Study On Optimum Coherent And Joint Feature-based Detection Methods Of Sea-surface Weak Targets

With the development of stealth technology and the miniaturization of various targets,weak targets on the sea surface have become the key objects for maritime radar to perform early warning,search and alert.Weak targets refer to the all kinds of sea-surface or low-altitude targets whose radar cross sections are small and radar returns are often submerged in strong sea clutter.To detect such weak targets,high-resolution radars are often used to reduce power level of sea clutter and enhance the signal-to-noise ratio of target returns.High-resolution sea clutter exhibits complex characteristics including non-Gaussianity in the probability distribution,temporal nonstationarity,spatial heterogeneity,and spatial-temporally varying characteristics of model parameters,which undoubtedly makes detection more difficult.In terms of the low detection probability and numerous false alarms for traditional detection algorithms,the optimal coherent detection theory and joint feature-based detection algorithm for weak targets are researched in this dissertation.The former focuses on the statistical model of sea clutter and the design methods of optimum or asymptotically detectors matching the clutter characteristics,while the latter focuses on the multi-feature or multiscan joint detection methods in the long integration or observation time.The research results can be applied for the airborne early-warning radar in sea-to-sea mode,coastal surveillance radar,high-resolution anti-invasion radar for island or ship and high-resolution anti-submarine radar to improve their detection capabilities in sea clutter.The main research contributions in this dissertation can be summarized as follows.1.In terms of the performance loss that the optimum coherent detector in K-distributed clutter(OKD)causes in homogeneous clutter,an optimal coherent detector in homogeneous K-distributed clutter(Hom-OKD)and its corresponding adaptive detector are constructed.The homogeneity means that the clutter vectors of the cell under test(CUT)and reference cells share the same power,which often occurs in the case of low sea state and high-resolution clutter.The Hom-OKD is proposed by inserting the reference cells into the likelihood ratio test and it is proved that both Hom-OKD and adaptive Hom-OKD are constant false alarm rate with respect to the speckle covariance matrix,the Doppler steering vector,and the scale parameter of inverse Gamma texture.Moreover,the noniterative maximum likelihood covariance matrix estimation is developed.The experimental results by using simulation data and real data confirm the optimality of Hom-OKD in homogeneous clutter environments and its complementarity of the OKD in the real sea clutter environment.2.For the traditional adaptive detectors derived under the assumption of independent and identically distributed(IID)textures encounter performance loss,two coherent detectors under compound Gaussian clutter with inverse Gamma texture and inverse Gaussian texture are proposed based on texture structure.From the prior knowledge that texture is induced by large-scale wave modulation,the correlation coefficient is introduced to measure the texture correlation along range.According to this correlation coefficient,the number of range cells having the same texture as the CUT can be determined to provide texture information for CUT.The experimental results by utilizing open real data verify that the proposed detectors fully exploit the texture information conveyed by reference cells and attain better performance improvement for marine radar in comparison with the existing detectors in IID textures.3.Due to the low detection probability of low-velocity and floating small targets in sea clutter,an adaptive composite generalized likelihood ratio test detector(GLRT)using the forward-backward income-reference particle filter(FB-IRPF)for state estimation.Long time observation is a necessary way to improve the integrated gain for small and weak targets.However,it leads to the nonlinearity of target's Doppler offset and the nonstationarity of sea clutter,making small targets almost undetectable.Thus,the piecewise linear frequency modulation signal model and the piecewise spherical invariant random variable clutter model are developed and then the adaptive composite GLRT detector is proposed under the two models.In order to effectively estimate the unknown instantaneous Doppler frequency of target,the FB-IRPF algorithm is proposed in sea clutter,where the optimum coherent detector under inverse Gamma texture serves as the income function.In this way,adding the FB-IRPF state estimation into the adaptive composite GLRT detector forms the new proposed detector,realizing the short-time coherent integration and long-time noncoherent integration simultaneously.Experimental results using real data show that the proposed detector can be applied to detect low-velocity and floating small targets in sea clutter,and that it achieves obvious performance improvement when the Doppler spectrum of target and sea clutter are separate.4.Focusing on the problem that single feature lacks robustness detection of small target under complex clutter environments,the multi-feature joint detection method in timefrequency(TF)feature space is proposed.First,the TF distribution is normalized to mitigate sea clutter.On the normalized time-frequency distribution(NTFD)plane,the power of sea clutter is distributed randomly on the whole plane,while that of targets is concentrated near its instantaneous Doppler curve in a rule.To reflect the above different characteristics between sea clutter and targets,three quantitative TF features are extracted from NTFD.Then,the detection problem is transformed into the one classifier design problem in 3D TF feature space.Cooperating with the modified convexhull learning algorithm,a feature-based detector using three TF features is proposed and it enhance the separation of sea clutter and target in TF feature space.Finally,experiments based on real sea clutter data show that the proposed detector attains robust and obvious overall performance improvement compared with the fractal detector,the consistency factor detector and the tri-feature-based detector.5.To solve the problems that the long-time integration are unavailable in fast-scan mode and sea spikes result in false alarms in high resolution,a Doppler-guided detector via multiscan integration is proposed to detect sea-surface small targets.It consists of two-stage detections within one scan and among scans and Doppler information is the bridge to join two stages.Within one scan,the optimum coherent detection extracts Doppler information and highlights the trajectory of small target as much as possible at a high false alarm.Because sea spikes are decorrelated in the time scale of seconds,multiple scans are jointly processed to mitigate sea spikes effectively and reduce false alarm rate.It accumulates in a small region along the main direction of Doppler guidance,which improves the gain of long-time integration of small targets and reduces the computational cost.Besides,a simulation model for sea spikes is proposed and it includes the characteristics of appearance time,amplitude modulation and random distribution for sea spikes.Finally,experimental results by using simulation data and real data verify the better performance improvement of the proposed detector and it can detect small moving targets steadily and effectively in the high-resolution sea clutter.