Spectral Asteroid Classification Using Machine Learning
AJAS · 2025 Physics and Astronomy (inferred)
Overview
The mining of Earth's finite metals has devastating environmental and humanitarian impact. Asteroid mining provides an elegant alternative to the limited resources on Earth, as well as playing a critical role in the quest for interplanetary travel. Thus, a system to reliably determine the composition of asteroids is imperative to seek out the correct targets. Currently, asteroid spectral graphs are matched using mathematical error methods with known spectra of elements to determine their composition. Unfortunately, these approaches are unidimensional and blind to other factors that influence asteroid composition, breaking down when faced with more unique asteroid spectra. In this paper, a novel approach to asteroid spectral analysis is taken by using a convolutional neural network, CNN, engineered to analyze patterns within asteroid spectral graphs supplemented by auxiliary asteroid data, diameter, absolute magnitude, and albedo, to provide insight not found in current asteroid classification methods. Data was sourced from MIT's Small Main Belt Asteroid Spectroscopic Survey, Phase II, SMASSII, and the space rocks database. A program was created to automatically aggregate data from the two sources, filling empty values with simulated data. Data augmentation algorithms were created and employed to balance the dataset, ensuring sufficient unique data was present in each spectral class for the CNN to learn from. Reclassifying images of spectral graphs into 18 common spectral classes from the Bus DeMeo taxonomic system, the CNN predicted the correct asteroid class 98.8% of the time, over a 13% improvement from a model solely using spectral graphs. Thus, this study not only showcases the accuracy of CNNs in analyzing asteroid spectral graphs, but also underscores how auxiliary data enhances asteroid composition prediction.
Competition history
- AJAS 2025
Related projects
ISEF · 2024
New Population of Asteroids Discovered! Rapidly Finding and Tracking Faint Near-Earth Asteroids Using an Accurate Deep Learning Based Process
ISEF · 2017
Spectral Measurements of Near-Earth Asteroids
ISEF · 2021
The Fast and Inconspicuous: New Near Earth Asteroids Discovered Using Deep Learning and Synthetic Data Are Fainter and Move Faster than Those Previously Discovered
CSEF · 2023
Detecting Potentially Life-Threatening Near Earth Objects in NASA Wise Data Using an Efficient Deep Neural Network
JSHS · 2022
A Novel Deep Learning/Machine Learning Hybrid Technique for Automatic Classification of Nebulae
CSEF · 2017
Saving the World, One Asteroid at a Time: A Comparison of Different Methods of Orbital Determination
ISEF · 2025
Optimizing Exoplanet Detection: A Comparison of Neural Networks and Traditional Machine Learning Models on Light Curve Data
CSEF · 2019
Galaxy Morphological Classification through Convolutional Neural Networks
Closest projects by meaning, across every fair and year in the corpus.
Browse more like this
Source: AAAS Annual Meeting (Confex) / American Junior Academy of Science