Tatget Detection Methods In High Range Resolution Sea Clutter

The main research in the dissertation is several target detection methods in high rangeresolution sea clutter. Detection in sea clutter is an important issue, and it is one of themost complex problems in radar signal processing. Target detection in sea clutter canpotentially be utilized in military and civil field. Due to the spatial-temporal non-stationarity and non-Gaussian property of high range resolution sea clutter, targetdetection methods become complicated for the aircrafts over the sea and the floatingtargets on the sea surface.Based on the real IPIX radar sea clutter, the paper focuses on the target detections forthe aircrafts over the sea and the floating targets on the sea surface. Adaptivenormalized matched filter (ANMF) detector is widely used in sea clutter for targetdetection. We analyze the intrinsic conflict in the ANMF detector, and propose someschemes to alleviate the conflict. The main work includes the utilization of range-oversampled samples, clutter suppression algorithm, B-ANMF detector, and subbandANMF detector. All the algorithms mentioned above are based on the traditionaladditive model, and are applied to detecting aircrafts over the sea. Finally, atri-features-based detection algorithm is proposed for better detecting the floatingtargets on the sea surface.The main content of this dissertation is summarized as follows.The first part introduces the model of sea clutter, defines the homogeneous sea clutter,makes a detailed analysis on the development of coherent detectors, and points out thatthe intrinsic conflict in the ANMF detector is between the long integration duration andthe finite secondary samples. Finally, we introduce the real sea clutter data collected byIPIX radar, which forms the basis for the following study.The second part focuses on the performance gain of ANMF detector in range-oversampling case. The performance of ANMF detector depends on the estimation ofclutter covariance matrix. As an average estimator, the sample covariance matrix (SCM)estimator has been thoroughly analyzed for independent and identically distributedGaussian clutter vectors, but not for the range-oversampled secondary samples. Manyradar systems on service use range-oversampled receivers. Thus, the secondary samplesare correlated in spatially. Referred to the range-oversampled model in the weather andoceanic radars, the spatial correlation model of range-oversampled clutter vectors isestablished, the error of the SCM estimator is analyzed, and the range-oversampling gain, relevant to the oversampling factor and the receiver’s bandwidth, is derived. Therange-oversampling gain is used to evaluate the gain of performance of ANMF detector.The experiments using real radar clutter data are made to verify the range-oversamplinggain, showing that the range-oversampling improves the performance of ANMFdetector.The third part is contributed to clutter suppression algorithm and target detection inhigh range resolution sea clutter. The thesis basically consists of two parts: cluttersuppression and target detection. In the part of clutter suppression, it commences with aproposal of median-based estimator to estimate the power spectrum of high resolutionsea clutter by the time series observed in adjacent range cells and time intervals. Theestimator provides a robust estimation when just a few aberrant time series happen inobservation. Based on the estimator, a block-adaptive clutter suppression filter (BACSF)is designed to suppress the clutter prior to the pulse integration. In the part of targetdetection, due to that the residual clutter, the output of the BACSF, is modeled asspherically invariant random vector (SIRV), upon applying an ANMF detector to theresidual clutter, a residual clutter’s ANMF detector is derived. Moreover, in highresolution radar background, considering that the approximately stationary intervals ofsea clutter and residual clutter are much shorter than the coherent processing interval(CPI), another heuristic B-ANMF detector is proposed. It can integrate more pulses andachieve better performance than the ANMF detector does. The chapter concludes withexperiments of simulated target against the real sea clutter. Experimental resultsdemonstrate that, when target’s Doppler frequency is beyond strong clutter region, theANMF detector and B-ANMF detector perform better in residual clutter than in clutter.Hence, a double-channel scheme is given.The fourth part focuses on the subband ANMF detector in high range resolution seaclutter which is spatial-temporally non-stationary. The ANMF detector is limited inperformance, because the non-stationarity in spatially of the clutter restricts the spatialsamples available and the non-stationarity in temporally restricts the integrationduration. In order to overcome this limitation, a subband ANMF detector is proposed. Itconsists of a forward DFT modulated filter bank followed by a set of ANMF detectorsin individual subbands. The forward filter bank is used to decompose received signals ofhigh rate into subband signals of low rate. The subband decomposition via the filterbank realizes clutter suppression outside each subband and improves the short-termstationarity of sea clutter. Improved short-term stationarity and the low rate of subbandsignals allow the subband ANMF detector to have much longer integration durations than the traditional ANMF detector to do. The experimental results using real sea clutterdata show that the subband ANMF detector attains a much better detection performancethan the traditional ANMF detector does.The fifth part is contributed to the feature detection for floating small targets on seasurface. Presence of a surface target alters the scattering geometry of the sea surfacearound the target. It is shown that the observation in the high range resolution followsthe non-additive model rather than the traditional additive one. Under the non-additivemodel, target detection boils down to a special classification problem of the clutter-onlypattern. The average power, the Doppler offset and bandwidth are extracted andarranged into a three-dimensional feature vector. The tri-features-based detector isdesigned in the three-dimensional feature space. A convex-hull training algorithm isproposed to determine a three-dimensional decision region of the clutter-only patternfrom clutter-only data. The convex-hull’s polyhedron structure of the decision regionsupports a fast decision. The experimental results using the IPIX datasets show that thetri-features-based detector attains an excellent performance than available detectors doin detecting floating small targets on sea surface.