With the development of deep learning technology, some achievements have been made in the field of change detection. However, there are still problems such as inaccurate detection of the edges of the change region and incomplete detection of the interior of the change region in the existing change detection methods. In view of this, this paper proposes a change detection method that combines multi-level features and global feature, in which a feature extraction network with dense connection of features between different levels is designed based on the Siamese network and the encoder-decoder architecture to fully extract and fuse features from different levels. For the fused features, this paper also designs a global feature modelling module to model their global context information. Moreover, a difference feature enhancement module is embedded between the encoder and the decoder to strengthen the learning of difference features of bi-temporal images. The proposed method is compared with some mainstream methods on the large-scale public datasets CDD and SYSU-CD through experiments, and the results show that the proposed method has good performance on both datasets.