Abstract: Intelligent unmanned mining of coal mine is the technological development trend of green, intelligent and safe mining of coal resources. High-resolution mine images can provide key technical support for intelligent unmanned mining of coal mine. Aiming at the degradation phenomenon of mine images, in order to improve the resolution of mine images, a super-resolution reconstruction method mine image based on generative adversarial network is proposed. Based on SRGAN, this method improves the network structure and loss function. First, two 5×5 convolutional layers are used in the low-level feature extraction layer and reconstruction layer of the generator, and non-linearity is added after each convolutional layer of the low-level feature extraction layer, and the high-level feature extraction layer adopts the residual structure, and the sub-pixel convolutional layer is cascaded to achieve super-resolution reconstruction of different multiples. Secondly, the Wasserstein distance is used to improve the loss function, and the Sigmoid of the output layer of the discriminator is removed. The RMSProp method is used to improve the network optimization method, which can make the model training more stable. Finally, the trained generator model is used to reconstruct the mine images by 2 times and 4 times super-resolution, and subjective visual analysis and objective evaluation were carried out on the experimental results. The results show that compared with the traditional bicubic interpolation, SRCNN, and SRGAN, when the scaling factor is 4, the peak signal-to-noise ratio of the proposed method is increased by 2.68, 1.50, and 1.59 dB; the structural similarity is increased by 0.0334, 0.0048 and 0.0061. The method proposed achieves better reconstruction results in terms of subjective vision, peak signal-to-noise ratio and structural similarity. The improved method can reconstruct clear texture and detail information, and its overall performance is better than several other methods.