jokoum / satellite-image-classification

Semester project for the Deep Learning class of the MSc in Artificial Intelligence

Satellite Image Classification

---

Semester project for the Deep Learning class of the MSc in Artificial Intelligence by NCSR Demokritos and University of Piraeus

The project utilizes the JPG version of the satellite images dataset Planet: Understanding the Amazon from Space and utilizes the Attention mechanism to achieve Multi-label classification through a CNN encoder and an RNN decoder, investigating the application of Computer Vision and Recurrent Neural Network algorithms at detecting phenomena like deforestation.

Tensorflow v2 (Keras) platform was used for the neural networks creation. Network architecture based on Image captioning with visual attention tutorial.

You can clone the project from repository using the following command:

git clone https://github.com/JoKoum/satellite-image-classification.git

Google Colab environment was used for setting up and training the networks.

Colab Notebook link