Research On Radar Target Recognition Technology Based On One-dimensional Range Profile

Radar has been an important detection device in the field of military and civil on the account of its operating characteristic of all-day and all-night,strong penetation and so forth.High resolution range profile(HRRP)is the vector sum of the target scattering echo projected in the direction of radar line of sight,which contains the structural information such as the geometric size of the detected target.Radar automatic target recognition(RATR)is a vital technique for radar signal interpretation.With the rapid development of deep learning technology,high resolution range profile target recognition method based on deep learning has perform better than traditional one.Aiming to many shortages of existing HRRP RATR methods,this thesis focuses on target features deep extraction,adaptvie feature refinement,temporal feature modeling and so on,and proposes three target recognition models.The main contents of this thesis are follows:(1)Focusing on the global feature instability of radar HRRP data,a one-deimensional convolutional neural network with attention mechanism is proposed.The proposed method takes convolutional neural network as recognition framework,extracts local features by multiple convolutional operations,aiming to promote the feature staility of radar target.Meanwhile,a bi-parallel featue refinement module is developed to improve the discriminatation of local feature,which can highlight useful information and suppress unuseful one in both feature channel and spatial structure,thereby improving the recognition performance of the proposed method.(2)With regard to the sensitive problem of radar attitude and translation variation,a double branch variable scale temporal convolutional network is proposed for HRRP target recognition.The proposed method leverages convolutional operation in the low layer of network to effectively alleviate the translation sensitivity of HPPP target.In order to alleviate the attitude sensitivity of HRRP target,the proposed method considers the temporal relations between different range unit of HRRP data,and establishs feature extraction layer based on temporal convolution operation.Moreover,variable scale convolution operation is adopted to extract target deep semantic features,protoming target recognition performance by alleviating the attitude sensitivity of HRRP data.(3)Aiming to the problem of weak generalization ability of existing radar HRRP target recognition model,Conv-Transformer model-based HRRP target recognition method is studied.Considering that transformer model can extract the gobal temporal relations of target feature by a self-attention mechanism,a gobal temporal feature extraction model is established for HRRP target recognition.In order to make up for the fact that transformer can only model the whole sequence due to its self-attention mechanism,it is unable to encode the correlation of local features between range images,convolution operation layer is designed to encode the realtion of local features,thereby improving the generalization ability of HRRP target recognition model.