Project Description
Supervisors
Professor Michal Mackiewicz, Computing Sciences, University of East Anglia – contact me
Professor Graham Finlayson, University of East Anglia, School of Computing Sciences
Mr Peter Kohler, Cefas
Dr Julie Bremner, Cefas
Scientific background
Marine litter is a key threat to the oceans’ health and the livelihoods that depend on it. Hence, new scalable automated methods to collect and analyse data are needed to enhance our understanding of sources, pathways and impact of litter. Cefas is developing a visible light (VL) deep learning (DL) algorithm and collected a large 89 litter category training dataset. However, there is a recognition of the need of multispectral imagery to enhance the accuracy of the algorithms being developed when discerning material type. Consequently, Cefas is developing a new lab facility to assist in characterisation of multispectral reflectance of materials.
You will develop existing VL work on reflectance signature of materials extending it to multispectral imaging. The development of robust litter detection and classification algorithms will require an approach that considers the physics of the multispectral image formation including the three key variables: sensor spectral sensitivities, varying daylight illumination spectrum and wide range of relevant material reflectance spectra.
Research methodology
You will utilise the existing VL database of key materials, but importantly will also collect multispectral data with the enhanced lab setup with an aim to train the DL algorithms. Importantly, the algorithms developed must be robust to changing real-world illumination and utilised long-term, likely with imaging devices not existing during the development. Therefore, the algorithms are required to have a level of independence to the number of multispectral channels available and their spectral sensitivities. The research will examine several approaches including device independent data representations and/or various transfer learning and domain adaptation techniques.
Training
You will be based at the Colour & Imaging Lab at the School of Computing Sciences which has expertise in the design and evaluation of imaging solutions and will have an opportunity to work with scientists and engineers at Cefas. You will undertake training specific to this project including imaging principles, lab measurement, computer vision and ArcGIS, potential fieldwork and UAV flying training.
Person specification
Experience and/or enthusiastic interest in one or more of the of the following areas interest in environmental monitoring, AI, computer vision or multispectral imaging.
Acceptable first degree subjects: Computer Science/Physics/Maths or other numerate discipline