🔴 Infrared Small Target Detection via Wavelet-Driven Frequency Matching and Saliency-Difference Optimization

Qianwen Ma, Shangwei Deng, Bincheng Li, Zhen Zhu, Ziying Song, Xiaobo Li^†, and Haofeng Hu^†

^†Correspondence

Tianjin University, Beijing Jiaotong University

(a) Conventional downsampling and upsampling processes. SLVC is sliding local variance calculation, $4 \times$ is images with 4-fold downsampling followed by restoration to original size, where the processed texture images show differences and roughness compared to the pre-processed images. (b) Wavelet transform downsampling and upsampling processes, where the two texture images before and after processing exhibit extreme similarity.

---

(a) The overall framework of the proposed method. (b) The process of generating four low-resolution components from the input image through DWT downsampling and the process of upsampling back to a high-resolution image through IDWT. (c) The processing of two consecutive WFE modules. When the module is a DWFE, the feature maps will be concatenated with those from other nodes.

---

🔥 News

• [2025.09] Current Status: TGRS Accept!
• [2025.08] Current Status: TGRS R2!
• [2025.08] We release the code.

---

💻 Requirements

• PyTorch >= 1.13.1
• CUDA >= 11.3

---

🚀 Train and Inference Guide for DWTFreqNet

This section outlines the steps to run inference using the DWTFreqNet model.

---

📝 Step 1: Prepare the Dataset

Download the open-source infrared small target detection datasets we used: NUDT-SIRST, NUAA-SIRST, and IRSTD-1K.

Specify the dataset you want to train on and the path where the dataset is placed:

parser.add_argument("--dataset_names", default=['NUDT-SIRST'], type=list)
parser.add_argument("--dataset_dir", default=r'../Dataset')

---

▶️ Step 2: Run Train

Run the train script:

python train.py

The output results will be saved to the ./log/ directory.

---

▶️ Step 3: Run Test

Run the test script:

python test.py

Weights

Baidu Pan: https://pan.baidu.com/s/1nWugwqxK-KvL99A_F7_EKA?pwd=rph9 code: rph9

---

📊 Experimental Results

---

🙏 Thanks

Our code is based on SCT. You can refer to their README files and source code for more implementation details.

---

📖 Citation

If you find our work useful, please consider citing:

@article{ma2025dwtfreqnet,
  title={DWTFreqNet: Infrared Small Target Detection via Wavelet-Driven Frequency Matching and Saliency-Difference Optimization},
  author={Ma, Qianwen and Deng, Shangwei and Li, Bincheng and Zhu, Zhen and Song, Ziying and Li, Xiaobo and Hu, Haofeng},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  year={2025},
  publisher={IEEE}
}