Sentinel-1双极化数据舰船目标几何特性提取  

Ship geometric parameter extraction for Sentinel-1 dual-polarization products

作　　者：李博颖 柳彬[1] 郭炜炜[1] 张增辉[1] 郁文贤[1] 

机构地区：[1]上海交通大学智能探测与识别上海市重点实验室,上海200240

出　　处：《科技导报》2017年第20期94-101,共8页Science & Technology Review

基　　金：国家自然科学基金重点项目(61331015)

摘　　要：舰船目标几何特性提取是合成孔径雷达(SAR)图像海上目标检测识别的重要基础。在具有几何真值样本的基础上,通过参数寻优和拟合回归,能够提高几何特性提取的精度,这在Terra SAR-X数据上已有研究。本文考虑Sentinel-1大部分情况下均能提供双极化数据这一特点,探索双极化信息能否进一步提升几何特性提取的精度。基于Open SARShip测试库,首先使用二维度滤波进行图像处理,该图像处理过程中的关键参数使用交叉熵方法进行寻优,在大样本基础上,得到最优参数;之后,在目标几何特性的图像处理提取结果上,综合传感器、环境、目标3方面信息,特别是融合双极化信息,使用多元线性回归模型进行拟合,得到比仅用单极化信息更高的几何特性提取精度,证实了双极化信息的可用性。The ship geometric parameter extraction is an essential basis for the marine target detection and classification for the SyntheticAperture Radar（SAR） images. With the assistance of the ground true value sample of the marine target size, the improvement of thegeometric dimension extraction can be achieved by the parameter optimization and regression, as verified in Terra SAR-X datasets. Takinginto consideration of the typical characteristics of the dual-polarization for the sentinel-1 products, this paper explores the usefulness of thedual-polarization fusion information. Based on the Open SARShip, firstly we utilize a two-dimensional filter method for image processing.The parameters in the image processing are optimized by a cross-entropy method based on the large dataset. Next, with the preliminaryextraction results, we combine the information from the sensors, the environment and the target, and especially the information from thedual-polarization. We employ a multiple linear regression model to obtain the precise physical dimensions. The size extraction performanceby the dual-polarization fusion information is much better than merely using the single-polarization information, which proves theusefulness of the dual polarization information.

关 键 词：合成孔径雷达(SAR) Sentienl-1 OpenSARShip 舰船目标 几何特性提取 双极化 

分 类 号：TN957.52[电子电信—信号与信息处理]