Machine Learning Project Work

The majority of early machine learning work done by our researchers has been focused on the use of machine learning techniques to detect and classify features in images, video, and sound data (object detection). Image classification assigns one or more labels to the whole image, whereas object detection specifies the location of a particular object within an image and assigns it a label. Examples of this work include detection of fish species from video collected onboard fishing vessels, detection of Crown of Thorns starfish from towed video, detection of litter in water ways, and the detection of blast fishing events from audio collected by deployed hydrophones.

Several domains are using machine learning techniques for anomaly detection and behaviour classification. Historically there were significant investments in capability and technology to drive this work using Vessel Monitoring Systems (VMS) / Automatic Identification System (AIS) for vessel detection and analysis of behaviour.

There is a growing use of machine learning in modelling and forecasting work, including a move towards data-driven modelling, and forecasting to complement existing physics-driven models, particular for parameter estimation and gaining an understanding of complex systems.

We are using Nature Language Processing (NLP) to develop semi-automated approaches to monitor and understand social acceptance, social change and social conflict concerning marine resources.

As well as these more easily defined areas, machine learning is also widely used for data analysis, cleaning, and mining, in the same way that statistics is widely used across the program.