Dense Attention Fluid Network for Salient Object Detection in Optical Remote Sensing Images.

Area of Science:

• Computer Vision
• Remote Sensing
• Artificial Intelligence

Background:

• Salient object detection (SOD) is advanced for natural scenes but challenging for optical remote sensing images (RSIs).
• Existing methods struggle with the unique characteristics of RSIs, necessitating new approaches.

Purpose of the Study:

• To propose an end-to-end Dense Attention Fluid Network (DAFNet) for salient object detection in optical RSIs.
• To introduce a novel Global Context-aware Attention (GCA) module to capture long-range semantic context.

Main Methods:

• Developed the Dense Attention Fluid Network (DAFNet) integrating a Global Context-aware Attention (GCA) module.
• The GCA module features global feature aggregation and a cascaded pyramid attention mechanism.
• Constructed a new, large-scale optical RSI dataset with 2,000 pixel-wise annotated images for SOD.

Main Results:

• DAFNet effectively captures long-range semantic context and handles scale variations in RSIs.
• The proposed GCA module enhances attention map generation by enabling shallow-to-deep feature flow.
• Experiments show DAFNet significantly outperforms state-of-the-art SOD methods on the new benchmark dataset.

Conclusions:

• DAFNet provides a robust solution for salient object detection in optical remote sensing images.
• The developed dataset and DAFNet advance the field of SOD for remote sensing applications.