Semi-supervised based active class selection for automatic identification of sub-kilometer craters

Counting craters is a fundamental task of planetary science, because it provides the only tool for measuring relative ages of planetary surfaces. In this paper, we combine active learning with semi-supervised learning to build an new semi-supervised active class selection system for crater detection from high resolution panchromatic planetary images. We propose the Semi-supervised Active Class Selection Algorithm to iteratively enrich an original small training set, without additional human labeling effort, to detect craters from a large volume of images. We propose two strategies to improve detection accuracy by integrating classification with exploration on unlabeled samples. The Majority Vote Strategy is used to automatically obtain class labels by exploiting unlabeled samples from test images. In the same time, the Active Stability Strategy is used to obtain an appropriate class distribution in the constructed training set by detecting unstable classes. By using those two strategies, we actively select test instances from test images into an existing small initial training set while re-learning the classifier in the mean time. The proposed algorithm is empirically evaluated on a large challenging Martian image, exhibiting a heavily cratered Martian terrain characterized by heterogeneous surface morphology. The experimental results demonstrate that the proposed approach achieves a higher accuracy than other existing approaches to a large extent.