Automated classification of tomato plant disease

Deep learning based automated tomato plant disease classification covering over 40 disease classes and 4 healthy classes.

Get Started

Instal python virtualenv

sudo apt install virtualenv virtualenv --system-site-packages -p python3 py35

Activate the python virtualenv (run from root)

source py35/bin/activate

Install dependencies

sh requirements.sh

Note:

1. To setup nginx properly, follow the setup tutorial here: https://www.matthealy.com.au/blog/post/deploying-flask-to-amazon-web-services-ec2/

2. Delete the default nginx page
   sudo rm /etc/nginx/sites-enabled/default

3. Restart the nginx server
   sudo service nginx restart

Project tree

• Kisan_app

  • model: directory containing model files for API
  • templates
  • static

• tf_files

  • models: directory containing model files for training
  • bottlenecks: feature files for each image for each

• [data_dir]: get data_dir @{insert link to S3 bucket}

• scripts: contains python files for training the models, evaluation and quantization

• train.sh: bash script to run the training for inceptionV3 model

• train_mobilenet.sh: bash script to run the training for the mobilenet model

• retrain.sh: bash script to re-train the model on new images

• retrain_compare.py: utility script for retrain.sh

• README.md

Training the image-classification model

Activate the python virtualenv (run from root)

source py35/bin/activate

Change the dir to crop_classifcation_updated

cd KisanLab_CPU

Train the inceptionV3 model (more accurate, larger size, slow to train)

sh train.sh

or

Train the mobilenet model (less accurate, smaller size, fast to train)

sh train_mobilenet.sh

API

Kisan_app folder contains all the files for the API.

How to run the API?

Start the server (run from root)

source py35/bin/activate

Change dir to server dir

cd crop_classification_updated/kisan_app

Start the server (logging occurs in nohup.out file)

nohup gunicorn app:app -b localhost:8000 &

Access API via curl cmd

curl -F file=@/path/to/your/image ${PUBLIC_IP_OF_EC2_INSTANCE}/api_call

Output (in json)

[ { "Prediction1": "tomato fruit borer", "Confidence1": "0.490311", "Confidence2": "0.258647", "Prediction2": "cutworm on tomato" } ]

or

Access the web-api via the public IP of the ec2 instance

Output