Research On Radar Jamming Identification Method Based On Deep Learning

As technology continues to develop and integrate,the styles of radar jamming become increasingly diverse.Rapid and effective identification of jamming signals is crucial in electronic warfare,and it is a prerequisite for subsequent jamming suppression.Machine learning-based jamming recognition methods typically involve artificially selecting various feature parameters and designing different classifiers for classification and recognition.However,such traditional methods rely on the subjective judgment of the experimenter,have weak generalization ability,and are difficult to adapt to the current rapidly changing and complex adversarial environment.Therefore,there is an urgent need to explore intelligent antijamming methods that can adaptively extract features.With the rapid development of deep learning in recent years,radar jamming recognition has also entered a new stage,and the latest research on jamming recognition is also based on deep learning.However,most of the jamming recognition methods used in the research have long training times and are difficult to accurately distinguish unknown jamming.To address these issues,this article proposes a structure based on ECA and dynamic convolution combined with a lightweight network to reduce training time,aiming to achieve effective recognition of known jamming with fewer samples in a shorter time.For unknown jamming,a support vector data description recognition method based on differential evolution algorithm optimization is further proposed.Finally,a recognition method based on siamese networks is proposed to achieve differentiation and resolution of several types of jamming under unknown classes.The primary focus of the research comprises the following.First of all,building mathematical models for noise convolution jamming signals,noise product jamming signals,slice-type relay jamming signals,intermittent sampling direct relay jamming signals,intermittent sampling repeated relay jamming signals,intermittent sampling cyclic relay jamming signals,and frequency dispersion jamming signals.The jamming mechanisms of these signals on radar were analyzed,and their time-domain,frequency-domain,time-frequency joint information,and pulse compression information were analyzed to provide theoretical support for subsequent recognition.Next,this paper generates image datasets through time-frequency analysis as input for a lightweight network,achieving effective classification and recognition of radar jamming.To prevent the problem of high-dimensional feature loss caused by compressed computational complexity in lightweight networks,dynamic convolution kernels are introduced to enhance model expressiveness.By leveraging the interaction between attention concentration on kernels and channels to enhance expressiveness,ECANet is selected as the channel attention module,and a radar jamming recognition method based on ECA and dynamic convolution is proposed.Through simulation verification,the feasibility of this method is demonstrated,achieving higher accuracy and stronger robustness.Finally,regarding the unknown class of radar jamming,this paper proposes a support vector data description optimization based on differential evolution algorithm.After feature extraction using the above-mentioned Mobile Net V3 network,a feature description sphere of known jamming is obtained,and the extracted features of unknown jamming in the test input are compared with the known sphere to achieve effective recognition of known and unknown class radar jamming.In terms of distinguishing the styles of radar jamming in the unknown class,this paper designs a comparison recognition method based on the Siamese network,which sends the extracted features to the comparison module for similarity comparison,achieving effective recognition of the styles of unknown class radar jamming.