Numerical Focusing of a Wide-Field-Angle Earth Radiation Budget Imager Using an Artificial Neural Network

Abstract

Narrow field-of-view scanning thermistor bolometer radiometers have traditionally been used to monitor the earth’s radiant energy budget from low earth orbit (LEO). Such instruments use a combination of cross-path scanning and along-path spacecraft motion to obtain a patchwork of punctual observations which are ultimately assembled into a mosaic. Monitoring has also been achieved using non-scanning instruments operating in a push-broom mode in LOE and imagers operating in geostationary orbit. The current contribution considers a fourth possibility, that of an imager operating in LEO. The system under consideration consists of a Ritchey-Chrétien telescope illuminating a plane two-dimensional microbolometer array. At large field angles, the focal length of the candidate instrument is field-angle dependent, resulting in a blurred image in the readout plane. Presented is a full-field focusing algorithm based on an artificial neural network (ANN). Absorbed power distributions on the microbolometer array produced by discretized scenes are obtained using a high-fidelity Monte Carlo ray-trace (MCRT) model of the imager. The resulting readout array/scene pairs are then used to train an ANN. We demonstrate that a properly trained ANN can be used to convert the readout power distribution into an accurate image of the corresponding discretized scene. This opens the possibility of using an ANN based on a high-fidelity imager model for numerical focusing of an actual imager.

Artificial Neural Networks in Radiation Heat Transfer Analysis

“…A widely lamented disadvantage of the MCRT method is the excessive computational cost associated with achieving high accuracy when fine spatial resolution is required. The fact that rays are mutually independent entities permits massive parallelization, with a proportionate reduction in processor time; however, associated cost, power, volume, and weight penalties exclude massive parallelization in applications where real-time results are required for data interpretation and decision-making on board autonomous space probes [6] and fire-and-forget weapons [7]. The alternative to a slow or computationally ponderous highfidelity model in such applications would be a reduced-order model that provides comparable accuracy and spatial resolution but in real-time and with significantly reduced hardware requirements [8].…”