PhD Project in Self-supervised summarisation of hyperspectral image data

Hyperspectral imaging is an important technology in medicine, biosciences, materials sciences, remote sensing, astronomy, chemical analysis, and many other areas. Whilst the cameras that we are familiar with from day-to-day use acquire images that have three colour channels (Red, Green, Blue, or RGB), hyperspectral imagers acquire images with tens, hundreds, thousands, or even millions of colour channels. The images they acquire can reveal the world around us in tremendous details, for example, by allowing the chemical composition of an object to be understood at the molecular level, or by identifying subtle differences in the “colour” of vegetation that might allow the fertility of farmland (for example) to be understood from satellite images. The tremendous information content of hyperspectral image comes at a cost: the images can require tens to hundreds of gigabytes to store, and not only is this practically challenging, it is very hard indeed to properly examine all of the data. It would be highly beneficial to be able to both reduce the digital size of the data, and to reduce its complexity so that it can be more properly examined by those who wish to use it. Many techniques have been used to achieve this, for example, principal component analysis (PCA) can be used to project the data into a low-dimensional space for easier analysis. This is often useful, but breaks down when the data has a highly nonlinear structure. Other, more sophisticated techniques such as t-SNE and UMAP can be used to generate low-dimensional visualisations of the data that can reveal some patterns, but which tend to severely distort the data and to hallucinate structure that isn’t really there. In this project we will develop techniques for summarising very high dimensional hyperspectral imaging data. Our goal will be to generate a single “summary” image that enables the image content to be rapidly visualised in a controllable way. Given the extreme complexity of the data, we cannot rely on labelled data and so we will approach this using self-supervised learning approaches to determine what the salient features in the images are, and to embed them into a single image space that permits a simple and interpretable visualisation of the data.

This project will develop self-supervised learning approaches to image summarisation in hyperspectral imaging data. The approach will be to use self-supervised techniques to produce a single summary image that retains the salient image features. Potential techniques of interest include diffusion models, contrastive learning, and variational autoencoders.

The project will involve developing i) novel deep learning methods for image summaraisation, ii) information theoretic approaches to identify which parts of an image should be preserved in a summary image, and iii) evaluation and application of the new methods in a range of potential application areas including hyperspectral remote sensing, high-throughput screening in drug discovery, or mass spectrometry imaging of tissues and materials.

Funding Information

To be eligible for consideration for a Home DfE or EPSRC Studentship (covering tuition fees and maintenance stipend of approx. £19,237 per annum), a candidate must satisfy all the eligibility criteria based on nationality, residency and academic qualifications.

To be classed as a Home student, candidates must meet the following criteria and the associated residency requirements:

• Be a UK National,
or • Have settled status,
or • Have pre-settled status,
or • Have indefinite leave to remain or enter the UK.

Candidates from ROI may also qualify for Home student funding.

Previous PhD study MAY make you ineligible to be considered for funding.

Please note that other terms and conditions also apply.

Please note that any available PhD studentships will be allocated on a competitive basis across a number of projects currently being advertised by the School.

A small number of international awards will be available for allocation across the School. An international award is not guaranteed to be available for this project, and competition across the School for these awards will be highly competitive.

Academic Requirements:

The minimum academic requirement for admission is normally an Upper Second Class Honours degree from a UK or ROI Higher Education provider in a relevant discipline, or an equivalent qualification acceptable to the University.

Entrance requirements

Graduate
The minimum academic requirement for admission to a research degree programme is normally an Upper Second Class Honours degree from a UK or ROI HE provider, or an equivalent qualification acceptable to the University. Further information can be obtained by contacting the School.

International Students

For information on international qualification equivalents, please check the specific information for your country.

English Language Requirements

Evidence of an IELTS* score of 6.0, with not less than 5.5 in any component or equivalent qualification acceptable to the University is required (*taken within the last 2 years).

International students wishing to apply to Queen’s University Belfast (and for whom English is not their first language), must be able to demonstrate their proficiency in English in order to benefit fully from their course of study or research. Non-EEA nationals must also satisfy UK Visas and Immigration (UKVI) immigration requirements for English language for visa purposes.

For more information on English Language requirements for EEA and non-EEA nationals see: www.qub.ac.uk/EnglishLanguageReqs.