DGCFNet: Dual Global Context Fusion Network for remote sensing image semantic segmentation.

Area of Science:

• Computer Vision
• Remote Sensing
• Artificial Intelligence

Background:

• Semantic segmentation of remote sensing images is challenging due to complex backgrounds and feature variations.
• Convolutional Neural Networks (CNNs) excel at local feature extraction but struggle with long-range dependencies.
• Transformers are effective at capturing global context via self-attention but may overlook local details.

Purpose of the Study:

• To propose a novel semantic segmentation network, Dual Global Context Fusion Network (DGCFNet), for high-precision remote sensing image analysis.
• To integrate the strengths of CNNs for local information and Transformers for global context.
• To enhance segmentation performance by addressing challenges like high inter-class similarity and intra-class variations.

Main Methods:

• Developed DGCFNet, an encoder-decoder network combining CNN and Transformer architectures.
• Introduced a dual-branch global extraction module to enhance Transformer's global context modeling while preserving local information.
• Incorporated a cross-level information interaction module and a feature interaction guided module to improve feature correlation and segmentation consistency.

Main Results:

• DGCFNet achieved superior segmentation performance on Vaihingen, Potsdam, and BLU datasets.
• The proposed method reached mean Intersection over Union (mIoU) scores of 82.20%, 83.84%, and 68.87% on the respective datasets.
• Experimental results demonstrate the effectiveness of the integrated local and global context modeling approach.

Conclusions:

• The proposed DGCFNet effectively addresses the challenges in remote sensing image semantic segmentation.
• The network's ability to fuse local and global contextual information leads to significant performance improvements.
• DGCFNet offers a promising approach for accurate and robust semantic segmentation in remote sensing applications.