基于稳健海底趋势面滤波的多波束测深数据处理方法研究

Multibeam echosounders (MBES) are currently the best way to determine the bathymetry of large regions of the seabed with high accuracy. MBES systems measure simultaneously a series of depths in an athwartship direction. MBES is a dynamic measurement method on the moving platform. The measurement process is not only affected by the atmosphere, but also by the sea water movement. Due to the presence of instrument noises, harsh marine environment, survey vessel's sharp turn et al. The soundings contain a small amount of outliers, which are major factors influencing the accuracy of hydrography and seabed terrains production .The research in the field (automatic processing of MBES soundings) has drawn common concern of experts and scholars in home and abroad. The regular methods include interactive mode filter, trend surface filter, sounding statistical characteristic filter et al. .In this paper Sparse Weighted LS-SVM algorithm is applied to build seabed trend surface, while the series of problems must be researched which affects the seabed trend surface constructing, then the Multi beam soundings must be processed and the Multi beam outlier or noise must be removed. The Sparse Weighted LS-SVM solve function estimation (seabed surface structure) instead of traditional SVM using quadratic programming method to solve it by changing inequality constrained quadratic optimization problem in the original space into a feature space equality constraints by using nonlinear mapping. Based on the soundings analysis, the feature select and extract methods of the multi-beam soundings are put forward by Sparse Weighted LS-SVM algorithm（Sparse LS-SVM）; Based on the weight ratio relation of uncertainty, the computing method of training sample fuzzy membership degree are put forward（Weighted LS-SVM）, and the model of robust adaptive weighted bottom trend surface must be build. The achievements will construct the seabed trend surface which represents the real changes of seabed floor, then the multi-beam soundings processing results become reasonable. The project has important theoretical significance and very good application prospect.

多波束测深数据处理对海底地形的真实反映至关重要，常规的数据处理方法包括人工交互式滤波、基于测深数据统计特性的滤波、基于趋势面构造理论的数据处理方法。课题的主体思路紧紧围绕趋势面构造理论，将Sparse Weighted LS-SVM算法应用于稳健海底趋势面构建，验证常规的趋势面滤波法仅是Sparse Weighted LS-SVM算法在取特定参数时的特例，同时对影响海底趋势面合理构建的一系列问题进行研究，并在此基础上对多波束测深数据进行处理，探测多波束测深数据异常及削弱冗余噪声的影响。课题的主要研究内容：（1）LS-SVM算法的稀疏性研究（Sparse LS-SVM）；（2）LS-SVM算法的稳健性研究（Weighted LS-SVM）。项目成果将有效地构建出反映海底地形真实变化情况的趋势面，在此基础上进行的多波束水深测量数据处理更为合理，课题研究具有重要的理论意义及较好的应用前景。

项目背景：多波束测深数据处理对海底地形的真实反映至关重要，常规的数据处理方法包括人工交互式滤波、基于测深数据统计特性的滤波、基于趋势面构造理论的数据处理方法。项目的主体思路紧紧围绕趋势面构造理论，将Sparse Weighted LS-SVM算法应用于稳健海底趋势面构建，同时对影响海底趋势面合理构建的一系列问题进行研究，并在此基础上对多波束测深数据进行处理，探测多波束测深数据异常及削弱冗余噪声的影响。.项目主要研究内容：（1）在分析多波束水深测量数据采集特点和规律的基础上，结合实测的多波束数据，给出核函数选取及超参数计算的方法；（2）构建完备的算法模型，给出计算准确的拉格朗日乘子，对测深训练样本进行特征选择与抽取方法研究；（3）开展利用残差量和垂直不确定度归算测深训练样本权系数方法研究；（4）利用Sparse Weighted LS-SVM算法构建海底趋势面模型。.项目重要结论：项目取得的成果不仅是对LS-SVM算法在解决实际问题的进一步研究，而且对多波束测深数据的处理，特别是多波束测深异常数据的探测有着积极贡献。.项目科学意义：LS-SVM算法在海底趋势面模型构建的成功应用将进一步丰富LS-SVM理论，项目具有重要的科学与实践意义。