Study On Algorithms For Radar HRRP Target Recognition Based On Convolutional Neural Network

Radar high-resolution range profile(HRRP)is the amplitude of the coherent summations of the projections of broadband radar target scattering point echoes in the direction of radar line-of-sight(LOS).In recent years,HRRP has become one of the research hotspots in the field of radar automatic target recognition(RATR)due to its advantages of easy access,small data volume,simple processing,and rich target structure signatures.This dissertation first introduced the historical background and research significance of the RATR technology based on HRRP,and briefly explained the main content of this dissertation.Afterwards,starting from the traditional convolutional neural network(CNN)structure,this dissertation gradually introduced three CNN-based HRRP target recognition algorithms.The specific content is as follows:1?This dissertation studies a high-resolution range profile recognition algorithm based on one-dimensional convolutional neural network(1DCNN).Starting from the structure of the CNN model,a one-dimensional CNN is used to extract features from the original HRRP time-domain signal,and the extracted features are used for target recognition.Afterwards,the advantages of 1DCNN over other traditional HRRP recognition algorithms in recognition performance and robustness are proved through experiments.Finally,we summarize the advantages and disadvantages of the model.2?This dissertation studies a high-resolution range profile recognition algorithm based on two-dimensional convolutional neural network(2DCNN).In view of the problem that the HRRP time-domain signal used by the 1DCNN model only uses HRRP intensity information and cannot adapt to the CNN structural characteristics,this dissertation uses two-dimensional CNN to extract the HRRP spectrogram signal and use the extracted features for target recognition.Then we verified the superiority of the 2DCNN model over the 1DCNN model in terms of recognition performance through experiments.Finally,we also summarize the advantages and disadvantages of the 2DCNN model.3?This dissertation studies a high-resolution range profile recognition algorithm based on three-dimensional convolutional neural network(3DCNN).This dissertation first introduces the network structure of three-dimensional CNN.Then,by using the segmented recombination of the HRRP spectrogram signal as input,the three-dimensional CNN is used to extract the feature of the HRRP spectrogram recombination signal,and use the extracted features for target recognition.The 3DCNN model solves the network structure problem caused by the integrality of the frequency dimension of the HRRP spectrogram signal,and further improves the recognition performance of the model.