In this project, we create a variety of deep learning models in an effort to predict the species of different leaves.
In our Keras models, we use both (pre-extracted) structured and unstructured image data in order to build a good classifier.
The data can be found on Kaggle: https://www.kaggle.com/c/leaf-classification/overview
Our best model achieves perfect accuracy for the test data (on Kaggle).
We build:
- A fully connected Neural Network and use grid-search to optimize the hyperparameters.
- A Convolutional Neural Network with varying degrees of image augmentation.
- Transfer Learning with VGG and varying degrees of image augmentation.
- A merged Network that uses both the structured and unstructured data.
- A function that asks for human input in cases where the Neural Network is uncertain about a classification.
Should you have any questions or suggestions, do not hesitate to contact me!