On-Orbit Real-Time Avionics Package Identification Using Vision-Based Machine Learning Techniques

Abstract

Capability for on-orbit inspection using autonomous robots is important in manned space station/capsule operations. Due to the unpredictable lighting conditions on orbit, machine learning-based package identification and detection is a good candidate for inspection operations. In this work, the AI-based technique has been used to identify packages of interest from a set of packages in the orbital platforms. Images of different packages on the platform are used for the training using a simple neural network for fast inference of packages in the frame.