Research

My research focuses on developing self-supervised learning methods for hyperspectral imagery, with applications in ecological monitoring and precision agriculture. I work on creating methods that can learn from large amounts of unlabeled remote sensing data to improve performance on downstream tasks.

Self-Supervised Learning for Hyperspectral Imagery

Hyperspectral images capture hundreds of spectral bands, providing rich information about materials and vegetation. However, labeled hyperspectral data is scarce and expensive to obtain. My doctoral research explores how we can leverage self-supervised learning techniques to learn useful representations from large amounts of unlabeled hyperspectral data.

I’m particularly interested in adapting modern self-supervised learning approaches, such as Vision Transformers and Masked Autoencoders, to work effectively with the unique characteristics of hyperspectral data. The high dimensionality and spectral correlation in hyperspectral imagery present interesting challenges that require novel approaches beyond what works for standard RGB images.

Applications: Ecological monitoring, tree species classification, remote sensing analysis

Hyperspectral Imaging for Plant-Soil Systems

I’m also working on using hyperspectral imaging to understand plant-soil interactions at unprecedented spatial resolutions. Traditional soil moisture sensors provide point measurements, but hyperspectral imaging can potentially map soil water content at sub-millimeter resolution across entire root systems.

This research involves developing deep learning methods to predict pixel-level soil moisture from hyperspectral signatures, while handling challenges like spectral interference from plant roots. As part of this work, I contributed to the HyperPRI dataset to enable further research in this area.

Applications: Precision agriculture, plant phenotyping, understanding water dynamics in rhizosphere systems