| Automatic target recognition of multi-dimensional high resolution radar image can use image data to acquire information such as target category and attribute.It is an important part of radar image processing and interpretation.It is one of the key technologies to be solved urgently in radar technology application.It has high research value and practical significance.An important indicator of the increasing capability of radar imaging system is the gradual improvement of spatial resolution.High resolution radar images can provide more information,and also bring a series of new problems to target recognition.On the one hand,high-resolution radar,especially terahertz radar,as a frontier technology,is scarcely used in the civil field.On the other hand,the scattering characteristics of highresolution radar target are complex and the number of samples is small at this stage,which makes it difficult for traditional algorithms to meet the demand of target recognition of high-resolution radar.To counter the problems above,the theoretical analysis,method research and results of simulated data and measured data are carried out in this paper.The main contributions are as follows:1.A terahertz radar gesture recognition method is proposed.By extracting position features of one-dimensional range profile sequence and combining with Doppler information acquired by time-frequency analysis,the extended dynamic time-warping algorithm is studied.Aiming at the problem of multi-feature target classification,a decision-level fusion strategy is proposed,which realizes the recognition of terahertz radar gesture.A large number of terahertz radar gesture data were collected,and the performance of the algorithm was verified and analyzed by the measured data.2.A radar target recognition method based on multi-level reconstruction of scattering centers is proposed.The attribute scattering center model and energy ratio model are used to reconstruct multi-level high resolution radar images,which solves the problem of negative impact of radar image noise on recognition and low generalization ability of target recognition algorithm caused by a small number of training samples.The proposed method effectively improves the recognition performance of radar targets under the condition of limited original radar images.3.A method of radar target recognition based on Gabor feature decision fusion strategy is proposed.The contribution of Gabor feature in different scales and directions to recognition is measured by weight.The loss function is defined to obtain the optimal solution of weight.A new decision fusion mechanism is established.The problem of incomplete balance of multi-feature decision fusion in traditional methods is solved and the recognition rate of radar target is improved.4.A method of radar target recognition based on feature level fusion and decision fusion of the monogenic signal is proposed.Scale confidence model is established by Fisher criterion to realize feature level fusion based on scale selection.A weight-based decision fusion strategy is constructed based on scale confidence model to improve the recognition rate and generalization performance of radar target recognition algorithm.In this paper,a series of research on target recognition in multi-dimensional highresolution radar images are carried out.Aiming at the application of ultra-high-resolution radar images and the lack of data,a serious of explorations are made to provide guidance for target recognition in high-resolution radar images. |
