Abstract: Traditional three-dimensional (3D) modeling techniques suffer from low data acquisition efficiency, insufficient robustness in novel-view synthesis, and limited generalizability to complex scenes when applied to underground coal mine environments characterized by low illumination, large illumination contrast, and weak texture. To address these limitations, a rapid and lightweight 3D digitization method for coal mine roadways, termed Coal Mine Panoramic Gaussian Splatting (CM-PanoGS), is proposed based on a single panoramic image and 3D Gaussian Splatting (3DGS). To accommodate the large illumination contrast of panoramic images, the Retinexformer algorithm is employed to adaptively enhance image illumination while preserving details in dark regions. Subsequently, a model pretrained on panoramic images is used for depth estimation. The enhanced RGB panoramic image is then integrated with the estimated depth map, and a 3DGS scene is rapidly generated through back-projection. For large-scale roadway scenes, panoramic images are sparsely acquired at selected stations, with each station being independently reconstructed as a local 3DGS scene. A rapid scene loading and unloading mechanism is further introduced to enable efficient switching among different stations. Experimental results demonstrate that CM-PanoGS substantially reduces the amount of data required for 3D modeling from hundreds of images to a single panoramic image per station, while allowing model training to be completed within seconds. In terms of the Structural Similarity Index Measure (SSIM), CM-PanoGS achieves performance comparable to that of representative methods, including conventional 3DGS and Neural Radiance Fields (NeRF). Its Peak Signal-to-Noise Ratio (PSNR) is also close to those of the comparison methods, whereas significantly lower Learned Perceptual Image Patch Similarity (LPIPS) values are obtained. By accepting a moderate loss in geometric accuracy, CM-PanoGS achieves a substantial improvement in modeling efficiency and realizes lightweight and rapid digitization of underground coal mine roadways.