Machine Learning Edible Plants

A machine learning model that determines if a plant is edible or not. There are currently two ways this is achieved.

Upload an image

1. Select the choose button, upload an image, and it will say if it is an edible plant or not.

Select five edible plants and five non-edible images at random to make predictions

1. Select the button show sample predictions

Plan of action followed to achieve this

Getting the data together

1. Refer to 'List of edible stuff' - collect images
2. Refer to 'List of non edible stuff' collect images

Creating the model

3. Create the dataframe with the following columns (route for creating model)
   • url (name of the image)
   • name (can be a plant, car, plane...etc)
   • edible (car = 0, edible plant = 1)
4. Create the model (refer to section)
5. Use hyper parameters; save mutliple weight files and compare and use best
6. Train the model against edible and non-edible images.

Making predictions

7. Based on our prediction, if the plant is classified as edible, then show that its an edible plant, along with the name. (route for predicting)
8. Based on the prediction, return whether it is edible or not.

Extras

9. Add new images to DataSet if needed

List of edible stuff

• Edible plants

List of non-edible stuff

• Tree's.
• Cars
• planes
• Animals
• Fruits
• Poisonous
• Plants

How to run

1. Install requirements via requirements.txt
2. Bring in some test data under the static folder. The name of the folder should be called Model_data and the hierarchy should be:
   • Model_Data
     • test_dataset
       • edible
         • images
       • non-edible
         • images
     • train_dataset
       • edible
         • images
       • non-edible
3. Go to the application directory and run python3 ./app.py
4. Go to localhost:5000/create-model to create a model against the training data above. The weights will then be saved (by default this will save as edible_weights_v1.h5.
5. However, you don't need to train the model to use this, the weights file can be used to predict images straight off.