【作者】 黄贤源；

【导师】 翟国君； 隋立芬；

【作者基本信息】 解放军信息工程大学 ， 大地测量学与测量工程， 2011， 博士

【摘要】 本文对多波束测深数据质量控制的方法进行了较为系统和全面的研究。内容涵盖多波束测深系统误差源分析、多波束测深系统的声速改正技术、多波束测深异常值的探测技术、基于最小二乘支持向量机构建海底趋势面的算法研究、优化测深训练样本对海底趋势面构造的影响分析、多波束测深系统误差消除算法研究等等。取得的主要成果和创新点概括如下:1.针对传统经验正交函数分析方法在声速剖面构造方面存在的不足,提出了声速剖面经验模分解的方法,给出了修正后的声速剖面时间函数和空间函数,为测量海区声速变化模型的构建提供了必要的参数。通过实际观测数据的计算分析,有效验证了利用经验模分解的方法重构声速剖面的可行性和正确性。2.详细分析了多波束测深异常值产生的原因,并以此为基础对几种常见的多波束测深异常值探测方法从基本原理、适用条件、优缺点等方面进行了比较。结合遗传算法在异常值探测方面的优越性,提出了相应的测深异常值探测方法。根据遗传算法的择优选取原则,给出了测深异常值的探测步骤。实测算例表明基于遗传算法构建的异常值探测模型可以快速有效地标定出测深异常值。3.多波束测深是一个动态的过程,采集得到的水深数据具有不可重复性,针对这一特点,将Bayes估计理论和MCMC抽样设计方法引入到测深异常值的探测中。以水深观测数据为研究对象,详细推导了利用Bayes估计探测水深异常值的公式,并给出了详细的探测步骤。通过实际观测数据的计算分析,表明利用Bayes估计构造的算法能以简单明了的判断准则对测深异常值进行探测。4.深入研究最小二乘支持向量机算法的基本原理,并结合该算法在自由曲面重构方面的功能,将该算法引入到海底趋势面的构造中,并在此基础上对测深异常值进行探测;结合多波束测深数据海量性的特点,提出了有效的训练样本选取方案;给出了不同核函数的数学表达式,并进一步分析了不同的核函数对海底趋势面构造的影响;仿真算例表明为了避免海底微小地形的丢失,应结合海底地形的实际情况选取不同的核函数对海底趋势面进行构造。5.在分析不同核函数对海底趋势面构造影响的基础上,推导证明了趋势面滤波与最小二乘支持向量机算法在特定条件下的等价性,给出了公式推导过程;结合海底实际变化情况,提出了不同海底类型适用的核函数。利用实测数据对上述两种方法进行了分析和比较,得出在多项式核函数的阶数等于趋势面滤波的阶数,且权重系数等于0.5的情况下,利用最小二乘支持向量机构造的函数等价于趋势面滤波。6.最小二乘支持向量机在海底趋势面构造过程中无法消除较大偏差训练样本的影响,针对该缺陷,提出局部样本中心距优化测深训练样本的方法,从而降低了非正常训练样本对海底趋势面构造的影响。实测算例表明利用局部样本中心距能合理地优化测深训练样本,从而能合理地构造出海底趋势面,异常值也能得到有效地剔除。7.对多波束测深的各个误差源进行了定量的分析,给出了水深不确定度的计算公式;由于水深不确定度反映了测量水深的分散性程度,因此将其引入到测深训练样本的优化中,从而选取出对海底趋势面构造贡献率大的测深训练样本。实测算例表明利用不确定度能合理地优化训练样本,进而合理地构造出海底趋势面,异常值也能得到有效地剔除。8.详细地分析了多波束测深系统误差产生的原因,传统的两步滤波法虽然能有效地解决重叠区内测深数据的不匹配问题,但是并未顾及重叠区外的测深数据。针对该不足提出了基于不确定度的测深系统误差消除算法。实测算例表明利用不确定度修正的两步滤波法不仅能解决相邻条带重叠区内测深数据的不匹配问题,而且能顾及到重叠区外的测深数据,避免了海底地形发生扭曲。更多还原

【Abstract】 This dissertation mainly focuses on the method of multibeam echosounding data quality control which includes the principles of multibeam echosounding, the errors analysis, correction technology of sound velocity profiles for multibeam survey, detection technology of outliers for multibeam survey,research on constructing trend surface by LS-SVM, the influence of optimized train samples on elimination of sounding outliers in the LS-SVM arithmetic, correction technology of system errors for multibeam survey. The main works and contributions are summarized as follows:1. In the process of sound velocity profile conformation, the traditional empirical orthogonal function analysis method has some shortage, so a new approach based on empirical mode decomposition is presented. The orthogonal time functions and spatial functions of sound velocity profile are given. The new method supplies the essential parameters for the model conformation of survey area. By the calculation analysis of the actual observation data, the new method based on empirical mode decomposition could reconstruct sound velocity profile effectively and accurately.2. The cause of Multibeam echosounding outliers is analysed and a comparion of several common multibeam echosounding outliers’detection methods from the basic principle, the applicable condition, etc is given. Due to the abnormal value detection superiority of the genetic algorithm, the article puts forward the corresponding sounding outliers’detection method. Base on the selected principles of genetic algorithm, the steps for detecting outliers of sounding is given. The actual example shows that the model based on genetic algorithm could detect outliers quickly and effectively.3. Multibeam echosounding is a dynamic process; the depth is not repeating data. Aiming at the character; the Bayes estimation theory and MCMC sampling design method are imported to detecting outliers. Take the depth observation data as the research object, the formula of outliers detection be deduced by the Bayes estimation theory, and the steps of the detection is given; The actual calculation shows that the sounding abnormal value detection methods based on Bayes estimation theory could gain higher accuracy by simple judgment criterion.4. Further research the least squares support vector machine method, recur to the function of the free surface reconstruction, the ls-svm algorithm is imported to constructing the undersea trend surface, and the outliers could be detected by the trend surface.Multibeam echosounding data with the characteristics of mass, so the training sample selection plan is presented; Different kernel functions are given, and the influence of different kernel functions on seafloor trend surface are analyzed; Simulation examples show that in order to avoid the loss of seabed tiny terrain, should combine the actual situation of seabed terrain then choose the correct kernel function.5. On the basis of the influence of different kernel functions on seafloor trend surface structure, the equivalence between trend surface filtering and least squares support vector machine method is proved, and the formulas derivation is given; Combined with the actual situation, different kernel functions are suited for corresponding sea area; A comparion between trend surface filtering and least squares support vector machine method is given.The conclusion is that trend surface filtering is the especial result for least squares support vector machine method when the weight parameter equal to 0.5.6. In the process of sea floor trend surface construction, least squares support vector machines algorithm cannot eliminate the influence of large deviation training samples, aiming at this defect, the article puts forward local sample center distance method.The method could optimize sounding training samples, reduce the influence of abnormal training samples. The actual example shows that sounding training samples could be optimized reasonably by the local sample center distance, thus undersea trend surface could be constructed reasonably, so abnormal values could be eliminated effectively.7. The quantitative analysis is given about the multibeam echosounding error sources, the formulas of the depth uncertainty is obtained; Because of the uncertainty reflects the dispersion of the water depth measurement, the article import it to optimize the sounding training samples. The large contribution rate of the training samples should be chosed by the method. Real calculational results show that the sounding training samples could be optimized reasonably, thus undersea trend surface could be constructed reasonably, so abnormal values could be eliminated effectively.8. Analyzes the cause of system error in the process of multibeam echosounding, although traditional two-step filtering method can solve the problem effectively in overlapping area sounding data don’t match, did not attend the other sounding data. Aiming at the shortage, algorithm about eliminating sounding system error based on uncertainty is presented. The actual example shows that the improved two-step filtering method considers the other sounding data to avoid the seabed terrain distortions.更多还原