雷达海杂波非线性机理分析与多维特征学习

Due to the influence of radar parameters, marine environmental factors and the electromagnetic scattering coupling between the target and the sea surface, radar target echoes and sea clutter are characterized by different characteristics of nonlinearity, which are often difficult to be described exactly by the mono characteristic commonly used in traditional signal processing. Based on the nonlinear mechanism of sea clutter and the measured data under different conditions, the research project systematically studies the non-linear variation laws of sea clutter and target characteristics and the corresponding characterization method in multiple representation domains, i.e. time / space domain, frequency / dual-frequency domain, FRFT domain, HHT domain and so on. Then, the multidimensional characteristics of sea clutter in multiple representation domains are extracted, and deep learning methods such as the multi-modal convolution neural network are used to solve the construction problem of difference characteristics under high-dimensional nonlinear conditions. Here the multi-representation domain fine processing and deep learning methods are used to improve the radar echo information utilization and achieve the non-linear distinction between sea clutter and target in the characteristic space, in order to support the fine processing of sea radar clutter suppression and target detection. Specific research includes: (1) study on the cognition and nonlinear law of sea clutter fine characteristics under typical influencing factors; (2) study on the fine characteristics of the half-space target in multi-representation domain by the coupling mechanism; (3) study on the difference characteristic extraction and multidimensional characteristic learning for sea target detection.

受雷达参数、海洋环境因素以及目标与海面电磁散射耦合等因素综合影响，雷达目标回波与海杂波呈现特性各异的非线性变化特点，传统信号处理方法通常采用的单一特征难以准确描述。本项目基于海杂波非线性机理，利用多种条件下雷达实测数据，深入系统研究时/空域、频域/双频域、FRFT域、HHT域等多表示域中海杂波和目标特性非线性变化规律和表征方法，提取海杂波多表示域多维度特征，采用多模态卷积神经网络等深度学习方法解决高维非线性条件下差异特征构造难题，综合多表示域精细化处理和深度学习方法提高雷达回波信息利用率，在特征空间实现海杂波与目标的非线性区分，为对海雷达杂波抑制和目标检测精细化处理提供支持。具体研究内容包括：（1）典型影响因素下海杂波精细化特性认知与非线性规律研究；（2）耦合机制作用下半空间目标多表示域精细化特性研究；（3）面向海上目标探测的差异特征提取与多维特征学习。

受雷达参数、海洋环境因素以及目标与海面电磁散射耦合等因素综合影响，雷达目标回波与海杂波呈现特性各异的非线性变化特点，传统信号处理方法通常采用的单一特征难以准确描述。面向雷达海上目标探测性能提升的需求，开展了如下几方面的研究：.（1）研究了典型影响因素下海杂波精细化特性认知与非线性规律。采集了X、S、Ku波段雷达、不同擦地角、HH/VV极化方式、不同海况等级、不同浪向等多种条件下的对海探测数据，并基于此数据，分析了海杂波幅度分布特性、多普勒谱特性、时间与空间相关特性、非线性起伏特性、时频分布特性，梳理了海况、浪向、擦地角等典型因素影响下的特性非线性变化规律。.（2）研究了耦合机制作用下半空间目标多表示域精细化特性。基于实测数据，研究了半空间目标（即海上目标）的时域特征提取方法与统计特性分析、频域特征特征提取方法与统计特性分析、时频域和分数域特征提取方法与统计特性分析、双频域特征提取方法与统计特性分析，并给出了不同特征对海杂波与目标的差异化表征能力。.（3）研究了差异特征提取与基于多维特征学习的目标检测方法。利用海杂波特性和特征研究结果，提出了基于时-空/频域特征的复杂海杂波场景分类辨识方法、利用频域非线性特征的杂波抑制与目标检测方法、分数域海杂波抑制与目标检测方法，以及基于多维特征联合学习的海杂波中目标检测方法，并基于实测数据对所提方法进行了验证，相比于同领域经典方法，提升了目标检测性能。