基于非监督对比学习的火星地形特征提取方法

Intelligent surface terrain recognition of Martian is significant for the autonomous exploration of Mars rovers. At present, the methods used for feature extraction of Martian terrain images are mainly divided into two categories: traditional shallow visual feature extraction and deep feature extraction based on supervised learning. However, these methods tend to lose image information and require a large amount of labeled data, which are key problems to be solved. A Martian terrain feature recognition method based on unsupervised contrastive learning was proposed. By establishing the image dictionary dataset, a single image was compared with other images in the dictionary dataset by using two groups of neural networks, namely “query” and “encode” . Then, the similarity function was used as the loss function to train the network, so as to realize the feature recognition of Martian terrain images. The proposed method could also recognize new types of terrain images outside the training dataset and showed superior performance in subsequent recognition and classification tasks. Simulation results show that the recognition accuracy of the proposed method is 85.4%, and the recognition accuracy of new terrain images is 84.5%.