For cracks in small bridges, a segmentation method is proposed based on a generative adversarial network. This method introduces a segmental branch into the discriminator structure and combines the generative confrontation network with the semantic segmentation network. In addition, the method is capable of super-resolution image reconstruction and segmentation. To solve the problem of small-bridge-crack segmentation, this method transforms low-resolution small-bridge-crack images into super-resolution coarse-bridge-crack images, which are then segmented. The experimental results show that the proposed method facilitates the identification of small-bridge-crack and its segmentation is accurate. Compared with the traditional segmentation method, the recall rate and mean intersection over union of this method are improved by 6% and 10%, respectively.