Related Experiment Video
Updated: Sep 13, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
DBRSNet: a dual-branch remote sensing image segmentation model based on feature interaction and multi-scale feature
Yong Ji1, Wenbin Shi2, Jingsheng Lei1
1School of Computer Science and Technology, Zhejiang University of Science and Technology, Hangzhou, 310023, China.
Scientific Reports
|July 30, 2025
Summary
DBRSNet enhances remote sensing semantic segmentation by integrating CNN and Transformer features. This dual-branch approach improves detail retention and global context modeling for better land cover classification.
Area of Science:
- Remote Sensing
- Computer Vision
- Geospatial Information Science
Background:
- High-resolution remote sensing images contain rich geographical data, but accurate interpretation requires precise semantic segmentation.
- Current methods struggle with fine-grained local details and global context, leading to boundary fragmentation and small object degradation.
Purpose of the Study:
- To develop an advanced dual-branch framework, DBRSNet, for improved remote sensing semantic segmentation.
- To address limitations in detail retention and contextual modeling in existing approaches.
Main Methods:
- Introduced DBRSNet, a dual-branch framework integrating feature interaction and multi-scale feature fusion.
- Employed the Feature-Guided Selection Module (FGSM) to merge CNN and Transformer features.
- Utilized the Convolutional Attention Integration Module (CAIM) to enhance global dependencies and spectral correlations.
Main Results:
- DBRSNet demonstrated superior performance compared to 14 state-of-the-art models on ISPRS Vaihingen and Potsdam datasets.
- The framework achieved higher accuracy across all assessment metrics.
Conclusions:
- DBRSNet offers a robust solution for remote sensing semantic segmentation, outperforming existing methods.
- The proposed architecture effectively captures both local details and global context for complex scenes.
