SggNet: A Lightweight Semantic- and Graph-Guided Network for Advanced Optical Remote Sensing Image Salient Object Detection

---

📝 Overview

SggNet is a lightweight and efficient network for ORSI-SOD, achieving:

• Parameters: 2.70M
• FLOPs: 1.38G
• Inference Speed: 108 FPS

It demonstrates superior performance compared to state-of-the-art lightweight ORSI-SOD methods, delivering accurate saliency detection, sharper boundaries, and clearer activation maps.

---

📊 Results

Below are sample results showcasing the effectiveness of SggNet:

Example Outputs

• In Figure 1, we visualize saliency maps generated by SggNet compared to other state-of-the-art methods in challenging scenarios:

Key Metrics on EORSSD and ORSSD Datasets

Dataset	$S_m \uparrow$	$F^{max}_{\beta} \uparrow$	$F^{mean}_{\beta} \uparrow$	$F^{adp}_{\beta} \uparrow$	$E^{max}_{\phi} \uparrow$	$E^{mean}_{\phi} \uparrow$	$E^{adp}_{\phi} \uparrow$	$\mathcal{M} \downarrow$
EORSSD	0.9279	0.8770	0.8596	0.8386	0.9762	0.9689	0.9678	0.0068
ORSSD	0.9342	0.9032	0.8896	0.8884	0.9759	0.9695	0.9720	0.0111

---

📥 Installation and Usage

Clone the Repository

git clone https://github.com/LittleGrey-hjp/SggNet
cd SggNet

Install Dependencies

pip install -r requirements.txt

Training Configuration

The pretrained model(MobileNetv2) is stored in Google Drive and Baidu Drive (zskr). After downloading, please change the file path in the corresponding code.

Run `train.sh` to train.

Testing Configuration

Our well-trained model is stored in Google Drive and Baidu Drive (3knk). After downloading, please change the file path in the corresponding code.

Run `test.sh` to train.

Detection Maps

Our Detection Maps are stored in Google Drive and Baidu Drive.(qjbw). Please check.

Run `test.sh` to train.

### Evaluation

- Evaluate SggNet: After configuring the test dataset path, run `eval.sh` in the `srun` folder for evaluation.
- PR-Curves: We provide the code for obtaining PR-Curves through detection results. Please refer to 'PR_Curve.py'.

---

## 📬 Contact

For questions or feedback, feel free to open an issue on GitHub or contact us via email at [darrellduncan313@gmail.com](darrellduncan313@gmail.com).

---

### Citation
```bash
@Article{rs17050861,
AUTHOR = {Liu, Jie and He, Jinpeng and Chen, Huaixin and Yang, Ruoyu and Huang, Ying},
TITLE = {A Lightweight Semantic- and Graph-Guided Network for Advanced Optical Remote Sensing Image Salient Object Detection},
JOURNAL = {Remote Sensing},
VOLUME = {17},
YEAR = {2025},
NUMBER = {5},
ARTICLE-NUMBER = {861},
URL = {https://www.mdpi.com/2072-4292/17/5/861},
ISSN = {2072-4292},
DOI = {10.3390/rs17050861}
}

Let me know if you’d like additional customization!