract: In order to solve the problem of insufficient feature utilization in hyperspectral image classification, a hyperspectral image classification method based on Generative Adversarial Networks（GANs） is proposed. According to the correlation between the spatial and spectral domains of hyperspectral image, the method of GANs is used to mine its deep features and generate highly discriminable hyperspectral image. The generated hyperspectral images are classified by Support Vector Machine（SVM）. The method is verified on two sets of hyperspectral data, the results show that this method can learn better generative model with a small number of spectral bands, and achieve much better performance in land-cover classification.

Key words: hyperspectral image classification, Generative Adversarial Networks（GANs）, feature mining

摘要： 针对高光谱图像分类领域中特征利用不足的问题，提出了一种基于生成对抗网络（Generative Adversarial Networks，GANs）的高光谱图像分类方法。根据高光谱图像空间域和光谱域的相关性，利用GANs方法，挖掘其深层特征，生成可分性更高的高光谱图像，并通过支持向量机（Support Vector Machine，SVM）对生成的高光谱图像进行分类。使用两组高光谱数据进行实验，结果表明，该方法能够在少量高光谱波段的情况下，对抗学习到较好的生成模型，使得生成的高光谱图像在地物分类实验中具有更高的分类精度。

关键词: 高光谱图像分类, 生成对抗网络（GANs）, 特征挖掘