IEEE_TGRS_GSDG-Net

KO | EN

gitlite — search

#python #hacktoberfest #react #typescript #llm #go #rust #deep-learning #linux #macos #java #awesome #cli #vue

IEEE_TGRS_GSDG-Net

(★ 10)

Lianlian Zhao, Lina Yang, Lianhui Liang, Xinzhang Wu, Haoyan Yang. Geometric Spatial Prior-Guided Dynamic Graph Learning With Fourier-Modulated Superpixels for Hyperspectral Image Classification, IEEE Transactions on Geoscience and Remote Sensing (TGRS), 2026.

IEEE_TGRS_GSDG-Net Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

<h1>
  <a href="https://ieeexplore.ieee.org/document/11515211">Geometric Spatial Prior-Guided Dynamic Graph Learning With Fourier-Modulated Superpixels for Hyperspectral Image Classification</a>
</h1>

[![Python](https://img.shields.io/badge/Python-3.10+-3776AB?style=flat-square&logo=python&logoColor=white)](https://www.python.org/)
[![PyTorch](https://img.shields.io/badge/PyTorch-2.x-EE4C2C?style=flat-square&logo=pytorch&logoColor=white)](https://pytorch.org/)
[![IEEE TGRS](https://img.shields.io/badge/IEEE%20TGRS-2026-00629B?style=flat-square)](https://doi.org/10.1109/TGRS.2026.3692393)
[![DOI](https://img.shields.io/badge/DOI-10.1109%2FTGRS.2026.3692393-B31B1B?style=flat-square)](https://doi.org/10.1109/TGRS.2026.3692393)
[![Task](https://img.shields.io/badge/Task-Hyperspectral%20Image%20Classification-2E8B57?style=flat-square)](https://raw.githubusercontent.com/LanLanZhao/IEEE_TGRS_GSDG-Net/main/)
[![Model](https://img.shields.io/badge/Model-GSDG--Net-6A5ACD?style=flat-square)](https://raw.githubusercontent.com/LanLanZhao/IEEE_TGRS_GSDG-Net/main/)

<p>
  <a href="#architecture-overview">Architecture</a> |
  <a href="#dataset-and-hyperparameter-summary">Datasets</a> |
  <a href="#quick-start">Quick Start</a> |
  <a href="#results-reported-in-the-paper">Results</a>
</p>

**Authors:** Lianlian Zhao, <strong>Lina Yang</strong>, <strong>Lianhui Liang</strong>, Xinzhang Wu, and Haoyan Yang

</div>

## Citation

If this repository is helpful for your research, please cite:

```bibtex
@ARTICLE{11515211,
  author={Zhao, Lianlian and Yang, Lina and Liang, Lianhui and Wu, Xinzhang and Yang, Haoyan},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  title={Geometric Spatial Prior-Guided Dynamic Graph Learning With Fourier-Modulated Superpixels for Hyperspectral Image Classification},
  year={2026},
  volume={64},
  number={},
  pages={5513417-5513417},
  doi={10.1109/TGRS.2026.3692393}
}
```

---

## Contact Information

For questions about the code or paper, please contact: [lanlanzhao@yeah.net](mailto:lanlanzhao@yeah.net)

---

## Repository Structure

```text
IEEE_TGRS_GSDG-Net/
|-- Config/
|   `-- config.yaml              # Default experiment configuration
|-- CreateGraph/
|   |-- SLIC.py                  # LDA-SLIC superpixel generation and spatial prior
|   `-- create_graph.py          # Label and mask construction
|-- Model/
|   `-- GSDG.py                  # GSDG-Net model implementation
|-- assets/
|   `-- Graph2.png               # GSDG-Net architecture overview
|-- HyperLoad.py                 # Dataset loader
|-- Utils.py                     # Metrics, loss, visualization helpers
|-- scheduler.py                 # Optimizer and training schedule
`-- main.py                      # Main training and evaluation entry
```

---

## Installation

```bash
conda create -n gsdg python=3.10 -y
conda activate gsdg

pip install -r requirements.txt
```

If your CUDA/PyTorch environment requires a specific wheel, install PyTorch from the official selector first:

```bash
pip install torch torchvision torchaudio
```

---

## Dataset Access

We provide commonly used hyperspectral image classification datasets through Google Drive:

**Google Drive:** [Hyperspectral datasets](https://drive.google.com/drive/u/1/folders/1fl4Ag3979gGPjJzADrSZ58jR48JzR7CI)

Place the downloaded `.mat` files under:

```text
dataset/
```

Included example files:

```text
dataset/
|-- indian_pines_corrected.mat
|-- indian_pines_gt.mat
|-- Salinas_corrected.mat
`-- Salinas_gt.mat
```

---

## Dataset and Hyperparameter Summary

The following table summarizes the datasets included in the public Google Drive collection. A check mark indicates the datasets used in the GSDG-Net paper experiments. Hyperparameters are reported only for the datasets used in the paper.

| Dataset | Used in Paper | Loader Key | Scene Type | Files in Drive | Train Samples | Superpixel Scale `S` | Fusion `lambda` | `d_k` | Top-k | Spatial Prior k |
|---|:---:|---|---|---|---:|---:|---:|---:|---:|---:|
| Indian Pines (IP) | ✓ | `indian` | Agricultural | `indian_pines_corrected.mat`, `indian_pines_gt.mat` | 10/class | 100 | 0.95 | 16 | 8 | 15 |
| Pavia University (PU) | ✓ | `paviau` | Urban | `PaviaU.mat`, `PaviaU_gt.mat` | 10/class | 600 | 0.60 | 16 | 8 | 15 |
| Salinas (SA) | ✓ | `sali` | Agricultural | `Salinas_corrected.mat`, `Salinas_gt.mat` | 10/class | 100 | 0.95 | 16 | 8 | 15 |
| Utopia Planitia (UP) | ✓ | `UP` | Martian mineral scene | `Utopia.mat`, `Utopia_gt.mat` | 10/class | 800 | 0.60 | 16 | 8 | 15 |
| Xuzhou | ✓ | `xuzhou` | Peri-urban | `xuzhou.mat`, `xuzhou_gt.mat` | 10/class | 100 | 0.50 | 16 | 8 | 15 |
| Botswana |  | `botswana` | Wetland | `Botswana.mat`, `Botswana_gt.mat` | - | - | - | - | - | - |
| Dioni |  | `dioni` | Land cover | `Dioni.mat`, `Dioni_gt.mat`, `Dioni_gt_out68.mat` | - | - | - | - | - | - |
| Fanglu Tea Farm |  | - | Agricultural | `FangluTeaFarm.mat`, `FangluTeaFarm_gt.mat` | - | - | - | - | - | - |
| Holden |  | `HC` | Martian mineral scene | `holden.mat`, `holden_gt.mat` | - | - | - | - | - | - |
| Houston 2013 |  | `houston` | Urban | `Houston13.mat`, `Houston13_gt.mat` | - | - | - | - | - | - |
| Houston 2018 |  | `Houston2018` | Urban | `_houston2018.mat`, `_houston2018_gt.mat` | - | - | - | - | - | - |
| Kennedy Space Center (KSC) |  | `ksc` | Wetland vegetation | `KSC_corrected.mat`, `KSC_gt.mat` | - | - | - | - | - | - |
| LN01 |  | `ln01` | Hyperspectral scene | `LN01_HHSI.mat`, `LN01_GT.mat` | - | - | - | - | - | - |
| Loukia |  | `loukia` | Land cover | `Loukia.mat`, `Loukia_gt.mat` | - | - | - | - | - | - |
| MUUFL Gulfport |  | `muufl` | Urban | `Muufl.mat`, `Muufl_gt.mat` | - | - | - | - | - | - |
| Nili Fossae |  | `NF` | Martian mineral scene | `NiliFossae.mat`, `NiliFossae_gt.mat` | - | - | - | - | - | - |
| Pavia Center |  | `paviac` | Urban | `PaviaC.mat`, `PaviaC_gt.mat` | - | - | - | - | - | - |
| Trento |  | `trento` | Urban/rural | `Trento.mat`, `Trento_gt.mat` | - | - | - | - | - | - |
| WHU-Hi-HanChuan |  | `hanchuan` | Agricultural | `WHU_Hi_HanChuan.mat`, `WHU_Hi_HanChuan_gt.mat` | - | - | - | - | - | - |
| WHU-Hi-HongHu |  | `honghu` | Agricultural | `WHU_Hi_HongHu.mat`, `WHU_Hi_HongHu_gt.mat` | - | - | - | - | - | - |
| WHU-Hi-LongKou |  | `longkou` | Agricultural | `WHU_Hi_LongKou.mat`, `WHU_Hi_LongKou_gt.mat` | - | - | - | - | - | - |

### Global Training Settings

| Setting | Value |
|---|---:|
| Optimizer | Adam |
| Learning rate | 0.001 |
| Weight decay | 1e-4 |
| Epochs | 400 |
| Repeated runs | 10 |
| Default seeds | `2025, 2026, 2027, 2028, 2029, 2030, 2031, 2032, 2033, 2034` |
| Metrics | OA, AA, Kappa |

---

## Quick Start

1. Download the required `.mat` files from the Google Drive link above.
2. Put the files into `dataset/`.
3. Edit `Config/config.yaml` if needed.
4. Run:

```bash
python main.py --path-config Config/config.yaml --iterations 10
```

For a single quick sanity run:

```bash
python main.py --path-config Config/config.yaml --iterations 1
```

---

## Configuration Example

```yaml
model:
  model_name: "GSDG"
  lama: 0.95
  d_k: 16
  topk: 8

data_input:
  dataset_name: "indian"

data_split:
  samples_type: fixed
  train_num: 10
  train_ratio: 0.05
  superpixel_scale: 100

result_output:
  path_weight: ResultsModel/
  path_result: Results/
```

---

## Results Reported in the Paper

| Dataset | OA (%) | AA (%) | Kappa (%) |
|---|---:|---:|---:|
| Indian Pines | 90.10 ± 2.38 | 93.88 ± 1.36 | 88.76 ± 2.68 |
| Pavia University | 95.12 ± 2.67 | 97.53 ± 0.84 | 93.66 ± 3.40 |
| Salinas | 97.91 ± 0.85 | 98.90 ± 0.36 | 97.68 ± 0.95 |
| Utopia Planitia | 91.32 ± 1.99 | 94.05 ± 1.97 | 88.09 ± 2.68 |
| Xuzhou | 95.92 ± 0.84 | 98.32 ± 0.54 | 94.87 ± 1.05 |

---

## Notes

- Additional large `.mat` datasets are distributed through Google Drive instead of git.
- Result folders and model checkpoints are ignored by default to keep the repository lightweight.

---

## Acknowledgements

This project references the implementation of [quanweiliu/WFCG](https://github.com/quanweiliu/WFCG). We sincerely thank the authors for making their work publicly available.

IEEE_TGRS_GSDG-Net

IEEE_TGRS_GSDG-Net Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

IEEE_TGRS_GSDG-Net

IEEE_TGRS_GSDG-Net Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README