A CNN that uses transfer learning on mobilenet to decide if an image of a plant contains a healthy plant or a dying plant. This is essentially my catdogcnn code modified to work on 224x224 images of healthy and dying plants. I sourced my training images using the chrome extension fatkun batch image download and google images.
Copyright © 2019 Rory Sharp All rights reserved.
This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.
You should have received a copy of the GNU General Public License along with this program. If you have not received this, see http://www.gnu.org/licenses/gpl-3.0.html.
For a summary of the licence go to https://tldrlegal.com/license/gnu-general-public-license-v3-(gpl-3)
- Curl
sudo apt-get install curl - Bash
sudo apt-get install bash
curl https://raw.githubusercontent.com/qwertpi/catvdogcnn/master/rtinstall.bash | bash
- Git
sudo apt-get install git - Python 3
- ImageMagick (sic)
sudo apt-get install imagemagick - libhdf5 (only needed on some systems)
sudo apt-get install libhdf5-serial-dev
Install the python packages either running sudo -H pip3 install -U -r requirements.txt after downloading the repo (after step 0) or installing all the packages named in requirements.txt manually using sudo -H pip3 install -U package-name
0. Download this repo git clone https://github.com/qwertpi/planthealthcnn.git
1. Source some images of healthy and dying plants and put them in the respective folders in images
2. Resize your images cd images/healthy/ && for file in *; do convert $file -resize 224x224 -gravity center -background "rgb(0,0,0)" -extent 224x224 $file; done then cd .. then cd dying/ && for file in *; do convert $file -resize 224x224 -gravity center -background "rgb(0,0,0)" -extent 224x224 $file; done then cd .. then another cd ..
3a. Run aug.py to create new images with added noise, random brighnes and cropping python3 aug.py
3b. (Optional) Copy your old images to augmented as well cp images/healthy/* augmented/healthy/ && cp images/dying/* augmented/dying/
4. If you want to run viz.py you will have to replace the keract.py file in wherever your python packeges live with my forked verison
1. Run train.py to train the model python3 train.py
2. (Optional) Run metrics.py to see which images in the training data your model is failing to classify python3 metrics.py
3. Change the file variable in predict.py to point towards the image you want to check then run it, code for live prediction on webcam feeds may or may not come in the futre
4. (Optional) Run viz.py changing the file that is used if you want to see an image go through the network python3 viz.py, if you want to make a gif I used ezgif.com
If you get the /tmp direcotry doesn't use RAM message you can create a 10MB RAM backed directory called /ram by running the command sudo mkdir -p /ram && sudo mount -t tmpfs -o size=10m tmpfs /ram and then change tmp_dir to point to your RAM backed directory eg. tmp_dir="/ram/"
I have attached 3 images that I believe to be in neither the train nor test set for you to use in prediction testing if you want