Learning Conditional Diffusion Transformer for Salient Object Detection in Optical Remote Sensing Images.

Area of Science:

• Computer Vision
• Remote Sensing
• Artificial Intelligence

Background:

• Salient object detection in optical remote sensing images is challenging.
• Existing methods often use limited data and convolutional neural networks (CNNs).

Purpose of the Study:

• Introduce a novel architecture, the Conditional Diffusion Transformer Network (CDTNet).
• Improve salient object detection in optical remote sensing images (ORSI SOD).

Main Methods:

• Developed a Transformer-based progressive cross-stage fusion (PCSF) module for feature integration.
• Implemented a patch strategy (PS) for fine-grained feature aggregation.
• Utilized an encoder feature enhancement (EFE) module with spatial and channel attention.

Main Results:

• The CDTNet effectively integrates multi-scale features for enhanced saliency prediction.
• The patch strategy optimizes transformer layer utilization for detailed information.
• Experimental results show CDTNet outperforms state-of-the-art (SOTA) methods on benchmark datasets.

Conclusions:

• CDTNet offers a superior approach to ORSI SOD.
• The proposed modules significantly enhance feature representation and detection accuracy.
• This work advances the field of salient object detection in remote sensing imagery.