Change detection is one of the most important and challenging tasks in remote sensing images processing. Deep learning has gradually become one of the most popular technologies in remote sensing image change detection. Recently, the success of self-attention mechanism in computer vision provides new ideas for change detection task. In this paper, a new method based on deep siamese network with self-attention mechanism for bi-temporal remote sensing image change detection is proposed. In order to obtain more powerful image features, the contextual transformer module is added into the feature extractor. In order to make full use of the low-level and the high-level features from the feature extractor, the multi-scale fusion strategies are applied to integrate features. Furthermore, the obtained image features are input into the transformer to get more refined pixel-level features. The proposed model is testified on CCD dataset, and the results demonstrate its effectiveness.