Research On Recognition Technology Of Radar Active Jamming Signal Based On Deep Learning

Radar uses the reflection of the target to electromagnetic waves to find the target and determine its position,which plays a key role in the war.How to protect radar from the enemy’s interference and attack in the complex battlefield environment and ensure the normal operation of radar has gradually attracted great attention.In general,radar jamming signal recognition is the premise and foundation of radar anti-jamming.Only by accurately identifying the jamming patterns of jamming signals,especially those of more threatening active jamming signals,can we take targeted measures to suppress jamming and improve the survival rate of radar in the modern battlefield environment.Therefore,it is of great value to study the identification of radar active jamming signals.Aiming at the problems of complex model construction and low real-time performance in the traditional active jamming signal identification methods,this thesis deeply studies the radar active jamming signal identification technology based on deep learning.Aiming at the problem of long training time for deep learning networks,this thesis proposes a jamming signal identification method based on an improved CNN model.In order to reduce the time cost of building the model and shorten the training time,this thesis proposes a jamming signal recognition algorithm based on the transfer fine-tuning learning model.The main research contents of the thesis are as follows.Several common active jamming signal models are discussed,and their time-domain and frequency-domain characteristics are simulated and analyzed.The characteristic representation ability of five typical signal time-frequency analysis methods to eight kinds of jamming signals is studied.The simulation results show that the CWD transform is superior to the other four types of time-frequency analysis methods in three aspects,timefrequency resolution,cross-term suppression capability and jamming signal representation capability.The principle of CNN is analyzed and several common Deep Convolutional Neural Network(DCNN)models are discussed.In order to improve the recognition performance of CNN to jamming signals,a jamming signal recognition method based on the improved CNN model is proposed.The simulation results show that the model has a faster convergence speed while ensuring the recognition performance.In order to improve the time performance of deep learning,this thesis combines DCNN and transfer learning in radar active jamming signal identification technology,and gives an active jamming signal identification method based on deep transfer learning.To further reduce the training time cost,an active jamming signal recognition algorithm based on transfer finetuning learning is proposed.The deep transfer learning of five networks of AlexNet,ResNet18,VGG16,DenseNet121 and GoogleNet is implemented under the PyTorch framework,and the deep learning,deep transfer learning and transfer fine-tuning learning of five networks of VGG19,VGG16,ResNet50,InceptionResNetV2 and InceptionV3 are implemented under the Keras framework.An active jamming signal identification algorithm based on the VGG16 transfer fine-tuning learning model is proposed,and simulation experiments are carried out on the identification of jamming signals under different JNR.The results show that compared with the existing AlexNet transfer deep learning model and the CNN model,the VGG16 transfer fine-tuning learning model has better recognition effect and faster convergence speed,thereby avoiding the complicated process of building and training the network from scratch in the traditional deep learning methods.