Research On Adaptive Detection Algorithm Of Radar Targets In Sea Clutter

This thesis deals with the problem of radar targets detection in sea clutter.The sea clutter is always modelled by the Gaussian model for the low-resolution radars or high grazing angles.While,the sea clutter from high-resolution radars or at low grazing angles always exhibits obvious non-Gaussian characteristics.Under the circumstance,the traditional adaptive detectors for the Gaussian model will have performance loss in non-Gaussian clutter.In this thesis,the sea clutter is modelled by the compound-Gaussian model.There are some issues of radar targets detection in compound-Gaussian clutter.First,the target echo from high-resolution radars will scattered and spread more than one range cells.The point target model is no longer suitable for this case.The movement state of the target along the radar line of sight will be observed more finely.In addition,the target also has various movements such as rotation and pitch relative to the radar.It means that the rank-1 model also has a mismatch in practical scenarios.These two points may cause some performance loss.Second,the traditional adaptive detectors always need training data to estimate the covariance matrix.In the case of limited training data,the performance of traditional adaptive detectors will be bad.Third,the test statistic of optimal detector in compound-Gaussian clutter with generalized inverse Gaussian texture include the modified bessel functions of the second depending on data.Due to the high computational complexity of the test statistic,it is difficult to satisfy the real-time requirements of radar systems.In view of the above issues,the work of this thesis is as follows.1.Aiming at the problem of traditional target model mismatch in high-resolution radars,we model the range-spread target by using multi-dimensional linear subspace.The sea clutter is modeled by compound-Gaussian model,and the generalized Inverse Gaussian distribution is used to describe the texture component.Three adaptive detectors are proposed by using two-step generalized likelihood ratio test(GLRT),maximum a-posteriori(MAP)GLRT and Rao test design criteria.Aiming at the problem that the performance of these three detectors is limited by the training data sizes,three adaptive detectors based on the persymmetric structure of covariance matrix are proposed.The experimental result shows that the proposed detectors in this chapter outperform the traditional detectors in both simulated data and measured data,and the detectors based on the persymmetric structure of covariance matrix have better performance than their counterparts in the limited training data scenarios.2.Aiming at the problem that the performance of traditional detectors is limited by the training data sizes,two adaptive detectors for point-like targets based on bayesian criterion are proposed.The sea clutter is modelled by compound-Gaussian model with generalized inverse Gaussian texture.The covariance matrix speckle component is modelled as a random matrix that follows the complex inverse Wishart distribution.First,the detector without using the training data is deduced by GLRT design criterion,and the unknown parameters is instead by their MAP estimate(MAPE)or maximum likelihood estimate(MLE).Then another detector that combines a-priori knowledge and training data is proposed based on two-step MAP-GLRT design criterion.The training data and a-priori knowledge are used to estimate the covariance matrix.The experimental result shows that the detector without using the training data has better performance than the traditional detectors in the case of limited training data in both simulated data and measured data,and the detector that combines a-priori knowledge and training data has the best performance in different training data sizes.3.Aiming at the problem that the computational complexity the optimum coherent detector embedded in compound-Gaussian clutter with generalized inverse Gaussian texture(CG-GIG clutter)is too high to satisfy the real-time requirements of radar systems because of the modified bessel functions of the second depending on data.A near-optimum coherent detector with a new structure and its adaptive version are proposed,which decreases the computational cost.The near-optimum coherent detector contains two common detectors,the generalized likelihood ratio test with linear-threshold detector(GLRT-LTD)and the?-MF in K-distributed clutter.It also has a comparable detection performance of the optimum detector in CG-IG clutter.Theoretical analysis and numerical experiments illustrate that the proposed near-optimum detector has the constant false alarm ratio(CFAR)property relative to the speckle covariance matrix,Doppler steering vector and the mean power of clutter.The experimental result shows that the proposed detector has the optimum or near-optimum performance and outperforms the match filter detector and normalized match filter detector.The CG-GIG distribution contains three common distribution,the generalized Pareto distribution,the K-distribution and the CG-IG distribution.Therefore,the use of the detector proposed in this chapter can minimize the performance loss caused by the mismatch of the clutter model.