Change detection is an important task in remote sensing (RS) image analysis. With the development of deep learning and the increase of RS data, there are more and more change detection methods based on supervised learning. In this paper, we improve the semantic segmentation network UNet++ and propose a fully convolutional siamese network (Siam-NestedUNet) for change detection. We combine three types of siamese structures with UNet++ respectively to explore the impact of siamese structures on the change detection task under the condition of a backbone network with strong feature extraction capabilities. In addition, for the characteristics of multiple outputs in Siam-NestedUNet, we design a set of experiments to explore the importance level of the output at different semantic levels. According to the experimental results, our method improves greatly on a number of indicators, including precision, recall, F1-Score and overall accuracy, and has better performance than other SOTA change detection methods. Our implementation will be released at https://github.com/likyoo/Siam-NestedUNet.