Radar Radiator Signal Recognition Based On Deep Learning And Fuzzy Function

Electronic countermeasure technology is constantly developing and innovating towards intelligent and adaptive countermeasures.Its main task is to intercept enemy radiation source signals for analysis and processing to achieve the purpose of identification and positioning.But with the emergence of a large number of modern new system radars and the continuous improvement of anti-reconnaissance technology,in the current environment,the classification and identification methods based on traditional external characteristic parameters can no longer meet the deteriorating electromagnetic environment.Most of the existing intra-pulse feature extraction methods rely on artificial extraction,which is difficult to achieve fast and accurate recognition requirements.Deep learning technology has been widely used in other pattern recognition fields.It can form high-level abstract feature expressions by learning the underlying features of the data to achieve classification.Therefore,this thesis proposes to use the deep learning method to mine the abstract features hidden in the signal ambiguity function,so as to improve the recognition rate of the radiation source signal.The main research work is as follows.(1)This thesis proposes an algorithm that uses the stacked-autoencoder model to learn the main ridges of the ambiguity function of the signal.First,the main ridges of the ambiguity function are obtained from the signal,and after preprocessing,it is used as the input vector of the SAE.Finally,the classification results are output.Experimental results show that the method can maintain 100% recognition accuracy when the signal-to-noise ratio SNR is more than 2 d B,and the recognition accuracy can still maintain more than 82.8% when the SNR is-6d B,which is significantly higher than other methods.Even in a signal environment containing different parameters of the same modulation type,the recognition rate is stable above 95.0% when the SNR is greater than 0 d B,and the recognition rate can reach 79.0% when the SNR is reduced to-4d B.It is proved that SAE can effectively extract the deep features of signal AFMR,and has good anti-noise performance.(2)Using CNN to extract features of AFCL,first use a Gaussian filter to reduce the noise of the signal AF,and project AF onto the time-frequency joint plane to its contour map,and finally binarize it before reducing the dimension Entered into the CNN.The high-order features of CNN are obtained through multiple convolution and pooling operations,and the classification results are output by the fully connected layer.Experiments show that when the SNR is> 0d B,the recognition rate remains above 92.0%,and even in a low SNR environment of-6d B,the recognition rate can reach 83.7%,which is much higher than the artificially extracted features.At the same time,compared with other forms of input,the performance of AFCL is also better than these methods,which proves that AFCL can more fully retain the inherent modulation characteristics of the signal and is more robust in a low signal-to-noise ratio environment.