利用高光谱数据估算太湖水体CDOM浓度的神经网络模型  

Remote Sensing Retrieval of CDOM Concentration in Lake Taihu with Hyper-spectral Data and Neural Network Model

作　　者：孙德勇[1] 李云梅[1] 王桥[1] 乐成峰[1] 

机构地区：[1]南京师范大学虚拟地理环境教育部重点实验室,南京市文苑路1号210046

出　　处：《武汉大学学报（信息科学版）》2009年第7期851-855,共5页Geomatics and Information Science of Wuhan University

基　　金：国家自然科学基金资助项目(40571110;40701136);国家科技支撑计划资助项目(2008BAC34B05);江苏省普通高校研究生科研创新计划资助项目(CX08B_015Z);南京师范大学研究生优秀学位论文培育计划资助项目(181200000220)

摘　　要：利用2007-11-08~2007-11-21 14 d时间对太湖74个样点进行了水质取样分析和波谱实测。在分析水体固有光学特性的基础上,确定了CDOM浓度遥感反射比的敏感波段,建立了湖泊水体CDOM浓度反演的神经网络模型。结果表明,隐含层节点数为10的神经网络模型在各神经网络模型中效果最佳。利用验证样本对神经网络模型和其他算法模型进行误差分析,发现神经网络模型更适用于湖泊水体。CDOM concentration is an important parameter of water environment. In order to accurately retrieve CDOM concentration in lakes, a field experiment including water quality analysis and spectrum measurements was carried out on 74 stations of Lake Taihu during 14 days from Nov. 8, 2007 to Nov. 21, 2007. Based on the analysis of water inherent optical properties, sensitive wavebands were selected and neutral network models of CDOM concentration retrieval were established. The results show that a model with 10 nerve cells in hidden layer performs best, whose R is 0. 887 and RMSE is 0. 156. Meanwhile, the predicative errors of neutral network model and previous algorithm models were analyzed through validation samples. Average relative error of the former is （12.8±29.9）%, while others are very large, which indicates that neutral network model is more suitable for CDOM concentration retrieval in Lakes than other algorithm models.

关 键 词：遥感反射比 CDOM 神经网络模型 太湖 

分 类 号：P237.9[天文地球—摄影测量与遥感]