A Plant Disease Detector App based on Nested Transfer Learning

The objective of the project is to identify plant disease by using image of a plant leaf using deep learning model. Currently 5 Plants are supported Apple, Corn, Orange, Potato and Tomato. The Whole project has four components:- Deep Learning(Using Tensorflow), Command Line Interaction(Using ML-HUB), Linux Desktop App(MLHUB backend), and Android App(Tflite backend). This project has been done as a requirement of the course COM4560(ANU), under the supervision of Prof. Graham Williams.

Mobile App Demo

Desktop App Demo

Results on test images

Nested Transfer Learning Concept Map

The knowledge in the diagram refers to weights. The weights in model layers will be nested.

Trained Models with weights(h5)

1) Tansfer Learning Based EfficientB2 - 21 Classes
2) Nested Tansfer Learning Based EfficientB2 - 22 Classes

Dataset Sources

Plant Village Dataset - https://data.mendeley.com/datasets/tywbtsjrjv/1
Banana Leaf images - https://github.com/godliver/source-code-BBW-BBS/

Citation

If this repository is useful for your research, please cite as below:

@article{panda2022PlantDis,
  title={PlantDis: A Plant Disease Detector App 
  based on Nested Transfer Learning},
  author={Panda, Saswat and Williams, Graham},
  year={2022},
  repository-link="https://github.com/spsaswat/plantdis"
}

Funding

Starting from 11/06/2024, the project is supported by APPN (https://www.plantphenomics.org.au/).