This project aims to investigate unsupervised learning techniques for efficient hyperspectral image segmentation. The project will explore clustering and dimensionality reduction methods to efficiently process large and high-dimensional datasets, emphasizing graph-based approaches.  Goals: - Develop a comprehensive understanding of unsupervised learning techniques for hyperspectral image segmentation through literature review. - Develop new approaches to identify modes and coresets in the pixel distribution (i.e., subset of most representative pixels) by exploring the potential of combining multiple graph-based features and similarity measures such as the minimax distance or the number of shared nearest neighbours.  Tasks: - Conduct a literature review on unsupervised learning techniques for hyperspectral image analysis. - Implement clustering and dimensionality reduction methods, focusing on graph-based approaches. - Create a new methodology to combine multiple graph-based features for efficient mode detection and coreset identification. - Analyse and evaluate the performance of the developed methods using real-world remote sensing data. - Prepare a report detailing the methods, results, and conclusions of the project.

Links
Le Moan, Steven, and Claude Cariou. "Minimax bridgeness-based clustering for hyperspectral data." Remote Sensing 12.7 (2020): 1162.

Contact
steven.lemoan@ntnu.no