PhD Studentship: Machine learning for geospatial intelligence

Ordnance Survey (OS) manage many diverse geospatial datasets used to address significant issues in Great Britain and internationally, such as mapping land use changes, estimating greenhouse gas emissions, planning new development and energy installations, and tracking ecological processes. Data held includes huge amounts of aerial photography and satellite imagery, alongside other spatial land use data such as buildings, transport networks, ecological features, hydrology and topology. This project will develop new machine learning (ML) tools to enable effective use of imagery alongside other data types. These tools will enable better decision-making on important societal goals around land use, environmental stewardship and renewable energy deployment. It will directly influence government policy delivery.

Experts at OS currently use the imagery datasets as part of a largely manual process to update their land use maps. OS is greatly interested in how machine learning can be used to make this process faster, or more accurate, or to add new details to the maps, for exampl, details of roof shapes or street furniture such as lampposts and traffic lights.

The imagery datasets held are huge. For example, regular aerial photography provides imagery of the whole of Great Britain at 25cm grid resolution. This presents a major challenge to the development of machine learning tools, in that the dataset is too large for efficient processing. For example, it can take up to three weeks of high-performance computation to train a single ML model for land use classification.

The aim of this project is to explore techniques for data sampling and pre-processing that will improve performance by retaining important information and reducing information redundancy. The student will work with world-leading researchers at the University of Exeter and Ordnance Survey to generate simplified datasets, train machine learning models, and establish reliable and efficient pipelines for data processing. Possible solutions to the problem might involve data compression, feature selection, linking to alternative datasets, or improving the efficiency of training. The student will develop advanced knowledge of image processing, neural networks, high performance computation with GPU arrays, data handling and geospatial techniques.

Ultimately, the success of the project will be determined by application of the tools to real-world challenges faced by Ordnance Survey in its role advising the UK government and other clients. Thus the student will gain broad experience of the end-to-end deployment of advanced ML and geospatial analysis.

The student will be based within the Centre for Doctoral Training in Environmental Intelligence at the University of Exeter, within an interdisciplinary cohort of postgraduate researchers studying diverse topics in environmental data science. They will also spend time on placement with the Ordnance Survey research team during the project. After completion of the PhD thesis, the student will be employed by Ordnance Survey on a 12-week postdoctoral contract to facilitate knowledge exchange and implementation.

The studentship is funded via the EPSRC Industrial Cooperative Awards in Science & Technology (iCASE) scheme, via a grant awarded to Ordnance Survey. For more details about the Industrial CASE scheme, see https://epsrc.ukri.org/skills/students/industrial-case/intro/

If you have questions or for more information, please email Dr Chunbo Luo ([email protected] )