Aerial Target Recognition With Enhanced Micro-Doppler Dynamic Features Based on Frequency-Modulated Continuous-Wave Radar

The micro-Doppler (m-D) signal generated by the micro-motion of aerial target can be extracted for recognition. However, the m-D signal is weak and prone to be interfered by the bulk Doppler (b-D) signal and environmental noise. Moreover, existing m-D handcrafted features are limited for aerial target classification. In this article, a novel aerial target recognition method based on enhanced m-D dynamic features is proposed. First, a singular spectrum analysis method based on turning point (TP-SSA) is applied to divide the signal into two subspaces with interference and noninterference. The b-D interference signal is separated by reconstructing the noninterference subspace. In addition, an empirical mode decomposition method based on chopping pulse matching (CPM-EMD) is adopted to select the components with high matching degree through a scoring function to reconstruct the m-D signal. Second, the m-D signal is decomposed by sliding windows. Eight handcrafted features are extracted for each short period to form a feature sequence set containing statistical and time-varying information. Finally, a bidirectional gated recurrent unit (BiGRU) network is used to further extract the dynamic features for aerial target classification. The experimental results show that the proposed robust method achieves higher recognition performance with lower parameters.

Automated summary

In this article, a novel aerial target recognition method based on enhanced micro-Doppler dynamic features is proposed. The method separates the bulk Doppler interference signal using singular spectrum analysis and empirical mode decomposition, and extracts features through sliding window decomposition. Finally, a bidirectional gated recurrent unit network is used for aerial target classification based on the extracted dynamic features.