Acoustic Identication of Marine Cetaceans using Deep Learning Techniques

点击翻译按钮获取中文摘要

Modelling cetacean (dolphins, whales and porpoises) dynamics and behaviour is paramount to effective population management and conservation. Robust data is required for the design and implementation of conservation strategies and to assess risk presented by anthropogenic (man-made) activity such as offshore wind turbines and fishing. Moreover, cetaceans make prime candidates for modelling ecosystem change under the ecosystem sentinel concept as they reflect the current state of the ecosystem and respond to change across different spatial and temporal scales. As global climate changes and urbanisation of coastal areas intensies, it is imperative to develop methodologies for quick and effective assessment of the biological and ecological impact of rising sea temperatures, pollution and habitat degradation. This can be achieved through modelling the population, behaviour and health of large marine species such as dolphins.Methodologies of cetacean research include passive acoustic monitoring (PAM). Deployed PAM is less expensive and labour intensive compared to other marine mammal methodologies and can record data over long temporal scales. For dolphins, this is limited to the monitoring of echolocation clicks and some species identification using the whistles they produce, unable to identify individuals within a species and difference between certain species (such as bottlenose and white-beaked dolphins). The research will addresses these limitations by applying the methodologies and techniques of deep learning to the field of marine biology, mainly focusing on image analysis techniques. Methodologies will be designed to quickly identify individuals, analyse behaviour and incorporate remote sensing techniques. From these identifications estimations of group size will also be made. A statistical framework will be developed, which can be incorporated into PAM systems, to monitor species and individual occurrence over long spatial scales.Data collection will focus on a population of white-beaked dolphins (WBD) off the coast of North-East England. Recent research has identified sites where the species is regularly sighted and has shown seasonal and multi-year residency. A health assessment identified high incidence of skin disease and trauma suggesting conservation of this population should be high priority. Moreover, there has been recent evidence of this species producing unique whistles, however more evidence is needed, which this research will provide.The research will develop novel acoustic analysis algorithms which can be deployed on small low-cost Linux based computers, to be taken into the field to perform analyses in real time. These algorithms will be developed to work on both visualised and raw acoustic data collected from WBD, consisting primarily of whistles. The process for initial data collection has already begun using hydrophones to record dolphin vocalisations during expeditions off the coast of the North-East England. Initial visualisation of acoustic recordings have also begun, but experimentation with depth and distance from vessel placement of hydrophones are needed. The research will develop a novel method and statistical framework using deep learning models for acoustic analysis. This objective will be evaluated by testing the trained algorithm at successfully identifying group size or individuals from acoustic recordings.Another aim of the research is to assess the effect of anthropogenic activity on cetaceans, mainly wind turbines and boats. There is a large collection of wind turbines off the coast of Blyth where the collection of WBD and other cetaceans habitat. There has been no studies of how the noise pollution from these turbines affect the population but recent research has shown trauma to the white-beaked dolphins suggesting this could be a possible cause.