ASTROSENSE: applying astrophysics algorithms to remote sensing data
Lead Research Organisation:
University of Hertfordshire
Department Name: School of Physics, Astronomy and Maths
Abstract
Machine learning is a computational data analysis technique that offers tremendous benefits over traditional methods. In particular, algorithms can be developed that can automatically identify objects or features of interest in digital imaging. Some machine learning algorithms require extensive training through labelled examples, but unsupervised algorithms can learn from the data itself, requiring no pre-labelled training set. This makes such algorithms incredibly versatile and can be easily applied to many different types of imaging. In principle, the algorithm's performance should improve over time as it 'experiences' more examples of input data. We are developing just an unsupervised machine learning algorithm for use in large-scale astronomical surveys that can also be applied in other 'remote sensing' data, such as underwater sonar imaging of the sea bed and aerial/satellite imagery. Such an algorithm can, for example, help determine the local terrain and identify hazards in complex, changing environments that could be missed by a human inspector. This could feed into AI-assisted navigation units in autonomous vehicles for example. Our goal in this project is to develop a versatile, robust algorithm that can be deployed in a variety of practical areas, with a view to performing real-time image classification and analysis on input data, both from astrophysics and 'real-world' industrial sectors.
People |
ORCID iD |
James Geach (Principal Investigator) |
Publications
Spindler A
(2021)
AstroVaDEr: astronomical variational deep embedder for unsupervised morphological classification of galaxies and synthetic image generation
in Monthly Notices of the Royal Astronomical Society
Description | Novel machine learning algorithms that can be applied to tackle issues of unsupervised detection of objects/features in multi-spectral imagery. |
Exploitation Route | Versatile algorithms that could be applied to many types of imaging data, for example, one thing we are looking at now is real-time labelling of light scattering data for particle classification for airborne pollutants. |
Sectors | Aerospace, Defence and Marine,Agriculture, Food and Drink,Environment |
Description | Member of the Institute of Physics and Royal Astronomical Society LGBT+ Working Group |
Geographic Reach | National |
Policy Influence Type | Membership of a guideline committee |
Title | AstroVaDEr - Astronomical Deep Embedder for Galaxy Classification and Synthetic Imaging |
Description | AstroVaDEr is an implementation of a machine learning algorithm which performs unsupervised classification of galaxy morphology, by utilising a variational autoencoder and a learned Gaussian mixture model. The model can be used to rapidly classify the morphologies of nearby galaxies, and generate qualitatively realistic images of galaxies. The code is available via the following DOI: https://zenodo.org/record/4034802#.YDyxn5P7TUI |
Type Of Material | Computer model/algorithm |
Year Produced | 2020 |
Provided To Others? | Yes |
Impact | AstroVaDEr demonstrates a powerful potentital tool in the coming years for next-generation astronomy, and highlights many difficulties and approaches to solve them. In review, the work was said to "[reveal] fundamental insights and challenges in clustering galaxies with VAEs and with unsupervised learning in general, which are important to share with researchers working on the problem of clustering galaxies" and that it "guides the way for the field to address this difficult challenge". |
URL | https://zenodo.org/record/4034802#.YDyxn5P7TUI |
Description | National Student Space Conference - UKSEDS - Invited Talk |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | National |
Primary Audience | Undergraduate students |
Results and Impact | Invited to speak on the applications of machine learning in astronomy to the annual careers event hosted by UKSEDS. Hosted online, a prerecorded video was shown, followed by a short Q&A. |
Year(s) Of Engagement Activity | 2021 |
Description | Physicists Like Me - IOP/Sepnet |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | National |
Primary Audience | Undergraduate students |
Results and Impact | A panel of Researchers in academia and industry talked about their experiences in their careers, how they got there and how their diverse identities played a role. Undergraduate and postgraduate students were able to have small group chats with the panelists and discuss their options and career desires. |
Year(s) Of Engagement Activity | 2018 |
Description | Pride in STEM Research Showcasew |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | Regional |
Primary Audience | Postgraduate students |
Results and Impact | Six speakers from various career levels in Academia gave talks on their research and experience as queen scientists in STEM. The day took place at the University of Loughborough and was co-organised by the charity Pride in STEM. The day was attended by staff and students from the University of Loughborough, and other individuals from regional universities. Discussion revolved around the speakers research, and on policies for improving LGBT inclusion and diversity. |
Year(s) Of Engagement Activity | 2019 |