The objectives of this project are:

1. to investigate potential vulnerabilities of a CubeSat constellation (i.e., a network of CubeSats orbiting together);
2. to investigate potential cyberattacks exploiting these vulnerabilities of a CubeSat constellation, especially those that have a high effectiveness-to-cost or high reward-to-cost ratio; and
3. to devise countermeasures against these cyberattacks in a learning-based framework.

Machine learning has achieved great success in computer vision, natural language processing and other fields, especially in the accuracy of prediction and some may have exceeded the human capabilities. Nevertheless, users still need to understand the reasons for their conclusions in a more detailed and tangible way in application scenarios. Giving strong explainability to the model also helps to ensure its robustness, and usability of the method. This proposal focuses on developing interpretable models for early smoke wild-fire detection from satellite images.

Detecting fires at their early stages is essential to prevent fire caused disasters. Research has been conducted to detect smoke in satellite imagery for fire detection. Unfortunately, the imagery data used in previous research have low spatial resolution and only contain the RGB bands, which are ineffective for early fire detection. Our team (Data Analytics Group at UniSA) has been working on early fire smoke detection with multispectral multi-sensor satellite imagery for one year and the accuracy of wild fire detection can reach more than 90% accuracy. An AI framework of deep learning neural networks that identifies wild fires has been developed. It is necessary to present users with understandable reasons for the detection so the users can validate the detection and assess its severity. Current detection models are deep learning-based and have a black box detection kernel. This project aims to make the detection transparent to users so users can use and interact with the model easily. In the following, we will introduce the main techniques for explaining predictions made by deep learning / black-box
machine learning models.

This project aims to develop a dynamic system of analytical techniques operating on multiple modalities of data in Earth Observation. It will use insitu sensing to conduct on-board analytics, hence reducing the required bandwidth costs of communication. Reducing the need for human agents to interpret and respond to gathered data will give satellites advanced autonomy, accelerating the gathering and transmission of higher-fidelity data. While this project will target ISR, the resulting system can be configured to any domain, and can be quickly customised for industry use. Areas that can benefit include Disaster and Emergency Management, where tracking and responding to an evolving situation quickly means crucial information is communicated efficiently. Water and Environmental Management too, where real- me hydrological analysis is needed to gauge Mow changes, or real-time monitoring of forests and Agricultural crops can detect climate events or pest infestations. This project will also investigate the e1ects malicious actors can potentially achieve against the system. Exploring these and their associated defences will contribute to satellite resilience. Utilising multiple modalities of data as a defence and efficiency technique will also demonstrate a novel use of data fusion.

This research focuses on how to perform task offloading in LEO satellites. When a task arrives at an LEO, LEO can process this task itself or offload it to another LEO. This decision is made by considering several constraints like latency, energy, and queue. Task offloading requires LEO satellites to cooperate and make decisions based on local information, which aligns with the priority research area of Trusted Autonomous Satellite Operations. Also, to enlarge the capacity of the satellite at the same time, we consider the implementation of massive MIMO in the current system. Due to the high computational complexity of using a model-based approach to solve task offloading problems, we propose using a deep reinforcement learning-based method that can directly provide actions by interacting with the environment. To gain knowledge of the environment, we proposed a GCN-based scheme to learn the channel state information, which aligns with the subtopic of Explainable AI for Satellite Operations. LEO can provide global coverage, high-throughput, low-cost service to remote users. As for in remote area, satellite communication is the only low-cost way of communication. By increasing the capacity and computation speed of LEO, intelligent agriculture in vast remote area as well as disaster prediction and early warning can be made possible by LEO communications. Therefore, this research is beneficial to many important areas of application.

There are currently a number of issues with deploying AI onboard satellites, some of the largest being the efficiency and adaptability of the Neural Network. Compounding this, efficient onboard AI solutions are not scaleable for use in satellite constellations. Adapting novel techniques from prior state of the art research, namely “Once-for-All: Train One Network and Specialize it for Efficient Deployment”, this project aims to develop state of the art methods and approaches for implementing efficient onboard space AI in the context of Earth Observation nanosatellite constellations. This directly relates to SmartSat’s vision to be recognised as a leading contributor in transforming Australia’s space innovation, as well as SmartSat’s mission to create game-changing technologies.

This research aligns with the SmartSat research Program 2 as the methods developed have broader applications within machine learning applications in space. This could enable effective AI solutions for each of the topics mentioned. Additionally, the research also aligns with research program 3 as we are focusing on the specific application context of using AI systems to analyse satellite generated data in real time.

This research will focus on the onboard AI problems within the context of Earth Observation, making its potential applications as diverse as Earth Observation applications. Notable applications include, agricultural monitoring, disaster warning systems, and environmental system monitoring and management.

This project aims to develop a machine learning (ML) model that fuses satellite imagery and in-situ data for water monitoring and analysis for the benefit of water and environmental management. The result of this study will enable national monitoring of water bodies by using data fusion from satellite images and in-situ data; thus, limiting the risk of bush fires damaging ground-based water monitoring infrastructure.

This research aligns well with the AquaWatch Demonstrator as it develops artificial intelligence (AI) algorithm that will provide accurate forecasting of water degradation in Australia. The AI will map major environmental events such as algae blooms, bushfires, and other meteorological events to the degradation of Australian reservoirs, waterways and coastal environments. The proposed models will combine space and ground segments into an integrated metric for both near real-time decision-making as well as long term analysis and future forecasting.