Simen Berg is working on optimizing satellite operations. All satellites are controlled and tasked to perform specific operations, such as imaging, and defining the sequences is called scheduling. The research will bring forth frameworks to integrate optimization algorithms and utilize onboard processing for increased autonomy, with a goal to increase available resources on the satellite through e.g. energy-optimal control or making the satellite analyze the data and filter out captures of a poor quality metric.
As a case study, methods and algorithms from the research will be integrated and tested on the HYPSO satellites, and in practical terms, it aims to improve coverage and data acquisition of lakes such as Mjøsa.

Our understanding of bodies of water like Mjøsa depends on the collection and analysis of large amounts of diverse data. Automatic processing and clear representations of that data can therefore considerably ease the extraction of relevant information for various applications. Hence, I'm looking into how bodies of water can be visualized comprehensively in Extended Reality (XR) by applying methods from Computer Vision, Deep-Learning, and Human-Computer Interaction. 

The main challenges related to providing a user-friendly XR-Interface for the Digital Twin of Lake Mjøsa and similar bodies of water that I plan to address in this PhD project are twofold: Constructing a virtual environment that maps the landscape based on 2D images, especially those recorded underwater, and the visualization of additional information with spatial and temporal components within it. 

Project title: Unsupervised learning for hyperspectral remote sensing

Hyperspectral imaging (HS) is widely used in environmental applications, offering insights into the physical and biological properties of scenes. It helps in assessing health, detecting pollution, and measuring air and water quality. The technology has become more accessible due to cheaper and more efficient sensors and the use of drones in remote sensing. However, HS data are complex and challenging to analyse. Supervised methods need extensive annotated data for calibration or training, which is costly and hard to obtain. In contrast, unsupervised methods can extract valuable information without relying on ground truth data, making them cost-effective. This thesis will explore dimensionality reduction and semantic classification techniques to detect and monitor harmful algal blooms in Breton and Norwegian waters using multi-scale, graph-based, and unsupervised learning methods.

Imane Moulay Omar's PhD is a cotutelle, split between NTNU (Colourlab, IDI) and 

Her bio:
I hold a Bachelor's degree in computer science, followed by a specialized master in decision support and intelligent systems. These five years significantly contributed in building a solid background in a broad spectrum of computer science disciplines with a particular focus on data science and machine learning. My master project entitled "a study of combining remote sensing data and ground data for precision agriculture" has enriched my experience and provided me with a deeper understanding of the application of machine learning to remote sensing.