Mathematics to the Rescue: Addressing Deficiencies in Overhead Imagery Products

Overhead imaging or remote sensing is a wide ranging field which uses imagery products acquired by distant sensors to ascertain information of an area or target of interest.  Due to low spatial resolution, atmospheric effects, wide ranging collection geometries, and limited labeled data developing generalizable algorithms is difficult.  A recent trend has been to move away from traditional statistical and black box deep learning models to techniques which are inspired and informed by rigorous mathematical analysis.  These mathematical techniques borrow from theoretical results in geometry, topology, harmonic analysis, and algebra to better capture the intricacies of difficult sensing problems and leverage the limited labeled samples.  We will consider a wide problem domain including target and anomaly detection, atmospheric characterization, land cover analysis, and change detection but ask speakers to apply their techniques to well characterized data.

Organizers:

Tim Doster, Pacific Northwest National Laboratory

Adam Attarian, Pacific Northwest National Laboratory

Tegan Emerson, Pacific Northwest National Laboratory

Henry Kvinge, Pacific Northwest National Laboratory