Detection of Compact Binary Coalescence with U-net

• 

  Good Detection

• 

  Bad Detection

• 

  Error vs Signal-to-Noise ratio

• 

  Result Histogram

Inspiration

I like astronomy, I was fascinated by LIGO. A similar paper was published that classified when the waves occurred but did not segment the waves. Found an unexplored niche.

What it does

De-noises gravitational wave data of compact binary coalescence (collision and merging of black holes and neutron starts) detected by LIGO.

How we built it

Based on a U-net convolutional architecture originally used to segment images. Originally created a 1D U-net for just the raw signal. Then tried 2D U-net on the short form fourier transform of the signal.

Challenges we ran into

Fourier transform used too much memory, took very long to process.

Accomplishments that we're proud of

Actually deonises the wave.

What we learned

Learned how to implement a 1D convolutional architecture.

What's next for Detection of Compact Binary Coalescence with U-net

Finish Fourier transform variant that should be better since transforms the signal into frequency space which is better for de-noising tasks

Built With

• python
• tensorflow