Terrain identification is crucial for the safety of extraterrestrial surface exploration. Deep learning techniques have been applied to planetary exploration to strengthen rovers’ ability to identify terrain hazards. The scarcity of training data is a unique problem for deep learning applications for extraterrestrial surface exploration. To solve this problem, pretraining based on contrastive learning is considered an appropriate approach because contrastive learning can pretrain general representations using a large amount of unlabeled data, and these representations can be utilized by downstream tasks. This paper presents a terrain classifier for extraterrestrial surface exploration based on contrastive learning, which can complete terrain classification task with limited labeled data. A hypothesis is proposed that utilizing RGB images as 3-view data in the pretraining stage will help the model learn more general representations. Experimental results support this hypothesis and indicate that contrastive learning contribute to the improvement of the terrain classifier’s performance.