Exploring the utility of deep learning methodologies in the automated surveying of cliff nesting birds

点击翻译按钮获取中文摘要

Advances in artificial intelligence have enabled compelling capabilities that exceed the efficiency and accuracy of solutions attainable to humans. Applying such techniques in the context of wildlife monitoring has been shown to enhance the ecological picture beyond the means of manual observation and holds the potential for even greater advancements in the future. Due to their high trophic level, seabirds may be used as an indicator species within the marine environment. As a pathway to understanding the wider strength of the ecosystem, the monitoring of the health of bird populations serves as an invaluable biological resource. There are many factors that have the potential to adversely affect these species such as offshore industry, sea pollution or disease. However, the current standard framework for the collection of such data is often limited to manual observation. With the potential dangers associated with accessing the remote locations necessary for the data collection, the strenuous nature of the operation incurs a great cost. Further, there is often restricted opportunity for the standardisation and accuracy of results due to inconsistencies between observers and data collected being constrained to their viewpoint. The logistical implications pertaining to such methods also hinder the amount of data that may be collected, limiting the breadth of the survey and resulting in an unsuitably vague picture of the overall ecosystem. With recording intervals and accuracy being dictated by the efficiency of the data collection methods, streamlining the process through technological means holds the opportunity to demystify the ecological processes underpinning the observable cliff nesting bird dynamics. The aim of this project is to create, test and improve advanced deep learning architectures that will be tailored to expand the current bounds of ecological understanding of cliff nesting birds.