Leaf Disease Dataset Framework

This repository is the official instruction of Leaf Disease Dataset Framework, a cloud collection datasets published from article: "Towards Sustainable Agriculture: A Lightweight Hybrid Model and Cloud-based Collection of Datasets for Efficient Leaf Disease Detection".

This framework provides 38 public leaf disease datasets, collected from research platforms (e.g., Google Scholar, IEEE Xplore, Kaggle, Data Mendeley).

How to use this repository.

1. Clone this repository: !git clone https://github.com/iec2-uit/iec-models
2. Install prerequisites libraries:
3. cd into train-iec-models and install dependencies package:
   • pip install -r requirements.txt
     
4. or run this:
   • !pip3 install timm
   • !pip3 install pydicom
   • !pip3 install git+https://github.com/albumentations-team/albumentations
   • !pip3 install catalyst
   • !pip install -U albumentations==1.3.0
   • !pip install utils
5. Run python3 main.py or colab notebook (https://colab.research.google.com/drive/13WWKR97NsVPYmJOa_ojecC6aVdaO1WXk?usp=sharing ) with your parameters and models via the supported pipeline from our IEC library with 6 functions:

• IEC.download('name datasets')

  • This function supports download available leaf disease datasets from our cloud. These are currently 31 datasets you can download by name as described here.

• IEC.seed_everything('seed value')

  • This function sets a seed value for pseudo-random number generators, which guarantees the reproductivity of deep learning algorithm implemented by Pytorch.

• IEC.folds()

  • This function splits dataset into k consecutive folds, which keeps the ratio between classes in each fold constant as in the original dataset.

• IEC.preprare_dataloader()

  • This function prepares the training and evaluation datasets according to the training/evaluation ratio.

• IEC.trainoneepoch()

  • This function finds the best combination of weights and bias for minimizing the loss value.

• IEC.validoneepoch()

  • This function evaluates the model performance after training with data.

If you use this github for a paper please cite:

@article{thai2023towards, title={Towards sustainable agriculture: A lightweight hybrid model and cloud-based collection of datasets for efficient leaf disease detection}, author={Thai, Huy-Tan and Le, Kim-Hung and Nguyen, Ngan Luu-Thuy}, journal={Future Generation Computer Systems}, year={2023}, publisher={Elsevier} }