iLearn-Lab/TGRS25-HDNet

🚀 HDNet: A Hybrid Domain Network with Multi-Scale High-Frequency Information Enhancement for Infrared Small Target Detection

Mingzhu Xu¹  Chenglong Yu¹  Zexuan Li¹  Haoyu Tang¹  Yupeng Hu^1✉  Liqiang Nie¹

¹Affiliation (Please update if needed)

Official implementation of HDNet, a Hybrid Domain Network for Infrared Small Target Detection (IRSTD).

🔗 Journal: IEEE Transactions on Geoscience and Remote Sensing (TGRS), 2025
🔗 Task: Infrared Small Target Detection (IRSTD)
🔗 Framework: PyTorch

---

📌 Model Information

1. Model Name

HDNet (Hybrid Domain Network)

---

2. Task Type & Applicable Tasks

• Task Type: Infrared Small Target Detection / Remote Sensing
• Core Task: Small target detection under complex backgrounds
• Applicable Scenarios:
  • Infrared surveillance
  • Remote sensing target detection
  • Low-SNR object detection

---

3. Project Introduction

Infrared small target detection is challenging due to low signal-to-noise ratio and complex background interference.

HDNet proposes a Hybrid Domain Network that integrates spatial-domain and frequency-domain representations:

• Spatial Domain Branch: introduces Multi-scale Atrous Contrast (MAC) module to enhance target perception
• Frequency Domain Branch: introduces Dynamic High-Pass Filter (DHPF) to suppress low-frequency background
• Combines complementary representations to improve target-background contrast

Key Contributions:

• A hybrid-domain framework combining spatial and frequency information
• MAC module for multi-scale small target perception
• DHPF module for adaptive low-frequency suppression
• Extensive validation on three benchmark datasets

---

4. Training Data Source

Datasets:

• IRSTD-1K
• NUAA-SIRST
• NUDT-SIRST

Download datasets and place them in:

./datasets

---

🚀 Environment Setup

• Ubuntu 22.04
• Python 3.10
• PyTorch 2.1.0
• Torchvision 0.16.2+cu121
• CUDA 12.1
• GPU: NVIDIA RTX 3090

---

🚀 Training

python main.py --dataset-dir './dataset/IRSTD-1k' --batch-size 4 --epochs 800 --mode 'train'

---

🚀 Testing

python main.py --dataset-dir './dataset/IRSTD-1k' --batch-size 4 --mode 'test' --weight-path './weight/irstd.pkl'

---

📊 Quantitative Results

Dataset	mIoU	Pd	Fa
IRSTD-1K	70.26	94.56	4.33
NUAA-SIRST	79.17	100	0.53
NUDT-SIRST	85.17	98.52	2.78

---

📊 Qualitative Results

Visual results:

https://drive.google.com/drive/folders/1RfoxhoHpjfbRMZHBOvISrJSB5lpoz40t?usp=drive_link

---

⚠️ Notes

• Based on improvements over MSHNet
• Uses SLS loss
• Designed for research purposes

---

📝 Citation

@ARTICLE{11017756,
 author={Xu, Mingzhu and Yu, Chenglong and Li, Zexuan and Tang, Haoyu and Hu, Yupeng and Nie, Liqiang},
 journal={IEEE Transactions on Geoscience and Remote Sensing}, 
 title={HDNet: A Hybrid Domain Network With Multiscale High-Frequency Information Enhancement for Infrared Small-Target Detection}, 
 year={2025},
 volume={63},
 pages={1-15},
 doi={10.1109/TGRS.2025.3574962}
}

---

📬 Contact

For questions or collaboration, please contact the corresponding author.

---