RS DINO: A Novel Panoptic Segmentation Algorithm for High Resolution Remote Sensing Images

To address the suboptimal performance of current panoptic segmentation models in remote sensing image processing, this paper presents RS DINO(Remote Sensing Mask DETR with Improved Denoising Anchor Boxes), a novel algorithm for panoptic segmentation of multi-spectral high-resolution remote sensing images. First, a batch attention module was designed to calculate relationships between image patches, to effectively utilize long-range contexts in the image. Furthermore, a Channel Attention Module was used to extract interdependencies between multiple channels in remote sensing images and fuse spectral features of land cover targets. Finally, based on a high-resolution remote sensing panoptic segmentation dataset, typical normal image panoptic segmentation algorithms were compared with RS DINO through experimental evaluation. Results show that RS DINO achieves the best panoptic segmentation performance and has better applicability on multi-spectral high-resolution remote sensing images.