Daniel Pimentel-Alarcon, Georgia State University; Ashish Tiwari, Georgia State University; Douglas Hope, Hope Scientific Renaissance LLC; Stuart Jefferies, Georgia State University
Keywords:
Abstract:
High-resolution imaging of satellites through the Earths turbulent atmosphere plays a pivotal role in space domain awareness. When imaging through strong turbulence it is not uncommon to have to screen thousands of images to select a few promising ones that can be used in the image restoration process. Moreover, the quality of the restoration depends on the selected images.
We propose a novel image selection technique based on advanced machine learning methods. In particular, we use wavelet coefficients, known to be sparse and well behaved in images with similar characteristics to satellite images. Then we use group logistic lasso to identify the most relevant coefficients and the most promising images. We compare image reconstructions based on our selection method with a current state-of-the-art technique, which is centered on Fourier analysis.
Date of Conference: September 11-14, 2018
Track: Adaptive Optics & Imaging