基于无人机多光谱协同电磁感应技术的棉田土壤盐渍化监测研究

UAV-based multispectral synergistic electromagnetic sensing technology for salinization of cotton fields

利用无人机多光谱影像结合电磁感应技术准确估测新疆棉田土壤盐分,对棉田土壤盐渍化的防治及棉花产量的提高具有重要的意义。以新疆南疆阿拉尔垦区棉田为研究对象,将无人机多光谱波段反射率数据计算的植被指数、EM38-MK2大地电导率仪获得的土壤表观电导率数据与土壤盐分实测数据相结合,利用支持向量机(SVM)、随机森林(RF)、BP神经网络(BPNN)3种建模方法构建不同土壤盐分估测模型。结果表明,植被指数与土壤表观电导率数据融合模型对土壤盐分估测效果最好,单一土壤表观电导率数据模型或光谱指数模型估测效果较差。在棉花的花铃期,基于光谱指数与土壤表观电导率数据融合的RF模型是棉田土壤盐分估测的最佳模型,建模集R^(2)、RMSE、RPD分别为0.91、0.34、2.73,验证集R^(2)、RMSE、RPD分别为0.88、0.24、3.09;模型能够实现对棉花花铃期土壤盐分的高精度预测。

The use of UAV multispectral imagery combined with electromagnetic induction technology to accurately estimate soil salinity in cotton fields in Xinjiang is important for the control of soil salinity in cotton fields and the improvement of cotton yields.Taking cotton fields in the Alar Reclamation Area of South Xinjiang as the research object,the vegetation index calculated from the UAV multispectral band reflectance data and the soil apparent conductivity data obtained from EM38-MK2 geodetic conductivity meter were combined with the measured soil salinity data to construct soil salinity estimation models using three modelling methods:support vector machine(SVM),random forest(RF)and BP neural network(BPNN).The results showed that the fusion model of vegetation index and soil apparent conductivity is the best to estimate soil salinity.The single soil apparent conductivity data model or the spectral index model has a poor estimation effect.During the boll stage of cotton,the RF model based on the fusion of spectral index and soil apparent conductivity data is the best model for estimating soil salinity in cotton field,with modeling sets R^(2),RMSE and RPD of 0.91,0.34 and 2.73,respectively,and verification sets R^(2),RMSE and RPD of 0.88,0.24 and 3.09,respectively.The model is capable of predicting soil salinity at the boll stage of cotton with high accuracy.