Recognition of Ballistic Targets by Fusing Micro-Motion Features with Networks

Abstract

Ballistic target recognition is of great significance for space attack and defense. The micro-motion features, which contain spatial and motion information, can be regarded as the foundation of the recognition of ballistic targets. To take full advantage of the micro-motion information of ballistic targets, this paper proposes a method based on feature fusion to recognize ballistic targets. The proposed method takes two types of data as input: the time–range (TR) map and the time–frequency (TF) spectrum. An improved feature extraction module based on 1D convolution and time self-attention is applied first to extract the multi-level features at each time instant and the global temporal information. Then, to efficiently fuse the features extracted from the TR map and TF spectrum, deep generalized canonical correlation analysis with center loss (DGCCA-CL) is proposed to transform the extracted features into a hidden space. The proposed DGCCA-CL possesses better performance in two aspects: small intra-class distance and compact representation, which is crucial to the fusion of multi-modality data. At last, the attention mechanism-based classifier which can adaptively focus on the important features is employed to give the target types. Experiment results show that the proposed method outperforms other network-based recognition methods.

Scatterer-Level Time-Frequency-Frequency Rate Representation for Micro-Motion Identification

“…Compared to the precession and wobble, Equations ( 23) and (24) indicate that nutation causes a more complicated modulation law. It can be noted that the IF and IFR sequences of the scattering center p are weighted summations of high-order trigonometric functions with frequency ω c , ω s , 2ω s , ω c ± ω s , and ω c ± 2ω s .…”

“…Wang et al [22] presented a deep learning-based model combined with denoising CNN and MDS datasets for inertia parameter identification. To further improve the stability and robustness of micro-motion discrimination, some fusion recognition methods, which extract both narrowband and wideband features through DCNN, have been proposed, such as decision-level fusion [23] and feature-level fusion methods [24].…”