基于神经网络的水深插值研究

Interpolation of Charted Depth Based on Neuron Network

提出了利用神经网络技术,对离散的、分布不规则的海图水深数据进行插值处理得到连续水深数值的方法,研究了应用Levenberg Marquardt反向学习网络进行函数逼近和径向基函数网络进行函数插值二种算法,用函数曲面和实际水域数值对算法进行了测试和误差分析,并对实际应用中区域分块、规一化和等深线数据处理等问题作了说明。

Charted depth data are usually discrete irregularly. Based on the neuron network technique, interpolation methods of charted depth are suggested. Two algorithms based on LevenbergMarquardt backpropaganda and radialbasis function networks are investigated respectively. By using hyperbolic parabolid and the data of typical chart area the effectiveness of the algorithms is tested and error analysis carried out. Special process in practical applications such as division of area, normalization and depth contour data processing is also illustrated. 