基于探地雷达的树木根系相对介电常数估算

Estimation of relative permittivity of tree root system based on ground penetrating radar

【目的】本研究提出一种基于雷达波信号和树木根系参数的估算方法,以实现树木根系相对介电常数的定量估算。【方法】首先,仿真模拟雷达波在不同半径、不同相对介电常数的树木根系下的传播路径,并通过正演分析获得探地雷达图像中双曲线顶点处的A-scan曲线;然后,提取A-scan曲线中与根系相对介电常数关联的目标振幅参数ΔF;最后,结合土壤相对介电常数、根系半径、根系深度,建立相对介电常数估算的数据集。分别基于偏最小二乘回归(PLSR)模型、反向传播(BP)神经网络模型和粒子群优化反向传播(PSO-BP)神经网络建立估算模型,并对比分析这3种模型的估算精度。【结果】(1)在仿真实验中,PSO-BP神经网络估算模型的均方根误差、平均绝对误差分别为0.701、0.255,R2为0.990,各指标均优于PLSR和BP神经网络估算模型。(2)在实地预埋实验中,PSO-BP神经网络估算模型的估算精度均优于PLSR和BP神经网络估算模型,其最大绝对误差和整体平均相对误差分别为3.16和10.88%。【结论】利用本研究提取的目标振幅参数ΔF、土壤相对介电常数、根系半径和根系深度建立的数据集,结合PSO-BP神经网络估算模型,能够实现对树木根系相对介电常数的准确估算。这对于评估树木根系的生长和健康状况具有重要意义。

[Objective]In order to achieve the quantitative estimation of the relative permittivity of tree root system,an estimation method based on radar wave signals and tree root system parameters was proposed.[Method]Firstly,the propagation paths of radar waves in the root system of trees with different radii and relative permittivity were simulated,and the A-scan curves at the hyperbolic vertices in the groundpenetrating radar images were obtained through orthogonal analysis.Then,the target amplitude parameterΔF,which was related to the relative permittivity of root system was extracted from A-scan curves.Finally,the dataset for relative permittivity estimation was established by combining the relative permittivity of soil,the radius of root system,and depth of root system.The estimation models were established based on partial least squares regression(PLSR),back propagation(BP)neural network,and particle swarm optimizationback propagation(PSO-BP)neural network,respectively,and the estimation accuracies of the three models were compared and analyzed.[Result](1)In the simulation experiment,the root mean square error and average absolute error of PSO-BP neural network estimation model were 0.701,0.255,and the R2 was 0.990,and all indexes were better than PLSR and BP neural network estimation model.(2)In the field preembedding experiment,the estimation accuracies of PSO-BP neural network estimation model were all better than those of PLSR and BP neural network estimation models,with a maximum absolute error of 3.16 and the whole average relative error of 10.88%.[Conclusion]Using the dataset established by target amplitude parameterΔF,soil relative permittivity,root radius and root depth extracted in this study,combined with PSO-BP neural network estimation model,an accurate estimation of the relative permittivity of root system of trees can be achieved,which is of great significance for assessing the growth status and health of tree root system.