Research On Low Slow And Small Targets Deection With Copound Gassian Clutter

The research on the detection of low slow and small targets is of great military significance and application value in the aspects of security of major events,air supervision,and defense of low slow combat weapons.Low slow small target with low altitude flight,slow flight speed,radar cross-sectional area of the characteristics of small,so acoustic,optical detection is difficult,usually using airborne radar for its detection.Aiming at the problem that the training sample is low,the Doppler expansion is serious and the clutter is nonhomogeneous,many methods such as the subspace detection algorithm,the polarization domain detection algorithm and all kinds of prior knowledge are used to improve the detection performance.The main results of this paper are as follows:1.The geometric model and the detection model of detecting the low slow and small target are studied,and the difficulty of detecting the low-slow target is analyzed.The clutter model is established as a compound Gaussian model.The principle of adaptive detection and related detection are introduced.The method of estimating the commonly used clutter covariance matrix is studied,which provides a theoretical basis for the subsequent research work:2.Aiming at the problem that the multi-dimensional subspace signal is detected in the compound Gaussian clutter,the effective training sample is insufficient and the performance of the traditional algorithm is seriously degraded,a target space subspace detector based on knowledge is proposed.The detector uses the compound Gaussian model with texture component as inverse gamma distribution to describe the clutter.Using the persymmetric structure of the clutter covariance matrix,based on the two-step generalized likelihood ratio detection criterion,we first use the a priori information to obtain the covariance matrix.Then the exact maximum likelihood estimate is substituted into the likelihood of obtaining the knowledge assisted by the target signal subspace detector.Computer simulation shows that the detector has better detection performance compared with the traditional detector,and simulates the influence of the covariance matrix estimation method and the prior information mismatch on the detection performance.3.Aiming at the problem of detecting low-slow small target,low signal-to-noise ratio and muti-polarimetric channels interaction in compound Gaussian clutter,a polarimetric-adaptive detector based on knowledge is proposed.Firstly,the clutter is modeled as a compound Gaussian model with texture component as inverse gamma distribution.Based on the Rao detection criterion,the test statistic is obtained and the exact maximum likelihood estimation of the texture component is deduced when the polarization channel interacts with each other,Then use the a priori information to obtain the clutter polarimetric scattering matrix,and then substitute the test statistic to obtain the knowledge assisted polarimetric adaptive detector.Computer simulation shows that the detector has better detection performance compared wit h the traditional detector,and simulates the influence of the texture component estimation method and the prior information mismatch on the detection performance.4.Aiming at the problem that the low-slow small target is detected in the partially homogeneous clutter,the signal feature is difficult to extract and the effective training sample is limited.The low-slow small target is modeled as a multi-dimensional subspace model.A subspace detector based on multiple a-priori spectral models is proposed.The detector uses the linear combination of the multiple a-priori spectrum models to represent the inverse of the clutter covariance matrix and can detect low slow small targets in the partially homogeneous clutter.The simulation results show that the detector is better than the traditional detector based on the asymptotic maximum likelihood estimation covariance matrix and the performance of the detectors based on multiple prior spectrum and subspace alone,even the number of training samples is insufficient.