We developed a machine learning model that analyzes incoming seismic data, filtering out noise and isolating seismic events. The model uses a Long Short-Term Memory (LSTM) neural network, which is well-suited for time-series data, to differentiate between useful signals and background noise. Only the relevant data is prioritized for transmission back to Earth.

Using a machine learning model to analyze incoming seismic data, focusing on filtering out noise and isolating seismic events. The model is based on a Long Short-Term Memory (LSTM) neural network, which is ideal for handling time-series data like seismic signals due to its ability to retain long-term dependencies in the data. To train the model, we used the labeled seismic datasets from the Apollo and Mars InSight missions, which included both seismic events and background noise. This data was split into training and validation sets. Preprocessing steps included normalizing the data to ensure consistency across different signals and applying data augmentation techniques to increase the robustness of the model, given the limited availability of planetary seismic data.

Next, we trained the LSTM model using TensorFlow, feeding in sequences of seismic data points. The model learned to distinguish between useful seismic signals and noise by recognizing patterns in the input sequences. During training, we fine-tuned parameters such as learning rate, and batch size, and tweaked the LSTM layers to optimize the model’s performance.

Finally, once the model was trained, it was tested on the validation set to evaluate its accuracy in detecting seismic events. For the trained LSTM network to efficiently process continuous streams of seismic data, filtering out irrelevant noise and prioritizing the transmission of critical seismic events back to Earth.

The outcome:

- Energy Savings: Less power is required to transmit large data sets, enabling more efficient use of resources in space missions.

- Data Optimization: Prioritizes key seismic events, freeing up transmission bandwidth for other scientific instruments. As the number of specialists involved in the mission grows, each collecting data relevant to their own research objectives, the potential ways in which the data can be utilized expands.

- Advanced Seismology: Supports deeper exploration of planetary quakes with minimal data loss, potentially discovering new events.

Tools and Technology Used:

- Programming Languages: Python was used as the primary language.

- Machine Learning Libraries: TensorFlow was used to implement the LSTM model, and Scikit-learn (sklearn) was employed for additional data analysis and feature extraction.

- Software: The NASA Jupyter Notebooks were reviewed for experimentation and code development.

- Data: Real seismic data from Apollo and Mars InSight missions was used for model training and testing.

2024 NASA Space Apps Challenge

AvacadOS