Radar False Target Recognition Method Based on Reconstruction Error Distribution Characteristics of Frequency Response From Multi-CPIs

The existing artificial intelligence (AI)-based radar active false target jamming recognition method is mostly based on supervised learning and performs poorly when the signal-to-noise ratio (SNR) is low. For these reasons, this article proposes an unsupervised recognition method based on the reconstruction error distribution characteristics of fast-slow time-frequency response (FST-FR) from multiple coherent processing intervals (CPIs). Due to the different scattering characteristics, the FST-FRs of real and false targets show different spatial and temporal characteristics. The FST-FRs of real targets show dynamic characteristics, while the FST-FRs of false targets are generally static. In the proposed method, a long short-term memory auto-encoder (LSTM-AE) is constructed to train and reconstruct the FRs of radar-transmitted signals and determine the reference threshold according to the reconstruction errors. Then, the FST-FRs of the same “target” extracted from multi-CPIs are input into the LSTM-AE to estimate the reconstruction error differential. Finally, by comparing the reconstruction error differential of the “target” with the reference threshold, the unsupervised recognition of real and false targets is realized. Multiple sets of measured data prove that the proposed method can effectively recognize real and false radar targets in the case of low SNR, and the recognition rate is higher than 93%.