High-Order Semantic Decoupling Network for Remote Sensing Image Semantic Segmentation

I. Overview

Dense remote-sensing segmentation must distinguish visually similar regions while preserving relationships across large spatial extents. This project decouples high-order semantic interactions so the model can reason about complementary contextual structures without collapsing them into one feature stream.

II. Key Contributions

• Models high-order dependencies among semantic regions in remote-sensing scenes.
• Separates complementary contextual relationships before feature aggregation.
• Recombines decoupled semantics for pixel-level classification.

III. Methodology

A convolutional feature extractor supplies multi-level visual representations. The semantic-decoupling stages shown in the supplied overview build and separate higher-order contextual relationships, then fuse the refined features into a dense segmentation prediction.

IV. Research Focus

The method targets scenes where local appearance alone is ambiguous and broader semantic context is needed to assign consistent land-cover labels.

@article{zheng2023high,
  title={High-order semantic decoupling network for remote sensing image semantic segmentation},
  author={Zheng, Chengyu and Nie, Jie and Wang, Zhaoxin and Song, Ning and Wang, Jingyu and Wei, Zhiqiang},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  volume={61},
  pages={1--15},
  year={2023},
  publisher={IEEE}
}