Due to the complex underground environment of coal mine, the image collected by monitoring equipment appears fuzzy degradation under the influence of coal dust and water fog. Most of the existing image sharpening algorithms have problems such as excessive enhancement, detail loss and image darkening. A sharpening algorithm of coal mine dust image based on enhanced grid network is proposed to solve the above problems, which consists of three parts: pre-processing module, backbone module and image output module. First, a set of feature maps are generated by the pre-processing module as the input of the backbone module. Then, the feature maps are transformed by the backbone grid network based on attention to fully extract the features of different scales of the image. Finally, the image output module processes the fused feature information to output clear images. In the training process, the existing synthetic data set is used to train the network initially, and then the self-built data set is added to train the network twice. Experimental results show that compared with six representative sharpening algorithms such as the dark channel prior algorithm, the proposed algorithm has achieved varying degrees of improvement both subjectively and objectively, indicating that the proposed algorithm can effectively improve the clarity and visualization effect of downhole dust fog images.