Automated measurement of social movements in pheasants

Institution: University of Exeter

Lead Supervisor: Dr Joah Madden

Project Description

Collecting simultaneous, time matched, behavioural measures from multiple individuals is difficult, traditionally requiring a high degree of human effort and a corresponding risk of conflating human error. The deployment of automated behavioural measurement reduces such costs. An individual’s behaviour can be inferred from patterns of activity revealed by accelerometer traces, and individual ordering can be established by determining which individual moves first. This is a general problem, but also one specifically related to my existing ERC funded research project on the ecology and cognition of pheasants.

This Placement would constitute an independent facet of the main project yet benefit from the infrastructure and resources already in place. The work would take place on our field site and I would supervise the student directly, and s/he could also benefit from interactions with my team of post-docs and PhD students. We have established marked bird populations, computers and lab space the student can use. Part of the RTSG would be used to buy 5 new accelerometers to supplement the 10 already available from the project and a HDD to store data. Other equipment is supplied by the project.

The Placement involves fieldwork, with the student being responsible for the tagging of birds and the retrieval and processing of the ensuing data. The student will use Machine Learning methods to analyse the resulting large datasets. Each bird is monitored at 100Hz for 24hr periods, so with 15 accelerometers, logging ~20 days, I calculate ~300GB of movement data being generated. The student will gain skills in: practical bird handling and husbandry; direct behavioural observation; video analysis and synchronisation; machine learning focussed on supervised learning techniques; explicit hypothesis testing using conventional statistics. If the project progresses as planned, there will be an opportunity to contribute to writing the work up for publication.

Indicative timescale for project (subject to change)

During weeks 1-4 the student will collect representative behavioural data. This will entail responsibility for the deployment and retrieval of accelerometers, catching birds and attaching the tags to backpacks. The student will extract and archive the data. The student will be trained in bird catching and handling skills to ensure competency required by the Home Office Licence governing the project. The student will also deploy video cameras to collect training data, and using ELAN software, score the resulting videos and ensure their correspondence with the accelerometer data. The final six weeks will be spent coding the videos, developing the ML algorithms and making explicit tests of model accuracy. The student will remain at the study site with access to the test birds if refinements are required, and in contact with others on the project team to ensure biological grounding of data.

Powered by WordPress. Designed by WooThemes

css.php