A Double Res-UNET with Attention Mechanism for Change Detection of Remote Sensing Images

Remote sensing image change detection plays a great important role in disaster assessment and urban planning. Currently, most mainstream model is to build encoder-decoder models to detect some change regions of two different bitemporal images. In this paper, we develop a double Res-UNET with attention mechanism, which consists of two Res-UNETs and attention mechanism is designed to extract more shallow-level change features. Skip connections can be used to reduce training parameters. Extensive experimental results show that our model has better performance than several existing models, and training time is lower.