Current Research Projects

1. Real-Time Georectification of Visible, Thermal, and Hyperspectral Images

Our unmanned aerial vehicle captures images of the ground and water by utilizing a hyperspectral imager and high resolution visible and thermal cameras. The onboard GPS and IMU capture the position and orientation of the aircraft in real time, but how can we determine the location (latitude and longitude) of each pixel in the captured images?

2. Super-Resolution

The images captured by each imaging platform have different properties including spatial resolution, spectral resolution, and temporal resolution. How can we utilize (physics informed) machine learning techniques to cross achieve the maximum possible resolution in each of the categories? Can we infer reflectance spectra from an RGB image within a limited region and time window?

3. Machine Learning for Cross Platform Calibration

Our aerial vehicle samples surface reflectance spectra from the air. Meanwhile, our autonomous boat directly measures physical and chemical water properties including temperature, salinity, turbidity, hydrocarbon concentrations, and chlorophylls. Can we use physics informed approaches to map the remote observations of the aerial vehicle to the direct observations of the boat?

4. Scientific Machine Learning

Deep neural networks and many other machine learning methods behave as a black box; Carefully curated inputs are fed into an opaque model whose internal parameters are tweaked until sufficient agreement with target observations is achieved. How can we marry our existing scientific knowledge with machine learning methods so as to not throw out what we have already characterized through rigorous experiment? Can we achieve higher model generalizability by utilizing scientifically informed constraints and learning strategies?