lukerbs / radio-wave-classifier

GSI Internship Notebooks

Introduction

This repository contains code as part of my internship at GSI Technology in 2019. This contains:

• Custom ResNet neural network architecture in Keras for Signal Classification
  • Code training a resnet from scratch
  • Code for loading training data
  • Code for extracting hamming fingerprints from signal embeddings
  • Code for visualizing signal embeddings in t-SNE
• Introduction to Facebook Artificial Intelligence Similarity Search (FAISS)

Getting Started

• To start training from scratch, go to the notebook here:
• Here is a link to a blog I wrote describing the code: https://medium.com/gsi-technology/residual-neural-networks-in-python-1796a57c2d7

Signal Classifier: Extracting and Visualizing Embeddings

• Once you have trained your ResNet go to faiss_hamming.ipynb to extract binary fingerprints and run similarity search on the signals.

References:

This work was inspired by:

• Over the Air Deep Learning Based Signal Classification by Tim O'Shea et. al.
• FAISS Project: https://github.com/facebookresearch/faiss/wiki/Getting-started