SIParCS 2023 - Hayden Outlaw

Investigating Holographic Images of Clouds with Machine Learning

Recorded Talk

This project aims to improve the performance of a neural network processor for holographic images of cloud particles obtained using the HOLODEC instrument, an airborne cloud particle imager developed at NCAR that captures holographic images of liquid and ice cloud particles. A “U-Net” style neural network is used to recognize particles in the holograms after they have been computationally refocused. The objective of this project is to modify the neural net to reduce over-prediction and reduce data preprocessing requirements. Over the summer, the student will work with scientists in the CISL and the Earth Observing Lab (EOL) toward developing a new training dataset that utilizes multiple depth layers as additional “input channels.” Upon successful generation of the training dataset, the student will then modify the neural net to leverage this new data and train it. Depending on time, there may be opportunities to explore other potential solutions including mixed recurrent/computer vision modeling approaches as well as improving processing performance on the HPC systems at NCAR.

Mentors: John Schreck, Matthew Hayman (EOL), Gabrielle Gantos

Slides and poster