SAR图像近港舰船目标检测技术研究

发布时间：2018-08-08 19:45

【摘要】：利用合成孔径雷达(Synthetic Aperture Radar,SAR)图像进行舰船目标检测对于军事情报获取、海洋监视和渔业管控意义重大,已经成为海洋遥感领域的研究热点。在近港区域,舰船集中停靠,进出频繁,极具侦查价值,因此,研究SAR图像近港舰船目标检测技术具有重要的现实意义。本文以准确提取舰船目标为目的,针对近港SAR图像中排除陆地干扰、剔除杂波虚警等问题,采用理论结合实践的方法,对SAR图像海陆分割、舰船目标检测与鉴别等关键技术进行了深入研究。舰船检测本质上是数据逐级筛选问题,针对近港区域SAR图像,本文主要通过海陆分割、目标检测、虚警鉴别三级筛选达到舰船目标检测的目的。海陆分割是去除陆地区域,目标检测是从海洋中提取可能是舰船目标的ROI切片,虚警鉴别是从检测结果中剔除虚警,最终输出舰船目标。根据以上思路,开展工作如下:在SAR图像中,近港区域背景复杂,码头和舰船均属于强散射目标,灰度接近,当舰船停泊在码头附近时,在图像上表现为与码头连成一体,传统目标检测方法难以正确分离舰船与码头。针对这一问题,本文提出一种新的舰船检测方法,以海陆分割为基础,将同一区域光学图像作为先验知识,通过SAR图像和光学图像的自动配准,将港口的光学模板准确映射到SAR图像上,从而把靠岸舰船与码头分离,然后在限定的海洋区域内进行全局CFAR检测,快速提取舰船目标。基于特征的鉴别方法是目前应用最为广泛的目标鉴别方法,本文针对舰船目标和杂波虚警的差异,基于变化检测技术,提出一种新的鉴别特征—目标像素聚集度特征。该特征可定量评估切片分割后强散射目标像素在目标区域的聚集程度,从而将真实目标和杂波虚警区分开。此外,舰船目标的几何特征也是一种重要的鉴别特征,然而,由于SAR相干成像机理,舰船目标上容易出现“拖影”或“十字叉”,给舰船目标的几何特征提取带来困难。针对这一问题,本文根据舰船轮廓近似椭圆形的特点,提出一种基于椭圆拟合的舰船目标几何特征提取方法。通过多组真实SAR数据的实验验证表明,该方法可以在一定程度上克服“拖影”和“十字叉”的不良影响。
[Abstract]:Ship target detection using synthetic Aperture Radar (Synthetic Aperture) images is of great significance for military intelligence acquisition, marine surveillance and fisheries control, and has become a research hotspot in the field of marine remote sensing. In the near port area, ships call in and out frequently, which is of great value for detection. Therefore, it is of great practical significance to study the detection technology of ships near port in SAR images. Aiming at the problem of eliminating land interference and eliminating clutter false alarm in near-port SAR image, this paper uses the method of combining theory with practice to segment the SAR image by using the method of combining theory with practice. The key technologies of ship target detection and identification are studied in detail. Ship detection is essentially a data level screening problem. In view of the SAR images in the near port region, the purpose of ship target detection is mainly achieved by sea and land segmentation, target detection and false alarm discrimination. Land and sea segmentation is to remove land area, target detection is to extract ROI slices from the ocean that may be a ship target, false alarm identification is to eliminate false alarm from the detection result, and finally to output ship target. According to the above ideas, the work is as follows: in the SAR image, the background of the near port area is complex, the wharf and the ship belong to the strong scattering target, and the gray level is close. When the ship is moored near the dock, it appears to be connected with the wharf on the image. Traditional target detection method is difficult to separate ship from wharf correctly. In order to solve this problem, a new ship detection method is proposed in this paper. Based on the segmentation of land and sea, the optical image of the same region is taken as a priori knowledge, and the automatic registration of the SAR image and the optical image is carried out. The optical template of the port is accurately mapped to the SAR image, and then the docking ship is separated from the wharf, and then the global CFAR detection is carried out in the limited ocean area to extract the ship target quickly. Feature-based discriminant method is the most widely used target identification method at present. In this paper, a new discriminant feature, the pixel aggregation feature, is proposed based on the change detection technique for the difference between ship target and clutter false alarm. This feature can quantitatively evaluate the aggregation degree of the pixels of the strong scattering target in the target region after slice segmentation, and then distinguish the real target from the clutter false alarm. In addition, the geometric feature of ship target is also an important distinguishing feature. However, because of SAR coherent imaging mechanism, it is easy to appear "drag" or "cross" on ship target, which makes it difficult to extract geometric feature of ship target. In order to solve this problem, a method of extracting geometric features of ship objects based on elliptic fitting is proposed according to the approximate elliptical feature of ship contour. The experimental results of real SAR data show that the method can overcome the negative effects of "drag" and "cross" to some extent.
【学位授予单位】：国防科学技术大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：E925;TN957.52

【相似文献】