Gaussian Process Regression for CubeSat Data Interpolation with Uncertain Inputs and Multi-channel Outputs
Ruan, Weitong.
2018
-
Abstract: With their
greatly reduced sizes, low development cost and rapid construction time, CubeSats have
emerged as a platform of intense interest for a wide range of applications, including
remote sensing. Many applications require the interpolation of irregularly sampled
(scattered) sensor data onto a regularly spaced grid for the development of downstream
scientific products. This problem ... read moreis further complicated for the CubeSat platform by two
factors. First, there exist potentially significant uncertainties associated with the
spatial position of the satellite sensor, which if left unattended, will negatively
impact the interpolation performance. Second, instead of collecting a data point from
each sampling location, the CubeSat radiometer usually captures data from multiple
frequency channels, which turns the scattered data interpolation problem from single
output to multi-channel outputs. In this thesis, models and associated algorithms based
on the framework of Gaussian Process (GP) Regression are provided to solve both
problems. The Optimized Gaussian Process Regression (OGPR) approach provides both
estimation of the value on the grid and the geolocation by forming a Maximum A Posterior
(MAP) estimation which couples the statistical model for spatial variations in the
sensor data field provided by a GP model with a stochastic model for positional
uncertainties. For the multi-channel interpolation problem, we propose a Gaussian
Process Mixture Model (GPMM) that extends the naive GP model by constructing the
covariance matrix as a superposition of Kronecker products between spatial variations
for each output channel and weight matrices allowing for mixing across channels. Both
approaches are proven to provide superior performance compared with traditional
approaches in real data experiments. Although the algorithms that we discuss in this
thesis are motivated from the CubeSat application, they can be used in anywhere where
the model is applicable.
Thesis (Ph.D.)--Tufts University, 2018.
Submitted to the Dept. of Electrical Engineering.
Advisor: Eric Miller.
Committee: William Blackwell, Mai Vu, and Liping Liu.
Keyword: Electrical engineering.read less - ID:
- 5t34sx10g
- Component ID:
- tufts:28642
- To Cite:
- TARC Citation Guide EndNote
