Applications of transformer in remote sensing for image scene classification, semantic segmentation, and change detection

In recent years, with the proposal and widespread application of Transformer, which was originally applied in natural language processing (NLP), Transformer has become the preferred model for NLP tasks. Based on Transformer’s great success in the NLP task, this model has also become a hot research direction in the field of computer vision (CV). In the field of CV, models based on convolutional neural network (CNN) architecture have taken a dominant position. CNN adopts hierarchical feature representation, which can effectively extract deep features, but its local nature limits the processing of long-range dependencies. Transformer, based on a self-attention mechanism, is capable of effectively understanding the global context. The successful application of Transformers in visual tasks demonstrates their enormous potential in the CV field. Inspired by this, the remote sensing community has also made extensive exploration of the application of Transformers. This research involves a variety of remote sensing image tasks, and Transformer has shown excellent performance in these tasks. This paper briefly describes the classic research of several remote sensing image tasks, aiming to elaborate on the application status of Transformers in the field of remote sensing. These classic research cases are introduced in detail, and the research results are compared with the traditional methods. Finally, the problems to be solved and the feasible research directions are discussed.