🔥 MNHU-Net: A Multi-Scale Feature Fusion and Nested Structure-Based High-Order U-Net for Infrared Small Target Detection [📄 Paper Link]

Xiaoyang Yuan, Chunling Yang, Yu Chen, Yan Zhang, IEEE Transactions on Aerospace and Electronic Systems 2025.

If the implementation of this repo is helpful to you, just star it！⭐⭐⭐

The weights of three datasets are provided in the MNHU-net/result path

Introduction

We present a multi-scale feature fusion and nested structurebased High-order UNet (MNHU) for IRSTD. We evaluate the proposed high-order UNet-based methods (MNHU-E, MNHU-D, and MNHU) on three public datasets (e.g. SIRST, IRSTD-1k, NUDT-SIRST), which demonstrates the effectiveness of our methods. Our main contributions are as follows:

1. A novel high-order UNet paradigm is proposed. This paradigm enhances feature representation by leveraging high-order skip connections to fuse the feature maps from adjacent layers with strong correlations.
2. Building upon the paradigm, we propose three high-order U-Net architectures (MNHU-E, MNHU-D, and MNHU) to calibrate infrared feature maps by distinguishing small targets from background textures and modeling long-range dependencies.
3. To enhance segmentation performance, a high-order interactive feature extractor integrating a residual channel-spatial attention module and a high-order fusion supervision module is incorporated.

Usage

1. Data

• The SIRST dataset download dir [ACM]
• The NUDT-SIRST dataset download dir [DNANet]
• The IRSTD-1k dataset download dir [ISNet]

Our project has the following structure:

├──./dataset/
│    ├── IRSTD-1K
│    │    ├── images
│    │    │    ├── XDU0.png
│    │    │    ├── XDU1.png
│    │    │    ├── ...
│    │    ├── masks
│    │    │    ├── XDU0.png
│    │    │    ├── XDU1.png
│    │    │    ├── ...
│    │    ├── 80_20
│    │    │    ├── train.txt
│    │    │    ├── test.txt
│    ├── NUDT-SIRST
│    │    ├── images
│    │    │    ├── 000001.png
│    │    │    ├── 000002.png
│    │    │    ├── ...
│    │    ├── masks
│    │    │    ├── 000001.png
│    │    │    ├── 000002.png
│    │    │    ├── ...
│    │    ├── 80_20
│    │    │    ├── train.txt
│    │    │    ├── test.txt
│    ├── ...
│    ├── ...
│    ├── NUAA-SIRST
│    │    ├── images
│    │    │    ├── Misc_1.png
│    │    │    ├── Misc_2.png
│    │    │    ├── ...
│    │    ├── masks
│    │    │    ├── Misc_1.png
│    │    │    ├── Misc_2.png
│    │    │    ├── ...
│    │    ├── 80_20
│    │    │    ├── train.txt
│    │    │    ├── test.txt

2. Train.

python train.py 

3. Test.

python test.py 

• This code is highly borrowed from AMFU. Thanks to Won Young Chung.
• This code is highly borrowed from DNANet. Thanks to Boyang Li.
• This code is highly borrowed from IRSTD-Toolbox. Thanks to Xinyi Ying.

4. Results and Trained Models

The weights of three datasets are provided in the MNHU-net/result path

| Dataset      | mIoU  | DSC   | Pre   | Re    |
|--------------|-------|-------|-------|-------|
| SIRST        | 80.33 | 89.09 | 89.05 | 89.13 |
| IRSTD-1k     | 69.28 | 81.85 | 79.82 | 83.99 |
| NUDT-SIRST   | 90.47 | 95.00 | 96.20 | 93.82 |

Citation

If you find the code useful, please consider citing our paper using the following BibTeX entry.

@ARTICLE{10979417,
  author={Yuan, Xiaoyang and Yang, Chunling and Chen, Yu and Zhang, Yan},
  journal={IEEE Transactions on Aerospace and Electronic Systems}, 
  title={MNHU-Net: A Multi-Scale Feature Fusion and Nested Structure-Based High-Order U-Net for Infrared Small Target Detection}, 
  year={2025},
  volume={},
  number={},
  pages={1-16},
  keywords={Feature extraction;Data mining;Correlation;Clutter;Object detection;Encoding;Calibration;Attention mechanisms;Robustness;Complexity theory},
  doi={10.1109/TAES.2025.3564932}}

Contact

Welcome to raise issues or email to yuanxiaoyang1998@outlook.com for any question regarding our MNHU-Net.