Saliency and Tracking based Semi-supervised Learning for Orbiting Satellite Segmentation

The trajectory and boundary of an orbiting satellite are fundamental information for on-orbit repairing and manipulation by space robots. This task, however, is challenging owing to the freely and rapidly motion of on-orbiting satellites, the quickly varying background and the sudden change in illumination conditions. Traditional segmentation usually relies on a large annotated dataset and needs to be pre-trained for each target, which exhausts much time in both training and testing due to the large number and resolution of the images. In this paper, we proposed a STSS (Saliency and Tracking based Semi-supervised Learning for Segmentation) algorithm that provides the segmentation binary mask of target satellites at 12 frames per second without requirement of annotated data. Our method, STSS, improves the segmentation performance by generating a saliency map based semi-supervised on-line learning approach within the initial bounding box estimated by tracking. Experiment is evaluated on our generated dataset, which contains various challenges including variation in target, background and illumination condition.