Non-cooperative Spacecraft Detection Based on Deep Learning Under Complex Space Environment

In recent years, approaching non-cooperative spacecraft is necessary to multitudes of on-orbit missions, which are more difficult for non-cooperative spacecraft than cooperative spacecraft because there are no markings on the surface of non-cooperative spacecraft and service spacecraft can not access to pose information. To solve this problem, several navigation methods based on vision images have been proposed. However, most of those methods are validated in ideal environment, they do not consider the interference caused by complex environment in real space. In this work, a method using Ringed Residual U-Net (RRU-Net) to detect the non-cooperative target under complex space environment is proposed to improve the performance and robustness of on-orbit service missions for non-cooperative spacecraft. In addition, we compare the results of Binary Cross Entropy Loss (BCELoss) function and Tversky Loss function to improve the accuracy and robustness of this method. Experiment results demonstrate the proposed method achieve high accuracy and stability under complex space environment.