基于BP神经网络GPR反演滨海盐渍土含盐量模型构建
2018-08-14赵学伟王萍李新举刘宁
赵学伟 王萍 李新举 刘宁
摘要:滨海盐渍土水盐含量较高,水盐运移规律显著,利用探地雷达(GPR)反演土壤含盐量具有重要意义。针对目前GPR反演土壤含盐量模型多为内陆盐渍土单一影响因素模型现状,本研究采用BP神经网络方法,探究GPR信号的土壤介电常数、土壤层次振幅比与土壤含盐量之间的非线性关系,并以Matlab 为平台,采用自编程序对实验数据进行网络学习和仿真,构建了多重GPR信号反演的土壤含盐量模型。模型输出准确率达到86.53%,表明该模型可以预测滨海盐渍土含盐量。
关键词:滨海盐渍土; GPR; 土壤介电常数; 分层振幅比; 含盐量;BP神经网络
中图分类号:S126:S156.4+2文献标识号:A文章编号:1001-4942(2018)05-0152-04
Abstract The soil water and salinity in coastal area is high and changing rapidly, so it is significant to using GPR to invert soil salinity. Different from the most present models of inland saline soil salinity inverted by GPR based on single influencing factor, we consctucted the soil salinity model based on multiple GPR signals using the BP neural network. The nonlinear relationships were studied between the soil dielectric constant and soil layer amplitude ratio of GPR signals and the soil salinity. Then, the BP neural network is applied to train and simulate the experimental data with Matlab platform. The accuracy of model output is up to 86.53%, which indicates that the model could predict the soil salinity of the saline soil.
Keywords Coastal saline soil; GPR; Soil dielectric constant; Stratified amplitude ratio; Soil salinity; BP neural network
黃河三角洲地区多为滨海盐渍土,土壤水盐含量高[1],时空变异性强,运移过程复杂[2]。调控土壤盐分是滨海盐渍土改良和农业利用的最重要内容,而土壤盐分的定量监测是水盐动态运移研究的前提[3]。传统的盐渍土含盐量测定方法有TDR传感器法[4]、残渣烘干法[5]、电导率法[6],其中,传感器法可实时测定土壤盐分,测定结果精确度高,但需埋设较多探头,观测与维护成本高;残渣烘干法、电导率法需开挖土壤剖面,对土壤破坏程度大。探地雷达(GPR)可沿测线解析土壤剖面结构,反演分层土壤盐分含量,耗时短、费工少、土壤挖损小且精度高、测量深,更适宜于中尺度土壤水盐调查[7]。
但目前探地雷达对土壤的调查大多集中在土壤层次和厚度以及水分的测定等方面[8],对盐渍土,尤其是滨海盐渍土水盐变化的应用研究不多。……
