PlantMD is a image-based disease prediction app that can be installed locally using Docker or launched through CyVerse Discovery Environment. The webapp currently displays two outputs. The prediction % which indicates how confident the prediction is and a brief description of the disease.
The PlantMD is based on transfer learning method with VGG16 architecutre that has been trained on a dataset consisting of 54,000 labelled RGB images of both healthy and diseased leaves from PlantVillage website. Since the data is highly unbalanced for the 38 different classes of diseases, a custom script to augment the data. The data augmentation mainly consists of rotating the image into different angles as shown here. This resulted in 87,000 RGB images that are highly balanced compared to original data. After data augmentation, the data was split into training and validation datasets in 80-20 ratio.
- Install Docker and launch Plantmd Docker container (or use Discovery Environment)
- Check http://localhost:8501/ (or use Discovery Environment)
- Upload a test image
- Done! 🎉
👇Screenshots:
Docker can be installed on any of three platform using the instructions from Docker website. You can also try Play-With-Docker without installing Docker on your computer
$ git clone https://github.com/upendrak/Disease_Predictor.git && cd Disease_Predictor
$ docker run --rm -p 8501:8501 upendradevisetty/diseasepredictor:1.0 Open http://localhost:8501/ on your computer to load the dashboard
Alternatively you can run the Plantmd on CyVerse Discovery Environment.
Note: You need to register to CyVerse before you can use. More information about user registration can be found here
After you register, you click the button below to launch Plantmd on CyVerse Discovery Environment
