025-83701931 设为首页 加入收藏
文章摘要
粒子群算法优化SVR的天德湖总氮遥感反演
Remote Sensing Retrieval of Total Nitrogen in Tiande Lake Using SVR Optimized by Particle Swarm
  
DOI:
中文关键词: 支持向量机回归  粒子群优化算法  总氮  遥感反演  高光谱数据  天德湖
英文关键词: Support vector machine regression  Particle swarm optimization  Total nitrogen  Remote sensing retrieval  Hyperspectral data  Tiande Lake
基金项目:国家自然科学联合基金资助项目(U1704125)
作者单位
李爱民 郑州大学地球科学与技术学院 
许有成 郑州大学水利科学与工程学院 
王海隆 郑州大学水利科学与工程学院 
闫翔宇 郑州大学地球科学与技术学院 
康轩 郑州大学地球科学与技术学院 
摘要点击次数: 244
全文下载次数: 84
中文摘要:
      以郑州天德湖为研究水域,利用实测的水质总氮数据和光谱反射率数据,获取反演总氮浓度的最佳波段组合,以此建立支持向量机回归(SVR)模型,并利用粒子群优化算法(PSO)优化SVR模型,建立PSO-SVR模型,再利用珠海一号高光谱(OHS)数据验证该模型的精度和适用性。结果表明:PSO-SVR模型的决定系数(R2)为0.923 6,均方根误差(RMSE)为0.103 3mg/L,平均相对误差(MRE)为3.656%;SVR模型的R2、RMSE和MRE分别为0.808 0、0.203 2mg/L和6.583%,PSO-SVR反演结果优于SVR模型。利用粒子群算法优化SVR模型,有助于提高天德湖总氮浓度反演精度,即利用实测光谱数据与卫星影像数据结合来反演总氮浓度可行。
英文摘要:
      Taking Tiande Lake in Zhengzhou as the research area, the optimal band combination for retrieving total nitrogen concentration was obtained by using the measured total nitrogen data and spectral reflectance data, and a support vector machine regression(SVR) model was established. Particle swarm optimization(PSO) was used to optimize SVR model, and PSO-SVR model was created, and the accuracy and applicability of the model were verified by using Zhuhai 1 hyperspectral(OHS) data. The results showed that the determination coefficient(R2) of the PSO SVR model was 0.923 6, the root mean square error(RMSE) was 0.103 3 mg/L, and the average relative error(MRE) was 3.656%. The R2, RMSE and MRE of SVR model were 0.808 0, 0.203 2 mg/L and 6.583%, respectively. The results of PSO-SVR inversion were better than those of SVR model. The optimization of SVR model by particle swarm was helpful to improve the inversion accuracy of total nitrogen concentration in Tiande Lake, that is, it was feasible to invert total nitrogen concentration by combining measured spectral data with satellite image data.
查看全文   查看/发表评论  下载PDF阅读器
关闭