SHAKE IT OFF SPACE!
Fabrizio De Fiore
Sorin Stoleru
Date: October 06 2024
Challenge outline
Write a computer program to analyze real data from the Apollo missions and the Mars InSight Lander to identify seismic quakes within the noise
.
Seismic Detection Across the Solar System
Plot - Plot - Plot
What does it mean????
Workflow
Double filtering process, using Bandpass and Wiener filter
Filtering
01
STA/LTA algorithm is being normalized and used by the neural network as well as the STFT
Algorithms
02
The net is being trained for not even 1000 epochs which decreased the loss down to 0.006
Training
03
After some testing, we were successfully able to detect seismic events resulting more than 90% of accuracy
Testing & Detecting
04
Why Convolution?
Seismic data is a series of measurements over time, making it well-suited for 1D CNNs, which are designed to process sequential information.
Temporal Pattern Recognition
Convolutional layers share weights, reducing the number of parameters and computational cost compared to fully connected networks
Shared Weights
1D CNNs can detect the same pattern regardless of its position within the seismic waveform
Translation Invariance
1D CNNs can learn hierarchical features, starting from simple patterns and gradually building up to more complex ones. This means the network can capture both fine-grained and coarse-grained information.(individual peaks or overall patterns)
Hierarchical Feature Learning
Check out the other Screen !
Since we are XR and Game Developers, we opted for Unity environment to represent the data, to make it more immersive and visualizable
Unity Demo
Future Applications
The time constraints of the challenge presented a unique set of challenges for model training, such as limited training time, data availability, and my laptop computational power
With more time for finding and cleaning data, and more training time for the model
IT CAN BE IMPROVED !
Accuracy Improvements
Using programming languages such as CSound, the seismic event can also be reproduced, for a more realistic and representive experience
Sound Implementation
Using a TCP Server-Client architecture, and multi threading python scripts and Unity can send data toi each other, making the all visualizzation process fully automated
Unity data representation automation
Thanks
We would like to express our sincere gratitude to
Our project mentor and Data Visualisation Professor, for his invaluable guidance and support throughout this project
Yuhang Ye
And the organizer’s team for their dedication, kindness, and the successful realization of this remarkable event.
Sahir Sharma
And all the other mentors who provided us with a wide range of support, from mental encouragement to practical assistance
Conor Keighrey