Space training team – Machine learning for recognition of planetary materials from multispectral datasets

Our team and mission

PANGAEA (Planetary ANalogue Geological and Astrobiological Exercise for Astronauts) is a geological and astrobiological field training course for astronauts and mission developers. It develops fundamental knowledge and skills in these disciplines, as well as observational and decisional skills such as identifying important geological features in the field, conducting efficient sampling and effectively communicating with scientists on Earth. Complementary to its training goals, PANGAEA is also used as a research and development platform to enable advances to be made in several technological, scientific and operational areas.  

CAVES (Cooperative Adventure for Valuing and Exercising human behaviour and performance Skills) teaches astronauts human performance and behavioural skills for exploration tasks through an expedition into a natural cave system. Cave systems present environments and situations very similar to spaceflight, making them ideal for this type of training. The scientific exploration conducted by the astronauts during CAVES also enhances human understanding of cave environments.  

Activities and learning areas

Geology and geo-microbiology research is a primary scientific goal in current satellite and rover missions to Mars, Moon and asteroids, and these disciplines will have an even greater importance in future manned planetary exploration.

In the coming years, European astronauts will be involved in defining and planning unmanned and manned missions underpinned by geological exploration. Consequently, knowledge about general geology and planetary geology will be crucial to their role as an astronaut. Additionally, many of the Earth Observation tasks currently performed from the International Space Station have a strong relationship with geology.

Future human missions that return to the Moon or land on the surface of Mars will require astronauts to perform geological exploration during Extravehicular Activities (EVAs). With the right preparation, astronauts will be required to quickly identify key mineral resources, perform on-site analyses and collect valuable samples. This process will be greatly enhanced by portable analytical tools (VNIR, Raman, LIBS, XRF, etc.) that allow rapid mineralogical and elemental analysis of prospective samples. To make these instruments effective, they must be equipped with the appropriate libraries and software to ensure the scientific data collected is accurate and easy to interpret. This will allow improved decision making and flexecutionto be applied by the astronauts, even without the support of the scientists from ground control. To enable this, our team is working to enable the Electronic Field Book (EFB) system to record, accurately classify, enriched with relevant information, and distributed to all relevant mission support teams, spectroscopic data collected in the field. Two key projects are currently being developed to help achieve this, the PANGAEA Mineralogical Database (MinDB), and a machine learning toolset (MLtools).

The MinDB is a structured set of information on all the rocks and minerals detected on the Moon, Mars and other planetary bodies, coupled with information on most effective methods to identify them in-situ. A key part of this is library of standard spectra. These spectra are collected based on available open access on-line catalogues as well as from our own bespoke spectroscopic measurements (VNIR, Raman, LIBS & XRF) of planetary analogue minerals from different collections. The spectral library is specifically designed to improve the classification of minerals by offering a bespoke database focused on planetary materials.

The main task of the internship is to work on the machine learning toolset (MLtools). This toolset works in tandem with the mineralogical database’s spectral library to classify spectra from minerals and rocks against multispectral datasets. This project focuses on combining together two types of material characteristics, a mineral structure (obtained with VNIR and Raman spectra) and its chemical composition (derived from XRF and LIBS spectra) to improve automatic mineral classification. Our current efforts are focused on developing efficient recognition of mineral mixtures via “unmixing”, which works by breaking down a spectral signal into a set of constituent spectra. The MLtools work is closely tied to the development of the EFB system, which will link with these algorithms with spectrometers in a deployable field support system.

Specific duties

Candidates will be assigned specific duties based on their background.

• Familiarise with the current version of the supervised classifier being developed in the team.
• Investigate applying state of the art machine learning algorithms for recognising  planetary materials from multispectral datasets.
• Continue development of the ML classifier for recognising planetary materials from multispectral datasets.
• Begin investigating incremental learning approaches for improving the algorithms for use during mission operations.
• Handle and pre-process of measurement data for importing to a database, and automating generation of datasets for training and validation.
• Get familiarized with the current version of the PANGAEA Electronic Fieldbook  system (EFB) and its technologies to assist in the integration of the classifier into that system.

Competencies and skills

Preferred educational background incomputer science, data science, software engineering, applied mathematics or statistics.

• Practical experience in the artificial intelligence context, or related to machine learning (or data mining, clustering, unmixing, recommender systems), or in analysis of data coming from analytical instrumentation, or with databases.
• Academic or professional experience with the programming languages and frameworks currently used in the project: Python, TensorFlow, Keras, Scikit-learn, matplotlib and Numpy. Additional experience in HTML, JavaScript, Jupiter notebook, AngularJS, REST paradigm, UNIX environment and the visualisation of scientific data/user interface design are considered pluses.

Having an interest in mineralogy, geology, (astro-)physics, (astro-)chemistry, or planetary science would be an asset.

Internship start dates are flexible. Candidates should be available for six months. Shorter internships can be negotiated on a case by case basis, the minimum length being four months. 

Applications are currently closed.

Interested students are asked to familiarise themselves with the general terms and conditions of ESA internships here.