Research On Low Probability Of Intercept Radar Singnal Recongnition Algorithm Based On Stacked Sparse AutoEndocer

Due to the emergence of complex system radar such as Low Probability of Intercept(LPI)radars,the conventional radar signal identification method has failed.Therefore,how to effectively identify LPI radar signals has become the focus of non-cooperative radar signal processing research in recent years.Deep learning artificial intelligence technology has been developing by leaps and bounds since 2006,significant results have been achieved in the field of image recognition,The stacked Sparse Auto Encoder(sSAE)is one of the most commonly used models for depth learning.Compared with the traditional radar signal recognition algorithm,Automatic recognition algorithm based on sSAE can use the signal data itself,automatically learn and extract the characteristics of the signal time-frequency images,it reduces the huge workload of artificial extraction of characteristics,increases the efficiency,and significantly improves the recognition efficiency of LPI radar signal under low SNR condition.The innovative work and achievements of the dissertation are as follows:Firstly,the image preprocessing method based on CWD time-frequency image was studied,thus the difference of time frequency images between different signals is enhanced significantly.the redundancy between related information is removed,and the feature dimension is reduced.While reducing the signal noise,it retains the characteristics of the time-frequency image of the radar signal more completely,and simplifies the operation of the recognition part.Secondly,in view of the low recognition rate of LPI radar signals and the difficulty of feature extraction,this paper studies an automatic classification and recognition system based on CWD and sSAE.The system starts with the time-frequency images that reflect the essential characteristics of the signal.Firstly the LPI radar signal is transformed into CWD time-frequency,the one dimensional time signal is convert into a two dimensional time frequency image.Secondly,the original image of the time frequency is preprocessed and sent to the multi-layer sparse self encoder(SAE)for off-line training.Finally,the characteristics automatically extracted by SAE are input to the softmax classifier to realize the online classification and recognition of radar signals.The simulation results show that when the low SNR,it is obviously better than the method of manual design to extract signal features.Finally,based on the advantages and existing problems of random forest(RF)classifier,this paper studies the improved sSAE-RF algorithm based on sSAE and RF classifiers.The training of the RF classifier can be highly paralleled,thus a significant speed advantage is shown in large training data and high characteristic dimensions.But when the SNR is low,the RF model is poor in classification because of overfitting,and the training time of is long.However,sSAE can automatically learn and extract signal features with excellent classification and recognition at low SNR.The simulation results show that the improved sSAE-RF algorithm not only improves the training speed of the RF model when the training data is large,but also improves the poor classification effect of RF when the SNR is low.