【摘要】 神经网络BP方法应用于DEM内插,可以不需要传统的内插拟合函数。本文构建了一种"4X"BP神经网络结构,以地面点的平面坐标(X,X,Y,Y)作为网络输入层的节点,以高程H作为输出层的节点。用这种方法分析一个工程实例,选取隐含层节点数N为15,构建4×15×1的神经网络结构,结果得到DEM内插中误差为±0.45m,而传统的平面插值法的中误差为±0.53 m。实例证明,神经网络BP算法的效果非常好,值得在工程中推广应用。更多还原

【Abstract】 When BP neural network algorithm is applied in DEM Interpolation,traditional interpolation functions are useless.A kind of BP neural network structure model called "4X" is constructed in the paper.The horizontal coordinates(X,X,Y,Y) of a point are regarded as the input of the network,and its elevation H is as the output.A project is analyzed by this method as an example.A 4×15×1 neural network structure is constructed when there are 15 hidden level nodes.The mean square error in interpolation is ±0.45m when BP neural network method is used in this project.At the same time,it is ±0.53m when traditional plane interpolation method is used.It is proved that BP neural network method is preponderant obviously and can be popularized in other projects.更多还原