基于RBF模型的太湖北部叶绿素a浓度定量遥感反演  

Quantitative retrieval of chlorophyll-a concentration in northern part of Lake Taihu based on RBF model

作　　者：曹红业 龚涛[1] 袁成忠[1] 蒋进钦 CAO Hongye GONG Tao YUAN Chengzhong JIANG Jinqin(Faculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, Chengdu 611756, China)

机构地区：[1]西南交通大学地球科学与环境工程学院,成都611756

出　　处：《环境工程学报》2016年第11期6499-6504,共6页Chinese Journal of Environmental Engineering

基　　金：北京市自然科学基金资助项目(8142014)

摘　　要：以大型内陆浅水湖泊一太湖为例,采用RBF(Radial Basis Function)神经网络建立该研究区域的叶绿素a浓度与同步影像数据的反演模型,较分析现今应用最广泛使用的BP(Back Propagation)神经网络模型,并通过对模型的验证、稳定性和鲁棒性分析评价了两种模型的泛化能力。结果表明,常规的BP神经网络模型收敛速度慢,极容易陷入局部最优解,而RBF比BP模型有更加优异的函数逼近、分类和模式识别能力,对反演叶绿素a浓度具有很强的泛化能力。Taihu Lake,a large shallow inland lake,was taken as an example in this study. A quantitative inversion model was established between chlorophyll-a concentration and synchronous remote sensing images by using an RBF neural network. A comparison of this model with a BP neural network model,which has been extensively used in recent works,enabled a comparative analysis of the performance of BP and RBF in terms of validation,stability,and robustness. The results showed that the BF neural network has several shortcomings,such as slow convergence and easy fall into the local minima. In contrast,the RBF neural network has many advantages; for instance,it has stronger abilities of function approximation,classification,and pattern recognition than the BP network. The RBF model provides a good generalization from the perspective of inversing the chlorophylla concentration.

关 键 词：RBF神经网络 水质监测 叶绿素A 太湖 

分 类 号：X87[环境科学与工程—环境工程]