| High-frequency surface wave radar(HFSWR)has been widely used in maritime surveillance(e.g.,exclusive economic zone)for its ability of over-the-horizon,widerange,all-weather and real-time surveillance.Since the electromagnetic environment of the high-frequency electromagnetic wave is very complex,the radar echoes of HFSWR systems contain various types of time-varying clutter and interferences,resulting in a multi-source and time-varying detection background.Additionally,due to the different generation mechanisms of clutter,the same/heterogeneous nonGaussian distribution echoes coexist and the detection background includes nonhomogeneous detection scenarios such as multiple targets and clutter edges,resulting in nonhomogeneous and non-Gaussian detection backgrounds.In the complex detection environment of HFSWR,ship target detection is a challenging task.Aiming at the problem of HFSWR target detection,this dissertation adopts the architecture of "detection scene analysis-adaptive detection".The clutter recognition and statistical characteristic analysis technologies are used to analyze the detection scene deeply and achieve the multi-source and time-varying environment information extraction and perception.Then,based on the perceived environment information,the knowledge-based adaptive constant false alarm rate(CFAR)detection algorithms are used for target detection to improve the detection performance in the complex environment of HFSWR.Based on the detection architecture,this dissertation will carry out research on HFSWR intelligent target detection method,which mainly includes the following contents:1.In the multi-source detection environment,to solve the problem that the firstorder sea clutter has target-like characteristics and affects the target detection performance,by extracting multi-dimensional features that can distinguish the firstorder sea clutter from the target and using the multi-dimensional information of the first-order sea clutter,this dissertation proposes a first-order sea clutter recognition method based on multi-dimension feature learning.The proposed method first extracts multi-dimension features such as dimensionless shunting color vision and Doppler frequency-related features to represent the prior information on the local dominance of the first-order sea clutter amplitude,the predictable Doppler frequency range,and the geometric shapes.Then it uses the support vector machine classifier to establish a multi-dimension feature interface,and finally realizes the automatic recognition of first-order sea clutter.By using simulated and measured data,it is verified that the proposed method can accurately recognize the first-order sea clutter region and reduce the adverse effect of target-like characteristics on target detection performance.2.In the multi-source detection environment,to solve the difficult problem of complex shape clutter recognition,by using the spatial continuity and geometric shape information of clutter,and the Doppler frequency information of sea/land clutter,this dissertation proposes an improved U-Net clutter recognition method by introducing an attention mechanism.To reduce the mutual interference between the clutter in the recognition process,the sea/land clutter and ionospheric clutter are firstly enhanced by using bilateral filters.Then,based on the location information of the sea/land clutter,the improved U-Net recognition network with the attention mechanism is used to extract multi-scale features,and recognize different clutter regions.It is verified by simulated and measured data that the proposed recognition method can accurately recognize different clutter regions.3.In the nonhomogeneous and non-Gaussian environment,to solve the problem of the detection performance degradation caused by multi-target interferences and the increase of false alarm rate caused by clutter edges,by directly or indirectly exploiting the prior knowledge in the detection background,this dissertation proposes three clutter and target knowledge-based adaptive CFAR detection algorithms.Under the assumption that the Weibull shape parameters are known,the robust variability index CFAR for the Weibull background(RWVI-CFAR)detection algorithm is proposed by indirectly using the statistical prior knowledge of clutter and synthesizing the advantages of various CFAR detection strategies.It uses the known shape parameters and statistical prior information of clutter to analyze the detection scene in the reference window and then selects different reference samples and CFAR strategies for different detection scenes.Under the assumption that the shape parameters of the Weibull distribution are unknown,when the target is a point target,the block matching 3-D filtering-based CFAR(BM3D-CFAR)detection algorithm is proposed by directly using the prior knowledge of the sparsity of the target,the non-local selfsimilarity and spatial correlation of distribution parameters of background samples based on the Bayesian framework;when the target has energy leakages,the low-rank representation-based CFAR(LRR-CFAR)detection algorithm is proposed by directly using prior knwoledge of the geometric shape and sparsity of the target,and the nonlocal self-similarity and spatial correlation of distribution parameters of background samples based on the Bayesian framework.The two algorithms are based on the Bayesian framework and use prior information to establish the parameter optimization model.Then,the model is optimized by designing the corresponding parameter estimation algorithm,estimates of the distribution parameters are obtained,and the target detection is carried out.The robust detection performance and false alarm control ability of the proposed three CFAR detection algorithms in nonhomogeneous environments are verified by simulated and measured data experiments.4.In the time-varying detection environment of the HFSWR system,to solve the difficulty of ship target detection,based on the architecture of "detection scene analysis-adaptive detection",two HFSWR intelligent target detection methods are proposed.By using the environmental perception information and the prior knowledge of clutter statistics,this dissertation proposes an environmental information-based intelligent multi-strategy(IMS)target detection method.It first uses the clutter recognition and statistical characteristic analysis techniques described above to obtain enough environmental perception information.Then,in the adaptive detection stage,to make full use of the perception information,it adaptively adjusts the shape of the reference window according to the region where the target is located and uses the multi-strategy RWVI-CFAR detection algorithm based on the prior knowledge of clutter statistics to detect targets.To further improve the detection performance in complex environments,by using the environmental perception information and the prior knowledge of geometric shape information and spatial correlation of the background clutter,this dissertation proposes an environmental information-based intelligent multi-parameter(IMP)target detection method.It uses clutter recognition technology and statistical characteristic analysis to obtain realtime environmental perception information.And at the adaptive detection stage,it uses the multi-direction dictionary learning-based CFAR(MDDL-CFAR)to detect targets.As a knowledge-based multi-parameter Bayesian CFAR detector,the MDDLCFAR detector not only uses the online environment perception information but also uses the geometric shape information and spatial correlation of the background clutter to establish the distribution parameter optimization model.The model is optimized by designing a parameter estimation algorithm,and the distribution parameter estimates are obtained for target detection.The effectiveness of the proposed two target detection methods is verified by a combination of simulated and measured data. |
