Computer Science: Fully Funded PhD Scholarship at Swansea University: Computer vision-based analysis of multimodal images of the Martian surface (Joint Research Study)

  • Full cost of UK/EU tuition fees, plus a stipend
  • Deadline: June 14, 2018
Visit Scholarship Website

This scholarship is funded by Swansea University and Communauté Université Grenoble Alpes.

Subject areas: Computer Vision, Machine Learning, Planetary Sciences, Orbital Imaging, Hyperspectral Imaging

Start date: October 2018 or January 2019

Context:

Successful Mars observation missions in the last two decades have acquired large image datasets. These images allow research on the Martian atmosphere and surface, including climate, geology, geomorphology, mineral distribution, etc. They are extremely heterogeneous, covering different spectral domains and having huge differences in their spatial and spectral resolutions. They are also complementary, and the fusion and joint interpretation of this multimodal data are fundamental for the characterization of the planet. However, this joint analysis is still in a stage of infancy since products coming from the separate analysis of individual instrument datasets are most of the time simply superimposed during visualization. The supplementary information provided by combining complementary images of different sources is thus not well exploited. In addition, the dramatic increase in the amount and complexity of the data calls for the development of new intelligent analysis methods.

Work packages:

The proposed multidisciplinary project will address these shortcomings with new computer vision and machine learning methods, combined with physics models. They will address the joint analysis of multimodal images through the following steps: (i) precise orthorectification, joint co-registration of images and topography estimation; (ii) radiometric/atmospheric correction of the Martian images to enhance surface features and to improve the compatibility of the different images; (iii) fusion of multi-modal images at the data level (pansharpening); (iv) characterization of terrain properties from the fused multi-modal images; (v) visualization, joint interpretation, and cross-validation in order to create an augmented and reliable view of the Martian surface.

The fusion of multimodal images requires their fine co-registration as a pre-requisite, despite their large resolution differences. The traditional methods, which align onto a reference image by matching control points, are too inaccurate on Martian images due to the paucity of recognizable ground control points on Mars. We will seek to combine registration and 3D modelling for a greater robustness and accuracy.  The registered images will then be fused by adapted pansharpening methods developed to produce high resolution (multi-) hyperspectral images. The fused multimodal images will contain rich information on the terrain type and composition. We will assess the capability of deep learning models, trained and validated with synthetic images, to segment terrain types and, at the same time, estimate the pixel-wise abundance of minerals of interest. Planetary scientists need to explore the relationships between different surface properties (topography, roughness, composition, etc.) to understand geological and climatic processes. This interpretation will be facilitated by a new fusion and visualization strategy for the features extracted from the images.

An immediate application of this research will be the morphological, compositional, and textural characterization of sites representing various geological contexts involving different periods in the history of Mars. This information will allow a better understanding of the Martian environment past and present, the possible appearance of a prebiotic activity and its current habitability for humans. In addition, our developments will also advance the state-of-the-art in visual computing for scientific image analysis, which may then be generalized to other future applications.

Training:

The PhD candidate will receive a multidisciplinary training in visual computing and machine learning at Swansea University, and in computational physics and planetary imaging at Université Grenoble Alpes. The team of supervisors will assist in developing multidisciplinary team work skills as well as technical experience in their respective specialties. The student will be able to engage in any combination or all WPs based on their background and interests, under the guidance of the supervisors who will ensure the coherence and quality of the final doctoral project.

Required skills:

This project combines advanced methodological activities in computer science, physics, and mathematics. It requires solid knowledge in image processing, signal processing, and simulation. Data processing requires proficiency in programming (C++, Matlab, Python) and image manipulation. Knowledge and experience in machine learning and deep learning are recommended. The use of geographic information systems is a plus as well as knowledge in planetary radiative transfer, orbital navigation, and ephemerides. Minimum training in general planetary science would be appreciated.

Resources available:

Within the Department of Computer Science at Swansea University, you have access to the computational resources of the department and to the College of Science (CoS) Doctoral Training Centre (DTC) community that provides training and support for PhD students.

Within the IPAG, you have access to the "laboratory of planetary imagery" equipped with a workstation hosting a Martian Geographic Information System and with a GPU workstation for the production and visualization of high-resolution digital elevation models. IPAG and OSUG servers for other kinds of computation and processing.

Eligibility

As this is a joint degree, applicants must meet entry/funder requirements of both Universities: a master's degree or equivalent in computer vision, machine learning, planetary imaging, physics modelling and simulation. Candidates should be able and willing to broaden their experience and skills across the computer science and physics domains.

Due to funding restrictions, this scholarship is open to UK/EU candidates only.

We would normally expect the academic and English Language requirements to be met by point of application. For details on the University’s English Language entry requirements, please visit – http://www.swansea.ac.uk/admissions/englishlanguagerequirements/

Test your English

Prepare for your study by taking the most popular and most accepted English language proficiency test. Schedule an IELTS test >

Benefits

The scholarship covers the full cost of UK/EU tuition fees, plus an annual stipend of £14,777 (50% Swansea University and 50% Université Grenoble Alpes).

There will be an additional £500 per annum from Swansea University to assist with travel and accommodation.

Application

Please visit our website for more information.

Sign Up for Scholarship Updates

Get an email every week that 10.000's of students use to get the latest scholarships.

Name
Email