基于PSO优选参数的SVR水质参数遥感反演模型  

A Model for Water Quality Remote Retrieval Based on Support Vector Regression with Parameters Optimized by Particle Swarm Optimization

作　　者：何同弟[1] 李见为[1] 黄鸿[1] 

机构地区：[1]重庆大学光电技术及系统教育部重点实验室,重庆400030

出　　处：《信息与控制》2011年第4期532-536,共5页Information and Control

基　　金：国家自然科学基金资助项目(40671133);重庆市科技攻关重点资助项目(CSTC2009AB2231)

摘　　要：为进一步提高多光谱图像水质反演的精度,提出了一种基于PSO优选参数的SVR水质参数遥感反演模型.该模型利用高分辨率多光谱遥感SPOT-5数据和水质实地监测数据,采用交叉验证CV(cross validation)估计模型推广误差并使用PSO优选SVR模型参数,实现了模型参数的自动全局优选,在训练好的SVR模型基础之上对水质进行反演.以渭河陕西段为例进行实证研究,实验结果表明,本文提出的水质反演模型较常规的线性回归模型有更高的反演精度,为内陆河流环境遥感监测提供了一种新方法.In order to improve water quality retrieval accuracy of multi-spectral image,a model is put forward for water quality remote retrieval based on support vector regression（SVR） with parameters optimized by particle swarm optimization（PSO）.Based on high-resolution multi-spectral remote SPOT-5 data and the water quality field data,The model uses CV（cross validation）to estimate the generalization error and adopts PSO to optimize parameters of SVR model.Thus,automatic global optimization of model parameters is achieved,and the water quality is retrieved by the trained SVR.The proposed model is applied to the water quality retrievals of Weihe River in Shaanxi province.The experiment result shows that the developed model is more accurate than the routine linear regression model.It provides a new approach for remote sensing and monitoring of inland river environments.

关 键 词：高分辨遥感影像 粒子群优化算法 支持向量回归 参数优选 水质反演 

分 类 号：X832[环境科学与工程—环境工程] TP79[自动化与计算机技术—检测技术与自动化装置]