海杂波多维特性与条件物理场映射的深度认知方法研究

The cognition methods and cognition degree of the characteristics of sea clutter, which is the working background of sea radar target detecting system, are of vital importance to the performance of target detecting. It has been a hot spot and a challenge to cognize accurately and deeply. With the large sample of sea clutter measured data and the time-varying multi-scale electromagnetic scattering theory of sea surface employed, this project analyzes the law of the influence of the physical parameters of the experimental condition such as sea environment and radar parameters on the sea clutter characteristics, and reveals the corresponding mechanism. This project breaks though the efficient diagnosing technique of the effective area of massive raw data of sea clutter, and realizes the effective extraction of multi-dimensional characteristics of sea clutter, which provide large sample data base to the deep cognition of sea clutter characteristics. Using the multi-layer auto-encoding deep learning network, this project investigates the high order, hidden and non-linear mapping relationship between multiple physical parameters and sea clutter characteristics, constructs the deep cognition model of the mapping relationship between the sea clutter multidimensional characteristics and the physical parameters of radar and sea environment, and thus achieve the deep cognition of sea clutter multidimensional characteristics based on the physical parameters of sea clutter experimental conditions. This research can not only support the requirement of multidimensional and high order sea clutter characteristics in oceanic nontraditional “low, small and slow” target detecting application, but also provide a deep theoretical foundation for oceanic remote sensing based on the sea clutter.

海杂波作为对海雷达目标检测系统的工作背景环境，其特性的认知方法和认知程度对目标的检测性能具有重要影响，如何实现精确、深入地认知一直是研究热点和难点。本项目结合大样本海杂波实测数据和时变多尺度海面电磁散射理论，分析和揭示海洋环境参数、雷达参数等测量条件物理场参数对海杂波特性的影响规律和机理，突破海量海杂波原始数据有效区域的高效诊断技术，实现海杂波多维特性的有效提取，为海杂波特性深度认知提供大样本数据基础。利用多层自编码深度学习网络，探究多物理场参数与海杂波特性的高阶、隐层、非线性映射关系，建立海杂波多维特性与雷达、海洋环境物理场映射的深度认知模型，实现基于海杂波测试条件物理场参数的海杂波多维特性深层次认知。研究成果不仅为海上“低、小、慢”等非传统目标检测应用对海杂波多维、高阶特性的需求提供支撑，还为基于海杂波的海洋环境遥感应用研究提供深层理论基础。

本项目针对传统的雷达海杂波认知方法和认知程度不足，严重制约海上雷达目标精细化检测的问题，开展了从海杂波测量条件物理场参数出发的海杂波多维特性深度认知方法研究。首先通过结合实测大样本海杂波数据和多尺度海面电磁散射数据，揭示了海杂波特性的主要影响因素，构建了用于海杂波认知的条件物理场参数数据集；其次基于计算机视觉中采用的深度学习特征提取技术，提出了雷达海面回波数据中海杂波有效区域的自动化诊断方法，并实现了海杂波的多维有效特征提取，解决了海量雷达海面回波原始数据的预处理问题，构建了用于海杂波特性认知的特性参数数据集；以上述构建的条件物理场参数数据集和杂波特性参数数据集分别作为输入和输出，本项目研究了海杂波特性参数的深度学习认知方法，建立了海杂波多维特性与条件物理场参数的映射关系模型，实现了从多个海洋环境参数出发的海杂波特性参数精细化预测和认知。研究成果可支撑海上雷达目标检测算法的改进设计和海洋遥感应用研究。