Our project focuses on seismic detection across the Solar System for missions on planets like the Moon or Mars. We developed a deep learning-based solution that processes and filters seismic data, ensuring only scientifically important seismic events are transmitted back to Earth. By reducing the volume of data that needs to be sent, we conserve power and improve efficiency, which is critical for long-duration planetary missions. This solution directly addresses the challenge of power management and data transmission on planetary missions, making seismology missions more sustainable and impactful.
Our solution consists of 5 steps:
- Defining the window that splits the seismic wave to a fixed number of timestamps.
- Creating a proximity/heat feature that defines a value for each time stamp, and marks how much each time stamp is close to the timestamp that marks the start of the seismic wave.
- A deep learning model consists mainly of CNN-1D, Pooling, Dropout, and BatchNormalization layers.
- A post-processing step to focus on the area that contains the predicted seismic event is based on splitting the seismic wave into smaller windows and summing all the predictions in each window. Then, we choose the window with the highest summed predictions.
- We then choose the timestamp that has the highest predicted value in that window.
Important
Our results are saved in CSV file at results file
In the file_name column, files are named as {APOLLO_NUMBER}-{GRADE_LETTER}-{EVENT_ID}.csv
APOLLO_NUMBER is a number from 2 digits, GRADE_LETTER is a letter and EVENT_ID is a number from 5 digits
For the first 10 rows, there is no information about APOLLO_NUMBER and GRADE_LETTER due to they belong to Mars data so it is just EVENT_ID.