Multi-scale Feature Fusion Network with Transformer for Change Detection of Remote Sensing Images

With the increasing attention of the state to land resources, change detection is an important task in land resource management, cropland and forest protection and road damage detection. Although image difference-based methods can effectively detect changes between before and after images, they are not ideal when facing significant pseudo-changes. However, considering that the vision transformer (ViT) is difficult to accomplish the pixel-level semantic segmentation task and the transformer-UNet structure is too complex, we propose a multi-scale feature fusion network (MSFF-net). It combines the feature information of two images and improves the accuracy of prediction results. The experimental results show that the change detection performance of MSFF-net is significantly better than several current classical methods.