Parameter estimations of uncooperative space targets using novel mixed artificial neural network

Abstract Estimating the parameters of an uncooperative space target is vital important in the on-orbit servicing missions (OOS). Considering the measurement failures caused by the complexed space environment during the measuring procedure, this paper proposes a novel dual vector quaternions based mixed artificial neural network estimating algorithm (DVQ-MANN) to estimate the parameters of the uncooperative space target. Firstly, the dual vector quaternions (DVQ) are utilized to set up the relative kinematics and dynamics model. When the measurements are available, an Extended Kalman Filter (EKF) of the DVQ-MANN will operate to accomplish parameter estimations. When the measurements failures occur, the artificial neural networks(ANN) of the DVQ-MANN will work. In the designed DVQ-MANN, the first ANN is a three-layer feedforward neural network to estimate the states of the uncooperative space target. Besides, a novel deep convolution neural network is designed to estimate the covariance matrix of the states for its advantages in processing high dimensional inputs. By training well off-board and updating onboard, the proposed ANNs of the proposed DVQ-MANN can make reliable estimations under measurements failures. Finally, the proposed DVQ-MANN is validated by mathematical simulations to show its robust performances.