Cookies

We use cookies to ensure that we give you the best experience on our website. You can change your cookie settings at any time. Otherwise, we'll assume you're OK to continue.

Durham University

Department of Geography

Deep Learning and Computer Vision Techniques for Automated Microscopy in the Processing of Microfossils within Geoscience

Diatoms found in saltmarsh sediments from Greenland

Diatoms found in saltmarsh sediments from Greenland

Supervised by Dr Sarah Woodroffe & Dr Toby Breckon (School of Engineering and Computer Science)

Understanding historical changes to our environment is key to predicting future environmental patterns but this requires obtaining long-term records (over hundreds to thousands of years) by reconstructing environmental activity via proxies (microfossils).

Pollen and diatoms (tiny animals) are deposited in salt-marshes, peat bogs and lake basins building up over time to leave a rich environmental archive. We sample this archive by taking a sediment core, partitioning the core into time slices and counting the microfossils that remain (e.g. pollen, diatoms) - their relative levels of occurrence allows us to reconstruct the conditions at the time of their death and deposition. However, most of these microfossils are very small and have to be viewed under a powerful microscope (at 400x magnification). Identifying and counting microfossils is hence an extremely time consuming process, both to learn to identify the different sub-types and to count them in sufficient quantities to produce a statistically robust indicator of prior environmental conditions.

This project would look to leverage recent advances in computer vision based automatic image/pattern recognition, enabled by recent advances in deep learning, to develop an identification and classification algorithm and work-flow for this task. Leveraging such techniques, highly novel within the study of geoscience,this project will look to take advantage of these advances in information processing from computer science to automate the accurate identification and quantification of individual taxa (sub-types) of diatoms in the sediment samples based on their size, shape and ornamentation. This would significantly improve sample throughput, reduce variability in (otherwise manual) sample processing and increase the statistical robustness of microfossil based methodologies. Applications of a computer-based microfossil identification and counting solution include palaeo-environmental reconstruction but also petroleum exploration and forensic geoscience.

To apply for this project please visit the How to Apply page for further information.