PWD-Synthetic-Dataset

PWD(Pine Wilt Disease) synthesis data generated using 3D rendering tools

Highlights

• Pine Wilt Disease requires early detection due to its severity and lack of cure.
• Our synthetic dataset creation outperforms traditional PWD data collection process.
• Real and synthetic data combination improves PWD F1 Score to 92.88%.
• Synthetic data method aids forest preservation, applies to other agriculture.

Environments

• python version : 3.11.4
• pytorch version : 2.0.1
• GPU : 2080Ti*8EA

## conda env setup
conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia

Dataset & Pre-Trained Model Downloads

Fill out the form through that link and we'll give you access to the dataset.

Pre-trained model downloads link

Desciption Table for Dataset and Pre-Trained Model

R	S	Spr1	Spr2	Spr3
Real Dataset	Synthetic Dataset	Synthetic Image Translation Dataset (one)	Synthetic Image Translation Dataset (two)	Synthetic Image Translation (three)

+ meaning is mixed dataset

Training and Inference

traininig and Inference code reference and execute this code

testset is available this folder

Authors and Citation

Authors : Yonghoon Jung, Sanghyun Byun, Bumsoo Kim, Sareer Ul Amin, Sanghyun Seo

@article{JUNG2024108690,
      title = {Harnessing synthetic data for enhanced detection of Pine Wilt Disease: An image classification approach},
      journal = {Computers and Electronics in Agriculture},
      volume = {218},
      pages = {108690},
      year = {2024},
      issn = {0168-1699},
      doi = {https://doi.org/10.1016/j.compag.2024.108690},
      url = {https://www.sciencedirect.com/science/article/pii/S0168169924000814},
      author = {Yonghoon Jung and Sanghyun Byun and Bumsoo Kim and Sareer {Ul Amin} and Sanghyun Seo}
}

Acknowledgements

We are very grateful to the CSLAB researchers at Chung-Ang University (Prof. Park Sang-Oh, Prof. Lee Jae-Hwan, Dr. Nam Sang-Hyuk, Dr. Cho Min-Gyu, M.S. Lee Yo-Seb, and M.S. Kim Dong-Hyeon) and Prof. Kang Dong-Wann at Seoul National University of Science and Technology for their great help in collecting the real dataset. We are also very grateful to M.S. Yoo Jae-Seok, Won-Seop Shin and students Lee Jeong-Ha, Lee Won-Byung, and Oh Chang-Jin for data labeling. Finally, we'd like to thank Assoicate Seung-Yong Ji (Monitoring & Analysis Department, Korea Forestry Promotion Institute) for responding to so many requests.

This study was carried out with the support of R&D Program for Forest Science Technology (Project No. 2021338C10-2123-CD02) provided by Korea Forest Service (Korea Forestry Promotion Institute) and the National Research Foundation of Korea (NRF) grant funded by theKorean government (MSIT) (No.2022R1A2C1004657).

References

• https://github.com/hankyul2/EfficientNetV2-pytorch for implementation EfficientNetv2
• I2I techniques (one, two, three)