The door detection network is an important way of the autonomous boarding bridge detects the door. However, in the scenario where the door image has complex painting, the detection accuracy, recall rate and mAP of the door detection network for the door image are greatly reduced. To address this problem, a door image painting augmentation method based on generative adversarial network is proposed to increase the number of image samples with complex painting in the door detection network training set, thereby improving the detection accuracy, recall rate and mAP of the door detection network in complex paint scenes. The paint augmentation of the door image dataset is achieved by edge fusion of the edge image of the image to be augmented and the edge image of the painting image, and then coloring through the trained generative adversarial network. Experiments show that compared with the traditional basic augmentation methods, the door image painting augmentation method based on the generative adversarial network has a higher improvement in the detection accuracy, recall rate and mAP of Yolov5s.