基于多变换域特征信息的雷达波形匹配方法OA

In view of the degradation in waveform matching accuracy caused by noise interference and the limited representational capacity of single-transform-domain features,this paper proposes a multi-transform-domain feature fusion solution.Firstly,it constructs a three-channel joint representation by simultaneously extracting noise-resistant bispectral features,Hilbert marginal spectra suitable for analyzing non-stationary waveforms,and distinctive ambiguity function representative slices,thereby overcoming the incompleteness of single features.Secondly,a lightweight network with dual-attention mechanisms is designed to achieve adaptive feature weighting and noise suppression.Finally,a sliding window mechanism is employed to overcome phase alignment challenges.This approach establishes a technical chain of"multi-domain feature fusion-attention-based noise suppression-matching and localization".Experiments show that compared with the traditional ResNet method,the accuracy rate of the proposed method is improved by 4.71%in low SNR environment,and the proposed method can realize accurate matching in high SNR environment.Regarding matching speed,it reduces the time required for 100 training epochs by 46.5%compared to CNN.The proposed method has a wide engineering application prospect in the field of radar waveform detection and reconnaissance.