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	S_pr1	S_pr2	S_pr3
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)

dydgns2017 / pwd-synthetic-dataset Goto Github PK