The existing hyperspectral image generative adversarial network(GAN) classification algorithm cannot fully extract spectral and spatial-spectral features, which leads to the degradation of hyperspectral image classification accuracy. To resolve this issue, this study proposes a hyperspectral image classification algorithm based on a two-channel GAN. Improved one- and two-dimensional GAN classification frameworks are used to extract complete spectral and spatial-spectral features, respectively. Those features are nonlinearly fused to form a more comprehensive spatial-spectral features for classification. The experiments on two commonly used hyperspectral image datasets show that the proposed algorithm achieves the best classification accuracy; further, the results verify the effectiveness and advantages of the proposed algorithm.