Towards Automation of Tipping and Cueing Between Small Satellites in a Constellation
Abstract
The use of using low-fidelity sensors of satellites in a constellation for accurate surface target detection has the potential to lower costs while increasing flexibility, replacement time, and fault tolerance. This thesis investigates the possibility of utilizing an array of satellites with a heterogeneous mix of sensor types to optimize the validation process of surface target detection. Generation of synthetic scenes allows identification and extraction of optical features that are useful in remote sensing practices. Features of interest are generated from specified resolutions, representing different sensor types using Rochester Institute of Technology's Digital Imaging Remote Sensing Image Generation platform. Synthetic images are jittered to varying degrees to represent the pointing stability. These synthetic images are utilized to train The Berkeley Caffe Convolution-Based Deep Learning open source platform to automatically detect features of interest. Through these experiments, we demonstrate that remote sensing platforms can provide features of interest to an artificial intelligence platform to increase overall feature identification effectiveness. Through these experiments, we demonstrate that remote sensing platforms can provide features of interest to an artificial intelligence platform to increase overall feature identification effectiveness.