Improving Contrastive Learning on Visually Homogeneous Mars Rover Images

Contrastive learning has recently demonstrated superior performance to supervised learning, despite requiring no training labels. We explore how contrastive learning can be applied to hundreds of thousands of unlabeled Mars terrain images, collected from the Mars rovers Curiosity and Perseverance, and from the Mars Reconnaissance Orbiter. Such methods are appealing since the vast majority of Mars images are unlabeled as manual annotation is labor intensive and requires extensive domain knowledge. Contrastive learning, however, assumes that any given pair of distinct images contain distinct semantic content, whereas with Mars images, pairs are far more likely to be semantically similar due to the lack of visual diversity on the planet's surface. This results in incorrect pseudo-labels being assigned to training instances, creating an increased rate of pairs which are treated as different classes when they are in fact the same (false negative pairs). In this study, we propose two approaches to resolve this: 1) an unsupervised deep clustering step on the Mars datasets, which identifies clusters of images containing similar semantic content and assigns their pseudo-labels accordingly, and 2) a simple approach which mixes data from different domains to increase visual diversity of the total training dataset. Both cases reduce the rate of false negative pairs, thus minimizing the rate in which the model is incorrectly penalized during contrastive training. These modified approaches remain fully unsupervised end-to-end. To evaluate their performance, we add a single linear layer trained to generate class predictions based on these contrastively-learned features, and demonstrate increased performance compared to supervised models; we observe an improvement in classification accuracy of 3.06% despite requiring only 10% of the labeled data.