遥感与神经网络相结合的潮滩地形模拟方法

Modeling Methods for Tidal Flat Terrain Based on Remote Sensing and Artificial Neural Network

潮滩地形资料的获取是开发利用潮间带资源的第一步。以纳潮盆地为实验对象,从遥感影像面状信息入手,提取并组合影像中的信息,采用遥感与BP人工神经网络相结合的方法,构建遥感光谱信息、地貌特征与潮滩高程信息之间的关系模型。结果表明:在遥感光谱信息基础上,引入纳潮盆地纵、横剖面地貌特征因子的神经网络模拟效果更好;将潮滩滩面与潮水沟分别进行网络建模,生成地形,平均绝对误差达0.299m,这说明,神经网络在模拟高程起伏较大的区域时精度较低,适当降低神经网络输入数据的复杂度有利于改善网络的模拟精度。

To get the data of tidal flat terrain is the first step to develop and utilize the resources of the intertidal zone.This paper collected the information from the image,performed the experiments which took the tidal basin as the experimental object using Artificial Neural Network and eventually built the relational model among the spectral information,the geomorphology factor and the elevation.The results show that the neural network based on the spectral information and the geomorphology factor in the longitudinal and cross section of tidal basin is better.Ranging from-2.17to-1.0meters of height,MAE of the network model that based on the spectral information and the geomorphology factor in the longitudinal and cross section of tidal basin is 0.656 meters,while ranging from-1.0-1.32 meters,the MAE is 0.283 meters.To build models of the beach face and the tidal creek separately is effective,and the MAE is 0.299 meters.These data indicate that the neural network has lower precision where the terrain fluctuates strongly.The complexity of the input data of the neural network can be reduced to improve the accuracy of the network.In this paper,the final model has been applied to Dongsha,and the MAE is 0.25 meters,which is good.