There are more and more applications of change detection methods based on deep learning in high-resolution remote sensing images. However, downsampling and cropping strategies deployed to fit the GPU (Graphic Processing Unit) memory constraints on processing large-size remote sensing images often result in incomplete semantic information and loss of fine details. In this paper, a collaborative supervised network based on feature pyramids is proposed to enable the network to learn local and overall features from cropped and downsampled image blocks. In addition, a feature-sharing mechanism is introduced to fuse global features and local features. We evaluated the network on the LEVIR-CD (a remote sensing change detection dataset) and S2Looking (a building change detection dataset) by comparing it with some representative change detection networks. The comparison experiments show that the proposed network performs better in multi-scale change detection, with a 2.69% improvement in precision on LEVIR-CD, and 6.83% and 2.68% improvement in precision and recall on the S2Looking dataset, respectively.