Class-Guided Swin Transformer for Semantic Segmentation of Remote Sensing Imagery

Semantic segmentation of remote sensing images plays a crucial role in wide variety practical applications, including land cover mapping, environmental protection, and economic assessment. In the last decade, convolutional neural network (CNN) is mainstream deep learning-based method semantic segmentation. Compared with conventional methods, CNN-based methods learn features automatically, thereby achieving strong representation capability. However, local receptive field convolution operation limits from capturing long-range dependencies. contrast, Vision Transformer (ViT) demonstrates its great potential modeling dependencies obtains superior results Inspired by this, this letter, we propose class-guided Swin (CG-Swin) for images. Specifically, adopt Transformer-based encoder–decoder structure, which introduces backbone as encoder designs block to construct decoder. The experimental on ISPRS Vaihingen Potsdam datasets demonstrate significant breakthrough proposed over ten benchmarks, outperforming both advanced recent approaches.