Using Synthetic Satellite Imagery from Virtual Worlds to Train Deep Learning Models for Object Recognition