Automated Underwater Imagery Analysis

There are increasing possibilities of acquisition of large amounts of imagery for a range of environmental characterisation applications. One of the areas of interest is the marine environment where Gardline has been acquiring a large collection of imagery during their surveys including both video and stills. This data has been captured as part of their worldwide surveys, from shallow water estuaries to the deep sea continental margins and abyssal plains. Analysis of this large data collection poses significant challenges for human observers due to the size of the data, increasing acquisition rates and burdensome nature of the task. Clearly, there is a need for automated tools that would aid humans in analysing such imagery. Consequently, robust automated tools would open opportunities of more rapid and cost-effective data analysis and allow use of more data-heavy imagery acquisition techniques, such as automated underwater vehicles.

This project will aim to develop automated computer vision algorithms for categorisation of the marine environment imagery. This task will initially take the form of identifying statistically-distinguishable sets of images or add further information such as identifying specific habitats, fauna group and eventually also species where appropriate within the varied dataset.

The envisaged system will require a large dataset of annotated imagery for training and this will require some expert knowledge on the image appearance of the relevant objects and habitats. The project would look to use the existing datasets held within Gardline to identify suitable approaches. Many of the datasets comprise thousands of images, which have been manually identified and categorised into distinct habitats, with some also having individual species or habitat features digitally annotated. The data would primarily comprise benthic images acquired in the wide array of geographical locations using various underwater cameras and deployment systems.

We will employ various machine learning and computer vision techniques to achieve the aforementioned task. 'Deep Learning' (DL) is a family of algorithms usually utilising convolutional neural networks (CNNs) that has been recently reported to provide a significant improvement in performance in many computer vision applications. The key feature of DL and CNN based algorithms is that they replace the step of designing handcrafted features in the prior-art algorithms with the automated hierarchical feature learning. As part of their PhD, a successful candidate will investigate development and application of such algorithms in the field of marine imagery categorisation.

The NEXUSS CDT provides state-of-the-art, highly experiential training in the application and development of cutting-edge Smart and Autonomous Observing Systems for the environmental sciences, alongside comprehensive personal and professional development. There will be extensive opportunities for students to expand their multi-disciplinary outlook through interactions with a wide network of academic, research and industrial / government / policy partners. The student will be registered at University of East Anglia, hosted at School of Computing Sciences in the Graphics, Vision and Speech laboratory. The student will receive training in all areas relevant to the project including computer vision, machine learning as well as Matlab and Python programming. The student will spend periods of time at Heriot-Watt University and Gardline in order to familiarize with the images and the environmental aspects of the project.