基于RBF神经网络的水深遥感研究  

Remote sensing of water depth based on RBF neural network

作　　者：邓正栋[1] 叶欣 关洪军[1] 于德浩[3] 

机构地区：[1]解放军理工大学国防工程学院 [2]解放军63796部队 [3]沈阳军区司令部工程科研设计所

出　　处：《解放军理工大学学报（自然科学版）》2013年第1期101-106,共6页Journal of PLA University of Science and Technology(Natural Science Edition)

基　　金：国家863计划资助项目(2012AA062601)

摘　　要：为提高水深遥感反演的精度,以Landsat TM1～4波段为数据源,利用已知的水深数据作为训练样本,建立RBF神经网络模型对岱海水深进行反演试验。利用实测的水深数据检验RBF神经网络模型的反演精度,并与传统反演模型和BP神经网络模型进行对比。结果表明,RBF神经网络模型反演的水深与实测水深的决定系数为0.90,平均绝对误差为1.09m,均方根误差为1.45m,反演效果和精度明显好于传统反演模型;与BP神经网络模型相比精度也有提高,而且RBF神经网络模型的参数大多通过训练学习得到,应用更为便捷,在干旱内陆的咸水型湖泊水深遥感反演中有一定的应用价值。To improve the precision of remote sensing of water depth, the RBF neural network model was built to retrieve water depth from Landsat TM1-4 bands in Daihai Lake, which was trained by the known water depth data. Traditional retrieval models and BP neural network model were also built to compare with the RBF neural network model, and theie retrieving precision was tested by ground measured water depth. The results show that the R2, MAE and RMSE of RBF neural network model between the re- trieved and the measured water depth reach up to 0.89, 1.09m and 1.45m respectively, which is much better than the traditional models. Compared with BP neural network model, the RBF neural network model performs better too. Most of the RBF neural network model＇s parameters can be trained by learn- ing, which means it could be applied more conveniently to retrieving water depth by remote sensing in arid inland salty lake.

关 键 词：径向基神经网络 水深遥感 岱海 

分 类 号：S127[农业科学—农业基础科学]