Salient Object Detection in Optical Remote Sensing Images Based on Hierarchical Semantic Interaction.

Area of Science:

• Computer Vision
• Remote Sensing
• Artificial Intelligence

Background:

• Salient object detection in optical remote sensing faces challenges from complex backgrounds, scale variations, and interference.
• Current methods struggle with feature interaction and contextual information utilization in decoder architectures.

Purpose of the Study:

• To propose a novel Hierarchical Semantic Interaction Module for enhanced salient object detection in optical remote sensing.
• To improve feature synergy and contextual information exploitation for better detection of multi-scale and multi-type salient objects.

Main Methods:

• Developed a Hierarchical Semantic Interaction Module incorporating foreground content modeling.
• Implemented a hierarchical semantic interaction mechanism within a multi-scale feature space.
• Integrated the module into a salient object detection framework for optical remote sensing images.

Main Results:

• The proposed method significantly improved detection performance on benchmark datasets.
• On the EORSSD dataset, the full model increased max F-measure by 2.74% and S-measure by 2.69% over the baseline.
• The method demonstrated robustness and strong generalization capabilities in complex remote sensing scenarios.

Conclusions:

• The Hierarchical Semantic Interaction Module effectively addresses limitations in existing salient object detection methods for remote sensing.
• The proposed approach enhances feature interaction and contextual understanding, leading to superior performance.
• The method shows significant potential for applications requiring accurate salient object detection in diverse remote sensing environments.