Unsupervised Transformer Boundary Autoencoder Network for Hyperspectral Image Change Detection

In the field of remote sens., change detection is an important monitoring technology. However, effectively extracting feature still a challenge, especially with unsupervised method. To solve this problem, we proposed transformer boundary autoencoder network (UTBANet) in paper. UTBANet consists structure and spectral attention encoder part. addition to reconstructing hyperspectral images, also adds decoder branch for edge information. The designed module used extract features. First, global Then, can find maps reduce redundancy. Furthermore, reconstructs image information simultaneously through two decoders, which improve ability Our experiments demonstrate that significantly improves performance detection. Moreover, comparative show our method superior most existing methods.