Session
Technical Poster Session 1: Student Poster Competition
Location
Utah State University, Logan, UT
Abstract
With the increasing risk of collisions with space debris and the growing interest in on-orbit servicing, the ability to autonomously capture non-cooperative, tumbling target objects remains an unresolved challenge. To accomplish this task, characterizing and classifying satellite components is critical to the success of the mission. This paper focuses on using machine vision by a small satellite to perform image classification based on locating and identifying satellite components such as satellite bodies, solar panels or antennas. The classification and component detection approach is based on “You Only Look Once” (YOLO) V5, which uses Neural Networks to identify the satellite components. The training dataset includes images of real and virtual satellites and additional preprocessed images to increase the effectiveness of the algorithm. The weights obtained from the algorithm are then used in a spacecraft motion dynamics and orbital lighting simulator to test classification and detection performance. Each test case entails a different approach path of the chaser satellite to the target satellite, a different attitude motion of the target satellite, and different lighting conditions to mimic that of the Sun. Initial results indicate that once trained, the YOLO V5 approach is able to effectively process an input camera feed to solve satellite classification and component detection problems in real-time within the limitations of flight computers.
Article
Real-Time Satellite Component Recognition with YOLO-V5
Utah State University, Logan, UT
With the increasing risk of collisions with space debris and the growing interest in on-orbit servicing, the ability to autonomously capture non-cooperative, tumbling target objects remains an unresolved challenge. To accomplish this task, characterizing and classifying satellite components is critical to the success of the mission. This paper focuses on using machine vision by a small satellite to perform image classification based on locating and identifying satellite components such as satellite bodies, solar panels or antennas. The classification and component detection approach is based on “You Only Look Once” (YOLO) V5, which uses Neural Networks to identify the satellite components. The training dataset includes images of real and virtual satellites and additional preprocessed images to increase the effectiveness of the algorithm. The weights obtained from the algorithm are then used in a spacecraft motion dynamics and orbital lighting simulator to test classification and detection performance. Each test case entails a different approach path of the chaser satellite to the target satellite, a different attitude motion of the target satellite, and different lighting conditions to mimic that of the Sun. Initial results indicate that once trained, the YOLO V5 approach is able to effectively process an input camera feed to solve satellite classification and component detection problems in real-time within the limitations of flight computers.
