【作者】 李杨；

【导师】 张宁；

【作者基本信息】 哈尔滨工业大学 ， 信息与通信工程， 2010， 博士

【摘要】 大范围、全天候、实时超视距探测能力使高频地波雷达(High Frequency Surface Wave Radar, HFSWR)广泛应用于海态遥感、海面目标检测和经济专属区监视等军民两用领域。垂直极化电磁波沿海面绕射传播的超视距探测机理使HFSWR舰船目标的检测不同于传统微波雷达中的检测。多种杂波与噪声(如海洋表面回波、电离层杂波、电台干扰、流星余迹、大气噪声等)的同时存在造成了HFSWR检测背景的复杂化和多样化特征。分辨单元内多源散射造成同质(亦称均匀,homogenous)/异质(亦称非均匀,nonhomogenous)回波共存,导致检测背景不仅包含高斯类杂波,而且包含时变起伏的非高斯类杂波。这些都给舰船目标检测造成了很大困难。为此,本文通过对HFSWR检测环境的深入分析,总结了HFSWR检测环境的特点和检测难点,并根据实际系统需求提出了一种HFSWR背景感知信息处理系统。该系统包括背景感知信息提取(亦称检测场景分析)和多策略/参数检测两部分。前者通过特征提取或者模型的建立,从而完成对目标所在检测环境信息的提取和认知。包括对已有杂波的识别,对检测区域的分割和评价,检测背景分类及其统计特性分析等。后者则利用获得的背景感知信息,针对目标所处的不同环境,通过选择相应的检测策略、设置不同的检测参数实现优化检测的目的。本文将针对上述几个关键技术展开研究,主要内容如下:1.一阶海杂波(亦称Bragg峰)的识别对HFSWR具有重要意义:一方面,在海态遥感中,其位置参数可用于海洋参数反演;另一方面,在海面动目标检测中,由于一阶海杂波具有类目标特征,因此常常造成船目标检测结果中的虚警和漏警。针对上述问题,本文提出了一种提取距离-多普勒(Range-Doppler, RD)谱脊特征作为Bragg峰位置指示信息,并结合特征知识的高频一阶海杂波及其分裂谱峰的识别方法。通过实测数据验证表明,与经典的峰值检测和仅使用特征知识的方法相比,本文提出的方法无论在船目标干扰条件下还是在强/弱洋流切变环境下,都可以获得更好的识别效果。2.介绍了一种基于最大类间差的HFSWR检测区域分割方法。该方法利用检测背景在RD谱中的统计特性,将检测区域分割为强散射回波区、中等散射回波区、弱散射回波区和参考噪声区4个部分。该结果的一个重要应用是对扩展E层、F层电离层杂波的识别。由于覆盖范围广、强度大、时变起伏并具有不规则分布,扩展E层、F层电离层杂波对HFSWR性能造成很大威胁。为此,本文提出了一种基于区域分割结果并结合杂波区域特征的扩展E层、F层电离层杂波识别方法。首先,使用卷积模板确定杂波边缘,然后利用分割区域类型的采样占有率作为杂波区域的限定门限,从而成功的实现了对扩展电离层杂波区域的识别。实测数据结果表明,该方法可以准确的描述扩展电离层杂波对雷达检测背景的影响,对其定量的评价分析结果与实测结果相符。这一结果对杂波抑制、选频和雷达系统性能的评价有重要的参考和应用价值。3.使用检测区域分割结果并结合检测背景的物理特征,实现了对HFSWR检测背景的分类。为了剔除各类检测背景中的异质采样,使用一元非线性回归分析方法对实际检测背景进行了预检测处理,并对处理后的各类检测背景的统计特性进行了分析。实测数据结果表明,经过预检测处理后的各类非大气噪声背景,其统计分布分别服从形状参数不同的Weibull分布,并且具有比预检测处理前更好的似然效果;而对于成分单一的大气噪声类检测背景,无需进行预检测处理即可获得其背景统计分布信息。该结果是多策略/参数检测的重要依据。4.提出了基于背景感知信息的HFSWR多策略/参数检测方法。该方法利用前文所述技术,通过在线获取检测场景信息,经过杂波剔除、检测区域分割、检测背景分类、概率分布估计、峰值检测和Weibull双参数归一化处理等过程,实现了多策略/参数检测任务。由于根据不同类型背景的特点选择不同的检测策略/参数,本文提出的方法避免了以往检测过程中由于检测模型失配造成的检测损失。文中使用仿真目标加实测背景数据的方法进行了Monte Carlo仿真实验,并使用含有非合作目标的实测数据处理结果进一步验证了算法的性能。通过与国际上经典的和最新的高频雷达检测方法比较发现,基于背景感知信息的多策略/参数检测方法在保证虚警概率的前提下,具有更好的小目标发现能力。与经典的和基于知识的(Knowledge-Based, KB)检测结构不同,本文通过单一类型传感器在线实时提取检测环境信息,提高了对目标所处检测环境的感知能力,从而为不同环境下的多策略/参数检测提供决策依据,可为雷达检测性能的提高、系统评价和智能管理提供有力的帮助。更多还原

【Abstract】 High Frequency Surface Wave Radar (HFSWR) is widely used in both millitary and civilian areas such as sea state sensing, surface target detection and the surveilance of the Exlusive Economic Zone (EEZ) for its excellent capability of long range, all weather and real-time surveilance. Its mechanism utilizing a vertical polarization electromagnetic wave that follows the curvature of the earth along the air-water interface with low propagation loss on highly conductive ocean surface makes ship target detection in HFSWR different from that in microwave radar. Clutter such as sea echo, ionospheric clutter, radio frequency interference, meteor, and environment noise cause the complexity and diversity of the detection environment. Multi-source scatter makes homogenous and non-homogenous echoes coexistent which results in the detection background consists of not only Gaussian clutter but also unknown, time-variant, and fluctuated non-Gaussian clutter. All of these lead to great difficulties in ship target detection for HFSWR.Thus, by deeply analyzing the detection environment, this dissertation summarizes detection characteristics and difficulties of HFSWR. Then an architecture of background cognitive information processing system for HFSWR is presented according to these requirements. The system consists of background cognitive information extraction (namely detection scene analysis) and multiple strategies/parameters detection. The former extracts and cognizes the environment information of the detection background by characteristics extraction and model establition, which includes clutter identification, detection region segmentation and evaluation, detection background classification and statistical analysis. The latter utilizes the background cognitive information for selecting proper detection strategies or setting parameters to optimize the detection performance. This dissertation is composed of the studies on the above-mentioned key techniques, and the outline is as follows:1. First-order sea clutter (namely Bragg peaks) plays an important role in HFSWR. On the one hand, its location paramter can be inversion tools for sea state sensing. On the other hand, its quasi-target features often result in false alarm and missing alarm in ship target detection. To solve these problems, this dissertation proposes an algorithm to identification single and splitting first-order sea cluter in HF band which combines characteristic knowledge with the location indicative information extracted from ridge feature of Bragg peaks in Range-Doppler (RD) map. Real data experiments show that the proposed algorithm, comparing with the classical peak detection method and the characteristic- knowledge-based method, can have a better identification performance even in the environment with ship targets as interference or in the strong/weak current shear condition.2. A detection region segmentation method is intorduced based on maximizing the separability of the resultant classes. This method uses statistics of the detection background in RD map to segment the detection region into four parts: strong scattering area, medium scattering area, weak scattering area and reference noise area. One of its most significant applications is to identificate the spread E, F layer ionospheric clutter. Wide range covering, strong intensity, time-variant, flucuated and irregular distribution of the spread E, F layer ionospheric clutter badly affects the system performance of HFSWR. A spread E, F layer ionospheric clutter identification method is proposed based on the region segmentation results and region characteristics of the clutter. First of all, convolution template is used for locating the edge of the clutter, then the ratio of the number of the samples belonging to some segmented region and the total number of the samples in the region of interest is used for setting the determinative threshold of the clutter region. Experiments manifest that the proposed method can describe the effect of the spread ionospheric clutter to HFSWR. The quantitative analysis is consistent with the real data observation. The result can be used as a worthwhile reference for clutter mitigation, carrier frequency selection or radar system evaluation.3. Detection background classification is achieved by combining the region segmentation results and the physical features of the detection background. Unary non-linear regression analysis method is utilized for pre-detection which elimites the non-homogenous samples in the detection background. Then, statistical analysis is operated on the different types of the detection background. Experiments with real data show that the statistical distribution in the pre-detected non-atmospheric noise background follows Weibull distribution with different shape parameters with a better likelihood than that without pre-detection. While the statistical distribution information of atmospheric noise background consisting of single component can be got without pre-detection. The result is an important basic for multiple strategies/parameters detection method.4. Background-cognitive-information-based multiple strategies/ parameters detection method for HFSWR is proposed. Multiple strtegies/parameters detection method is implemented by utilizing on-line accessed detection scene information, including clutter elimination, detection region segementation, detection background classification, statistical distribution estimation, peak detection, and 2-parameter normalization of Weibull distribution, etc. Selecting proper detection stategies/ parameters according to the different characteristics of the background help this method reduce the detection loss which often happens in the classical detectors because of model mismatch. Monte Carlo simulation with sythetic targets in real data as background is operated and experiments with real data is used for further verifying the performance of the algorithm. Comparing the proposed method with the classical and latest HF detection algorithms, it’s found that the background-cognitive-information-based multiple strategies/paramters method has a better capability of detecting weak targets within a proper false alarm rate.In summary, compared with the classical and the Knowledge-Based (KB) detector, the proposed methods can extract the information of detection environment on-line by a single type of sensor. The capability of being aware of target’s surrounding environment is improved which is the decision basis for multiple strategies/parameters detection in different environment and can provide effective help in detection optimization, system performance evaluation and intelligent management.更多还原