基于反演因子筛选下的BP神经网络模型在水体叶绿素a含量反演中的可行性研究——以太湖为例  

Feasibility Study of BP Neural Network Model in Water Chlorophyll-a Content Retrieval Based on Retrieval Factors Selection:a Case Study of Taihu Lake

作　　者：朱婧婧 王庆[1] 战超[1] 刘亚龙[2] 董程 王红艳[1] ZHU Jingjing1 , WANG Qing1, ZHAN Chao1, LIU Yalong2, DONG Cheng1 , WANG Hongyan1(1. Coastal Institute, Ludong University, Yantai 264039, China ; 2. Yantai Marine Environment Monitoring Center Station, State Oceanic Administration, Yantai 264006, China)

机构地区：[1]鲁东大学海岸研究所,山东烟台264039 [2]国家海洋局烟台海洋环境监测中心站,山东烟台264006

出　　处：《鲁东大学学报（自然科学版）》2018年第3期259-266,共8页Journal of Ludong University:Natural Science Edition

基　　金：国家重点研发计划(2017YFC0505902);国家自然科学基金(41471005;41271016;U1706220)

摘　　要：本文利用HJ-1B卫星CCD遥感影像光谱反射率数据与实测的叶绿素a含量,建立了BP神经网络模型,实现星上点与地面实测点建模,从而得出用于反演水体叶绿素a含量的模型.在模型建立之前,分析反射率的不同波段组合与实测叶绿素含量的相关系数,依此选取不同的波段组合作为模型的自变量(反演因子).经过分析筛选,将相关系数大于0.5的35个反演因子作为模型的输入数据,建模结果显示:训练结果 R=0.91,模型测试结果 R=0.92,模型精度检验的APE为11.22%,相关系数为95%,RMSE为9,变异系数(PRMSE)为18%.由此可见,BP神经网络模型的建立结果较好,可用于水体参数的反演.In this paper,the BP neural network model was established by using the spectral reflectance data of HJ-1 B satellite CCD remote sensing image and the measured chlorophyll-a content to realize the modeling between the satellite points and the measured points on the ground,and the model for the inversion of the content of chlorophyll-a in the water body was obtained. Before the model was established,the correlation coefficient between the different wave band combinations and measured collorophyll-a content was analyzed. According to this,different band combinations were selected as the independent variables of the model（ inversion factor）. After analysis and screening,35 inversion factors of the correlation coefficient greater than 0. 5 were used as input data of the model type. The modeling results show that the trained result R is 0. 91,the model test result R 0. 92,the model accuracy test APE 11. 22%,the correlation coefficient 95%,the RMSE 9,and the coefficient of variation（ PRMSE） 18%. It can be seen that the BP neural network model has good results and can be applied to the inversion of water body parameters.

关 键 词：BP神经网络模型 叶绿素A 相关分析 定量遥感 

分 类 号：K903[历史地理—人文地理学]