A homemade convolutional neural network

This CNN can be used to clasify land cover from satellite images from the EuroSAT dataset.

Run locally

Installation of the dependencies

• Conda has been used to manage environments
• The only dependencies are numpy and pillow (requires the libjpeg prerequisite)

Training the model

• Running python main.py --should-preprocess will download the dataset and preprocess the data
• Just run python main.py the rest of the time

Results

The model achieves an accuracy of around 97% after 15 epochs

Credits

This implementation was inspired by :

• TheIndependentCode
• Riccardo Andreoni