Convolutional Neural Networks for Spacecraft Pose Estimation(NPO-51334-1)

The software trains and tests convolutional neural networks to determine spacecraft pose estimation. It also contains functionality to perform image preprocessing, data augmentation and tensorflow model compression.The software takes a unique approach that incorporates more image data than other approaches while retaining the original spatial dimensions. Other approaches will down sample images significantly which makes the neural network lose a lot of spatial information retarding robustness.The software is designed to estimate the pose of a spacecraft based on an image of that spacecraft. This includes the relative position and attitude of the target satellite with respect to a chasing satellite. Application of this technology aids function such as active debris removal and formation flying among other applications.