海杂波中小目标的特征检测方法

发布时间：2018-06-14 16:44

本文选题：海杂波 + 目标检测　；参考：《西安电子科技大学》2016年博士论文

【摘要】：海面监视雷达可以完成对大范围海面的预警、监视和探测,在军事和民用领域均有广泛应用。由于高分辨海杂波的空时非平稳特性,传统目标检测方法面临低检测概率,高虚警的问题,使得海杂波背景下的对海面目标特别是慢速、漂浮小目标的检测成为国内外专家和学者研究的重点与难点。本文在对实测雷达海杂波数据特性分析的基础上,主要研究在高分辨海杂波背景下海面慢速、漂浮小目标的检测问题。研究的问题包括：海杂波分形特性分析与目标检测方法、基于快速凸包学习的三特征联合检测方法、匹配与高分辨海杂波多普勒谱特性的检测方法以及基于块白化海杂波抑制的海杂波时频特性分析与目标检测方法。本文主要研究成果概括如下：1.分析了海杂波幅度时间序列的多尺度分形特性与分形特征的空时变性。在对海杂波扩展自相似建模的基础上,利用多尺度Hurst指数讨论了在三个尺度区间内海杂波时间序列的起伏的主导因素,定性的分析了噪声对序列分形特性的影响；通过对不同时间不同海况条件下采集的实测数据分析发现海杂波的分形特征如Hurst指数随外界条件如海态、雷达照射方向与浪向的夹角的不同而改变。提出了两个基于改进分形特征的海面漂浮小目标检测方法。实测数据表明,由于利用了海杂波序列的多尺度性和空时变性,基于多尺度Hurst指数和基于相对Hurst指数相比基于Hurst指数的方法有更好的检测性能。2.讨论了传统目标检测问题与异常检测中单分类问题的联系,为海面目标检测问题提供了新的解决途径。由于海面目标的复杂性和多样性,通常无法获取所有种类目标回波,因此我们将纯杂波回波看作正常观测,含目标回波看作异常观测,并从回波中提取三个在两种观测模式下具有明显差异的特征,分析了两种模式下特征向量在三维特征空间上的可分性。提出了利用正常观测训练回波样本,通过快速凸包学习算法在特征空间获得检测判决区域的非参数方法,并提出了三特征联合检测方法。与基于海杂波分形特性的目标检测方法相比,该方法可以在短的观测时间获得较好的检测结果。3.在实测海杂波数据基础上,对海杂波多普勒功率谱建模。与地杂波相比,海杂波具有较宽的多普勒带宽。我们在多普勒域将海杂波功率谱分为杂波占优单元、噪声占优单元以及杂波噪声混合单元,并将海杂波多普勒谱建模为一个正随机过程,该随机过程在每个多普勒单元满足具有不同形状参数和尺度参数的K分布模型。考虑到海面漂浮目标在多普勒域的能量扩散现象,提出了匹配于海杂波多普勒谱特性的双重检测方法。实测数据的实验表明当信杂比较高时该检测器具有很好的检测性能。4.高分辨海杂波的非平稳性导致在较长观测时间下,传统杂波白化方法在估计杂波协方差矩阵时无法得到足够的参考单元回波样本,降低了杂波抑制的有效性,进而限制了检测器在长时间积累条件下的目标检测性能。针对这一问题,我们提出了块白化的海杂波抑制方法,将海杂波背景下的检测问题转化为在近似白噪声背景下的检测问题,有效抑制时频平面上交叉项对回波能量累积的影响,在此基础上给出了：(1)基于时频脊引导的Hough变换的目标检测方法。提出的时频脊引导的Hough具有比传统Hough变换更小的计算复杂度,并能较好的积累目标能量,具有较强的实际应用价值；(2)基于改进凸包学习算法的时频双特征检测方法。利用特征分布的先验知识,改进的凸包学习算法可以更高效的获得检测判决区域。针对纯杂波回波和目标所在单元回波在时频平面上的时频脊的差异,给出了两种提取脊能量和脊全变差的方法,并用Bhattacharyya距离定量的评估了两种提取方法提取的两种模式下的特征在特征平面上的可分性。两个特征对于检测海面漂浮目标来说具有较好的互补性,通过实测数据验证,得到的双特征检测器具有很好的检测性能。
[Abstract]:Sea surface surveillance radar can achieve early warning, monitoring and detection of large scale sea surface. It is widely used in military and civil fields. Due to the non-stationary characteristics of high resolution sea clutter, the traditional target detection method faces the problem of low detection probability and high false alarm, which makes the sea clutter background to the sea target especially slow and floating. The detection of target has become the focus and difficulty of experts and scholars at home and abroad. On the basis of the analysis of the data characteristics of the measured radar sea clutter, this paper mainly studies the problem of low speed and floating small target detection in the background of high resolution sea clutter. The research problems include the fractal characteristic analysis and target detection method of sea clutter, based on the analysis of sea clutter and the method of target detection. The three feature joint detection method for fast convex hull learning, the detection method of matching and high resolution sea clutter Doppler spectrum characteristics and the time-frequency characteristic analysis and target detection method of sea clutter suppression based on the block white sea clutter suppression. The main research results are summarized as follows: 1. the multi-scale fractal characteristics of the sea clutter amplitude time series are analyzed. On the basis of the self similar modeling of sea clutter expansion, the dominating factors of the fluctuation of sea clutter time series in three scales are discussed on the basis of the self similar modeling of the sea clutter expansion. The influence of noise on the fractal characteristics of the sequence is qualitatively analyzed, and the measured data collected at different time and different sea conditions are measured. The data analysis shows that the fractal characteristics of the sea clutter, such as the Hurst exponent vary with the external conditions such as the sea state, the direction of the radar and the angle of the wave direction, are changed. Two methods for detecting the floating small targets on the sea surface are proposed based on the improved fractal features. The measured data show that the multiscale and space-time variability of the sea clutter sequence is based on the multiscale and space-time variability of the sea clutter sequence. The multiscale Hurst index and the method based on the relative Hurst index based on the Hurst index have better detection performance.2.. The relationship between the traditional target detection problem and the single classification problem in the anomaly detection is discussed, which provides a new solution for the problem of the sea surface target detection. For all kinds of target echoes, we regard pure clutter echoes as normal observations, which include target echoes as abnormal observations, and extract three features that have distinct differences in the two modes of observation from the echoes, and analyze the separability of the eigenvectors under the two modes in the three-dimensional feature space. The sample, using the fast convex hull learning algorithm to obtain the non parametric method of detecting the decision area in the feature space, and proposes a joint detection method of three features. Compared with the target detection method based on the fractal characteristic of the sea clutter, the method can obtain better detection results at short observation time, on the basis of the measured sea clutter data, the.3. is on the sea. Clutter Doppler power spectrum modeling. Compared with ground clutter, sea clutter has a wider Doppler bandwidth. In Doppler domain, we divide the sea clutter power spectrum into clutter dominant unit, noise dominant unit and mixed wave noise mixed unit, and model sea clutter Doppler spectrum as a positive random process. This random process is in each Doppler. The unit satisfies the K distribution model with different shape and scale parameters. Considering the energy diffusion of the floating target in the Doppler domain, a dual detection method matching the Doppler spectrum characteristics of the Yu Hai clutter is proposed. The experimental data show that the detector has a good detection performance.4. high resolution when the signal is high. The non stationarity of the sea clutter leads to the fact that the traditional clutter whitening method can not get enough reference unit echo samples when estimating the covariance matrix of the clutter in the long time of observation, which reduces the effectiveness of the clutter suppression and limits the detection performance of the detector under the condition of long time accumulation. The method of block whitening sea clutter suppression, the detection problem under the background of the sea clutter background is transformed into a detection problem under the approximate white noise background, and the effect of the cross term on the echo energy accumulation on the time frequency plane is effectively suppressed. On this basis, the target detection method based on the Hough transform based on the time frequency ridge guidance is given. The time frequency ridge is proposed. The Hough has a smaller computational complexity than the traditional Hough transform, and it can accumulate the target energy well, and has a strong practical application value. (2) a time-frequency dual feature detection method based on the improved convex packet learning algorithm. Using the prior knowledge of the feature distribution, the improved convex packet learning algorithm can get the detection area more efficiently. In view of the difference in the time frequency ridges of the pure clutter echo and the target unit echo on the time frequency plane, two methods of extracting ridge energy and ridge total variation are given, and the Bhattacharyya distance is used to quantitatively evaluate the separability of the characteristics on the characteristic flat surface of the two modes extracted by the two extraction methods. The two features are for the detection of the sea surface. The results show that the dual feature detector has good detection performance.
【学位授予单位】：西安电子科技大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TN957.52

【参考文献】