With the development of deep learning in remote 1 sensing image change detection, the dependence of change 2 detection models on labeled data has become an important 3 problem. To make better use of the comparatively resource-saving 4 unlabeled data, the change detection method based on semi-5 supervised learning is worth further study. This paper proposes a 6 reliable contrastive learning method for semi-supervised remote 7 sensing image change detection. First, according to the task 8 characteristics of change detection, we design the contrastive 9 loss based on the changed areas to enhance the model's feature 10 extraction ability for changed objects. Then, to improve the 11 quality of pseudo labels in semi-supervised learning, we use the 12 uncertainty of unlabeled data to select reliable pseudo labels 13 for model training. Combining these methods, semi-supervised 14 change detection models can make full use of unlabeled data. 15 Extensive experiments on three widely used change detection 16 datasets demonstrate the effectiveness of the proposed method.

No abstract