Abstract: Image acquisition in underground coal mining faces is often compromised by uneven illumination, coal dust, and water mist, resulting in low resolution, missing details, and poor texture clarity.These issues hinder the effectiveness of technologies like remote monitoring and machine vision.To address this, we propose a super-resolution generative adversarial network enhanced by a residual hybrid attention mechanism.By incorporating residuals in the generator's feature extraction process, we enhance the network's depth and learning effect.The introduction of channel and spatial attention mechanisms improves the recovery of texture features and local details at image edges.Our optimized loss function ensures that reconstructed images closely resemble real images in terms of high-level semantics and subjective quality, retaining detailed textures and avoiding blurring from averaging.The discriminator recognizes reconstructed images, enabling adversarial training between the generator and discriminator to refine the model using small sample datasets.Experiments demonstrate that our method significantly improves image quality in coal mining face equipment datasets, achieving a 4-fold scaling enhancement.Compared to existing algorithms, our approach increases the peak signal-to-noise ratio (PSNR) by an average of 8.57% and improves structural similarity (SSIM) by an average of 1.43%.