利用BP神经网络剔除多波束测深数据粗差

Detecting Outlier of Multibeam Sounding with BP Neural Network

针对多波束单ping水深数据多呈现较为复杂的曲线形式的现象,提出了基于逆传播(back propagation, BP)神经网络的多波束测深数据粗差剔除方法,即依据BP神经网络具有从输入到输出的映射功能,构建适应多波束单ping水深数据复杂曲线的训练学习算法进行曲线拟合。考虑地形之间的延续性进行相邻ping水深数据间的相关性分析,纵向检查定位并剔除粗差。通过实测多波束测深数据验证该方法的有效性,并与不确定性与测深学联合估值滤波以及交互式滤波方法进行比对分析,结果表明该方法可以有效剔除多波束测深数据中的粗差。

A method of detecting outlier of multibeam sounding with back propagation(BP) neural network is proposed in this paper for the complexity of bathymetric data of a ping. This paper constructs a training and learning algorithm for complex curve of multibeam single ping data for curve fitting based on the mapping function from input to output of BP neural network. Then it inspects the results from the previous steps lengthways by the correlation analysis of data of adjacent pings, and a vertical check to locate and remore outlier is also proposed. The experiment is conducted using the real bathymetric data, where there is a shipwreck in the middle. And also the result is compared with the combined uncertainty and bathymetry estimator(CUBE) algorithm, which is a popular method in detecting outlier of multibeam sounding at present. The experiment proves that the method proposed in this paper can detect the outlier more effectively.