【作者】 张涛；

【导师】 熊惠霖；

【作者基本信息】 上海交通大学 ， 控制科学与工程， 2019， 博士

【摘要】 舰船是人们生产生活中必不可少的交通工具,准确掌握舰船位置信息有助于人们更加合理地开发海洋资源、保护海洋环境。作为一种先进的对地观测技术手段,合成孔径雷达(Synthetic Aperture Radar,SAR)具有不受时间、光照、天气变化影响等特点。凭借这些优势,SAR技术目前已被人们广泛应用到各个领域。这其中,利用SAR数据特别是极化合成孔径雷达(Polarimetric Synthetic Aperture Radar,PolSAR)数据去检测海面舰船目标是研究的热点方向之一。然而,利用PolSAR图像进行舰船目标检测依旧存在着许多重、难点问题,如中高海况下的弱小舰船目标检测、舰船的方位向模糊去除、对弱小舰船目标散射特性探讨等问题。为了提高海面舰船目标检测精度,本文提出了几种新的舰船目标检测方法。本文主要贡献和创新概括如下:(1)大多数双极化舰船目标检测算法所用到的极化特征关联性较强,存在着相同信息重复使用的情况。针对这个问题,本文首先构建了一个新的6维极化散射特征向量,而后对该项量进行主成分分析(Principal Component Analysis,PCA),并提取出更深层的3维特征向量。最后,将提取出的3维向量与双极化几何扰动滤波器(Dual-pol Geometrical Perturbation-Polarimetric Notch Filter,GP-PNFd)相结合,构建新型舰船目标检测器GP-PNF6DPCA。实验结果在验明该算法有效性的同时也显示出经过PCA操作后的一系列GP-PNFPCA算法更加有助于中高海况下的弱小舰船目标检测。(2)对于全极化舰船目标检测,大部分检测算法只利用了目标像素点极化信息,而未利用目标周边杂波像素的极化信息。事实上,这些杂波像素信息也有助于舰船目标检测。考虑到这一点,本文在传统局部二值模式(Local Binary Pattern,LBP)特征的基础上,针对PolSAR图像数据特性,提出了一种计算目标散射差异信息的策略,并构建了极化协方差差异矩阵(Polarimetric Covariance Difference Matrix,PCDM)[P]。(3)在传统协方差矩阵[C]的基础上,提出了新的极化散射特征-基准船只等高(Pedestal Ship Height,PSH),并证明了它是目标去极化度的另一种度量标准。同时为了解决舰船检测任务中常见的方位向模糊问题,本文利用PSH特征提出了两种不同的方位向模糊去除方法:1)将ISPAN和PSH相结合的BPCDM方法;2)将 PSH 和全极化几何扰动滤波器(Geometrical Perturbation-Polarimetric Notch Filter,GP-PNF)相结合的IGP-PNF方法。实验结果表明这两种算法在去除目标方位向模糊的同时也能够有效地检测弱小舰船目标。(4)考虑到PolSAR图像中的相位信息有助于舰船目标检测,同时也为了弥补[P]矩阵中缺失的相位信息,本文在计算[P]矩阵策略的基础上又提出了一种新的目标差异信息计算策略,并进一步构建了完备极化协方差差异矩阵[(CP]。实验结果表明,和[C]和[P]矩阵相比,[CP]矩阵能够更加有效地改善相关舰船目标检测器的检测性能(5)本文在对舰船目标,特别是对弱小舰船目标的散射类型进行分析后发现,表面散射是弱小舰船目标的主要散射成分之一,在检测算法构建中不该被忽略。鉴于这个事实,为了更加有效地检测中高海况下的弱小舰船目标,本文提出了RsDVH舰船检测算法。该算法在利用目标[CP]矩阵信息的同时也利用了目标原始极化协方差矩阵[C]的信息,实验结果证实了 RsDVH的有效性。此外,不同的实验结果也反映出综合利用目标[C]和[CP]矩阵的信息能够更加有效地检测弱小舰船目标。更多还原

【Abstract】 Ship is the primary transportation on the sea and plays an important role in people’s life and production.Therefore,monitoring the geographical locations of ships are also crucial for us to more reasonably exploit and protect sea resources.Synthetic Aperture Radar(SAR)has been widely used in earth surveillance due to its unique advantage of independence of both time and weather conditions.As one important application,ship detection using SAR data,particularly Polarimetric SAR(PolSAR)data has attracted lots of attention in recent years.Although lots of research efforts have been devoted to ship detection,ship detection is still a challenging work due to the noise effects resulted from sea surface.In order to more effectively detect ships,especially the small ships in the case of moderate/high sea state,we propose some new methods in this paper.Generally,the contributions can be summarized as follows:(1)The polarimetric features used in most dual-pol ship detectors are correlated to each other.Under this condition,some same polarimetric information included in these features may be reused.To solve this problem,we propose a new 6D(6-Dimension)po-larimetric feature vector t6D in this paper.Then,PCA(Principal Component Analysis)is adopted to extract much deeper features from t6D and construct a new 3D(3-Dimension)feature vector t6DPCA.Finally,we replace the feature vector of GP-PNFd(Dual-pol Geo-metrical Perturbation-Polarimetric Notch Filter)with t6DPCA and develop a new method GP-PNF6DPCA.Experimental results show that the proposed method can detect ships more effectively,and a series of GP-PNFPCA methods after PCA operator are also useful to detect small ships in the case of moderate/high sea state.(2)Many ship detectors only use the characterises of ship pixel to detect ships in Quad-pol SAR image,which belongs to the category of ’pixel to pixel’.Actually,the clutter pixels surrounding ship pixels also include some significant information,and can help us to detect ships.Hence,based on this fact,we propose a new scheme in this pa-per to calculate the difference information between ship pixel and its surrounding pixels.In detail,the traditional LBP(Local Binary Pattern)model is first improved for PolSAR image,and then the polarimetric covariance difference matrix[P]is constructed.(3)A new polarimetric feature PSH(Pedestal Ship Height)is derived from[C]in this paper.Then,we verify that PSH is another metric of the depolarization of target.Tak-ing advantage of this fact,two methods are proposed for removing azimuth ambiguities in this paper.The first one is developed by combining I SPAN and PSH together,the second one is constructed by improving GP-PNF(Geometrical Perturbation-Polarimetric Notch Filter).Experimental results also show that these two detectors can not only remove azimuth ambiguities,but also detect small ships.(4)Considering that the phase information is useful for ship detection,thus,we pro-pose another scheme for calculating the scattering difference between ship pixel and its surrounding pixels.Then,the complete polarimetric covariance difference matrix[CP]is constructed.Compared to[P],the phase information is obviously included in[CP].Subsequently,a ship detection method DBSP is proposed using[CP].Experimental re-sults demonstrate thatin competition with[C]and[P],[CP]can more effectively help ship detectors improve their performance.(5)Analyzing the scattering mechanisms of ship,we find that the surface scattering is one main scattering component of small ship.However,many researchers neglect this fact and avoid using the surface scattering for detecting small ships.In this paper,one ship detector called RsDVH is proposed by combining[C]and[CP]together.Experi-mental results demonstrate that RsDVH is useful for ship detection.Besides,the results also show that the way of combining[C]and[CP]has the potential to detect small ships.更多还原