紧缩极化SAR海面油膜特征提取和探测方法研究

Polarimetric SAR can distinguish oil film and weak damping look-alikes on the sea surface more accurately. However, it has limited swath-width. The compact polarimetric SAR can acquire the partial polarization information of the target and obtain doubled swath-width of the full polarimetric SAR system, which is suitable for operational marine oil spill surveillance. The backscattering characteristics of the sea surface are significantly different in different radar polarization modes, and the feature extraction, optimization, and classification algorithm have a significant impact on the performance of the oil film detection method. This study intends to establish a uniform information matrix of general transmitting and linearly receiving (GTLR) polarization modes. Based on polarization signature, probability density function, and the statistical distance, the behavior of polarimetric features of the oil spill and look-alikes changing with radar polarization modes are quantitatively analyzed. Then the optimal features are extracted. Based on extended Bragg scattering (X-Bragg) model, polarimetric SAR features are analyzed and the parameters of oil films are retrieved. By using the feature reduction methods, the polarization characteristics are optimized, and the number of layers, the number of hidden elements and the activation function of the feed forward neural network (FFNN) are adjusted to achieve the accurate detection of oil slicks. The research is helpful to the all-weather monitoring of the marine environment.

极化SAR系统能够准确地区分海面油膜和弱衰减（weak damping）似然物，但是其观测带宽度有限。紧缩极化SAR能够在获取目标地物部分极化信息的同时获得两倍于全极化SAR系统的观测带宽度，更加适合业务化海面溢油监测。海面的后向散射特征在不同雷达极化方式下具有明显差异，并且特征的提取、优化和分类算法显著影响海面油膜检测方法的性能。本课题拟构建一般发射线性接收（GTLR）极化模式下统一的信息矩阵，利用极化特征图，概率密度函数和统计距离等手段定量分析海面油膜及其似然物的极化特征随雷达极化模式的变化特性，对最优特征进行提取。利用扩展的Bragg散射(X-Bragg)模型对海面后向散射进行建模分析和油膜状态参数反演。利用特征降维方法对极化特征进行优化，对前馈神经网络（FFNN）的层数、隐元数和激活函数进行调整，实现准确的海面油膜探测。研究有助于海洋环境的全天候监测。

目前，海面溢油已成为威胁海洋生态环境的主要因素之一，及时、准确的海面溢油遥感监测有助于溢油事故的处置及追责。合成孔径雷达（SAR）以其全天时、全天候的观测优势成为了当前常用的环境遥感监测手段。然而，生物油膜、海面禁风区、海洋内波等似然物的存在较大地限制着SAR海面溢油监测算法的精度。极化SAR系统能够基于海面目标的极化特性准确地区分海面油膜和弱衰减（weak damping）似然物，但是其观测带宽度有限；紧缩极化SAR能够在获取目标地物部分极化信息的同时获得两倍于全极化SAR系统的观测带宽度，更加适合业务化海面溢油监测。然而，海面的后向散射特征在不同雷达极化方式下具有明显差异，并且特征的提取、优化和分类算法显著影响海面油膜检测的准确度。本课题围绕紧缩极化SAR海面溢油监测的关键问题开展研究，利用多种统计分布对洁净海面及油膜的极化SAR后向散射特征进行拟合，获得不同极化特征下洁净海面及油膜的统计分布特性；提出了改进的倾斜调制布拉格散射海面油膜后向散射模型，为极化SAR和紧缩极化SAR海面油膜特征的分析和优化奠定了理论基础；建立了统一的海面油膜紧缩极化特征分析框架，探索海面油膜监测的最佳极化工作模式和极化特征，为紧缩极化SAR海面油膜监测平台的设计和算法的研发提供了依据；开展了基于深度学习的海面油膜检测算法研究，通过堆叠自编码器对神经网络权重进行预训练，提高了海面溢油检测模型的收敛速率和拟合精度，通过融合极化信息和后向散射强度信息改善了油膜检测结果；提出了一种基于卷积神经网络和超像素分割的极化SAR海面溢油监测算法，充分利用极化SAR图像的极化信息和语义信息，实现了海面油膜及其似然物的高精度分类。研究有益于海面溢油的预警监测及海洋生态环境的保护。