Precession Period Extraction of Axisymmetric Space Target from RCS Sequence via Convolutional Neural Network

Abstract

The precession period was used to identify space targets in radar target recognition, especially the axisymmetric ballistic missile warheads and the similar shaped decoys. There are many precession period extraction methods based on RCS sequence. However, these methods have many restrictions and often yield poor results under noise condition. Aiming at extracting the precession period from RCS sequence, this paper designed a one-dimensional convolutional neural network. The precession period extraction is converted to a signal parameter estimation problem where the RCS sequence is the input signal and the period is the expected parameter. The proposed method was trained and validated on simulated RCS sequences and compared with spectral method, including CAUTOC, CAMDF, CAUTOC/CAMDF and trigonometric fitting method. The results showed that the proposed method yielded more accurate estimation results. Moreover, it can tell that there is no valid period by yielding a value that is not in the range [2, N /2]where $N$ is the length of the RCS sequence.