Corona discharge images collected with night-type ultraviolet cameras are affected by the photographer’s environment and the degree of partial discharge, and the color of the discharge area is not only close to the background but also overlaps with the background, which makes it difficult to automatically segment corona discharge. This paper proposes a coarse-to-fine corona discharge ultraviolet (UV) image segmentation method. First, a deep-learning semantic segmentation model was constructed, and rough segmentation results of the corona discharge were obtained using a trained Unet network. Second, the UV image of the discharge region was converted into a gray image, and the rough segmentation result was accurately segmented based on the Otsu threshold segmentation method with foreground weighting. A total of 426 samples were tested, and all the corona discharge regions in the sample images were segmented using the proposed method. The error between the segmented discharge regions and the true value was close to 0. The proposed corona discharge segmentation method provides accurate data sources for the quantification and evaluation of corona discharges.