Power Efficient Deep Learning Models for Environmental Monitoring at the Edge

This is an exciting opportunity to showcase how deep learning can benefit Antarctic research and remote biological monitoring. Working in the Securing Antarctica’s Environmental Future (SAEF) program, this ARC-funded, multi-disciplinary PhD project aims to develop a set of novel low power, extremely accurate, deep learning algorithms to process a stream of data coming from various sensors, including a camera, deployed in Antarctica and extract (visual) information about the environmental conditions (moss he

Antarctica is one of the most challenging environments on Earth for plants to live. Plants need sunlight, warmth, and liquid water to be able to grow and survive. Antarctica is dark for half of the year, extremely cold, and water is frozen most of the time. Most plants are unable to survive this environment, but moss can because they have different physiological strategies to other plants which have enabled them to form large green turfs that stand out amongst the frozen white and grey landscape.

Because they are the dominant plant life in Antarctica, mosses are extremely important for providing habitat for invertebrates, microbes and fungi which make up more than 99% of terrestrial biodiversity in Antarctica. The moss beds are miniature forests.

However, long term monitoring has revealed that these moss forests are stressed in many places and have been dying off, but there is evidence of recovery as well, and this proposal seeks to monitor the moss and their environment, leading to an understanding of what drives the dying and recovery of the mosses, and whether we can predict these processes remotely for the purpose of biological monitoring.

The goal of this proposal is to develop a set of novel low power, extremely accurate, deep learning algorithms to process a stream of data coming from various sensors, including a camera, deployed in Antarctica and extract (visual) information about the environmental conditions (moss health, snow cover, water in the field of view, fauna, …).

Those sensors are part of an Artificial Intelligence of Things (AIoT) remote sensing platform currently being developed within the Securing Antarctica’s Environmental Future (SAEF) program, and designed around an NVIDIA Jetson edge computer enabling AI processing.

The AIoT platform aim to be deployed for a long-term period, performing real-time monitoring. Thus low power algorithms is critical for enabling this long term operation. This can be achieved for instance by investigating distillation, pruning, quantization, compression techniques as well as novel architectures.

A comprehensive long-term data set is available for developing and validating machine and deep learning models. The data set spans decades of research and includes photographs of the mosses, moss health, moss physiology, water availability, microclimate temperature, microclimate light intensity, multispectral drone imagery, and hyperspectral ground-based imagery.

This is an exciting opportunity to showcase how deep learning can benefit Antarctic research and remote biological monitoring.

This project is part of the Securing Antarctica’s Environmental Future (SAEF) program, a collaborative partnership of international researchers and practitioners that will deliver research to forecast environmental change across the Antarctic region, to deploy effective environmental stewardship strategies in the face of this change, and to secure Antarctica as a natural reserve devoted to peace and science.

SAEF is an Australian Research Council (ARC) Strategic Research Initiative and acknowledges the significant investment from the ARC and contributing organisations.

Duration:

4 years (or full time equivalent)

Application process:

Interested applicants need to prepare the following:

• a one page cover letter outlining relevant experience
• a Curriculum Vitae (three page max.)
• The most recent academic transcripts
• contact details for two academic referees

Applicants can begin in Autumn 2025

Eligibility requirements:

• Knowledge in deep learning for computer vision
• Good programming skills (PyTorch, Tensorflow)
• Ability to work in a multidisciplinary team environment, as well as independently
• A Masters or Class I Honours (or equivalent) undergraduate degree in a relevant field such as Artificial Intelligence, Computer Science, Information Technology, or Environmental Science.