Reseach On Adaptive Target Detection With Small Amount Of Training Data

Adaptive target detection is an important research direction in the field of radar signal processing,and it is the prerequisite for many functions of radar.For target detection with unknown covariance matrix,training data are usually needed to estimate the noise covariance matrix(NCM).Usually,the adaptive processing loss caused by the amount of training data is required to be at least double the rank of the NCM to maintain a 3d B loss with respect to NCM.Because of the complex environment and terrain,sufficient training data could not be obtained in practice.When the amount of training data(ATD)is small,the detection performance of traditional detectors will drop significantly.This paper studies the problem of target detection with limited ATD.The research content and innovations of this paper are as follows:(1)For subspace target detection with limited training data,this paper proposes a new detector by applying the persymmetric structure and reduced-dimension approach.Firstly,a particular linear transformation is applied to the persymmetry of the NCM.Then,the reduced-dimension approach is used by projecting the data onto the signal subspace.Last,the detector is designed based on generalized likelihood ratio test(GLRT).The detector is shown to possess the constant false alarm rate(CFAR)property with respect to NCM.Simulation results show that the proposed detector has higher probability of detection and less demand for ATD compared with the counterparts of other detectors.(2)For distributed targets detection with partially known steering vectors,a reduced-dimension approach with persymmetric structure is proposed.Firstly,the persymmetric structure is exploited by a particular linear transformation.Then,the data are projected onto the signal subspace to reduce dimension.Last,the detectors are designed based on GLRT,two step GLRT and Wald in homogeneous environment and partially homogeneous environment.These new detectors have CFAR properties with respect to noise.Numerical results demonstrate that the new detectors have lower demands for training data size and achieve higher probabilities of detection compared with the counterparts of other detectors.(3)For target detection without a priori information in a sample-starved environment,a new reduced-dimenison approach is proposed.The test data and training data vectors can be properly divided by rows into several groups.Providing that the steering matrices corresponding to all groups have full-column-ranks,the data in all groups can be individually used for target detection.And the final detection result is the synthesis of the results with respect to all groups.The simulation results show that the new detection strategy is effective in achieving larger probability of detection,especially when the ATD is small.