Change detection is an important research direction for remote sensing image application. Finding the change location automatically is the goal, which could provide useful information for disaster evaluation, disaster evaluation or urban development planning etc. In this paper, a Siamese feature fusion network is designed for change detection, which applies GSoP, a kind of attentional mechanism, to fuse the information of two feature extraction branches. The feature fusion strategy could build an information bridge which could also ensure the uniqueness of each branch, but also realize the information interaction of them to give more attention to important features during feature learning procedure. In the experimental section, many experiments were designed on many public datasets with some related methods. The results were shown that the proposed Siamese fusion network was efficient for change detection and had obvious advantage than some related methods.