基于探地雷达功率谱属性参数的土壤含水状态预测

Soil Moisture State Prediction Based on Ground Penetrating Radar Power Spectrum Properties Parameters

研究准确快速探测土壤含水率的方法,对实现农业生产灌溉、生态修复等具有重要的现实意义。通过物理实验获得不同土壤含水率的探地雷达信号,以AR功率谱为理论模型计算雷达信号功率谱并提取功率谱属性参数,利用互相关方法优化选择功率谱属性参数,将功率谱属性参数作为输入样本使用BP神经网络进行土壤含水状态预测。实验结果表明:基于探地雷达功率谱属性参数与BP神经网络相结合的土壤含水状态预测方法对土壤含水状态识别的准确率为96.3%,其土壤含水率反演结果与实际含水率的平均绝对误差为1.2%,均方根误差RMSE为0.015;在野外实测中利用该方法对16组土壤富水性识别出现1次错误,对土壤含水率反演的绝对误差和相对误差分别在3%、10%以内。该方法对土壤含水检测具有较高精度,预测误差较小,对土壤含水率快速探测具有一定参考意义。

Researching accurate and rapid methods for detecting soil moisture content holds significant practical importance for agricultural irrigation,ecological restoration,and other applications.In this study,we conducted physical experiments to acquire ground-penetrating radar(GPR)signals at different soil moisture levels.The power spectrum of the radar signals was calculated using the AR power spectrum model,and power spectrum attribute parameters were extracted.Cross-correlation was employed to optimize the selection of power spectrum attribute parameters.These parameters were then used as input samples for training a backpropagation(BP)neural network to predict soil moisture states.The experimental results demonstrated that the combination of GPR power spectrum attribute parameters and the BP neural network yielded an accuracy of 96.3%for identifying soil moisture states.The average absolute error between the predicted and actual soil moisture content was 1.2%,with a root mean square error(RMSE)of 0.015.In field measurements,utilizing this method,only one error occurred in identifying the water-rich nature of 16 soil samples,and the absolute and relative errors in soil moisture content inversion were within 3%and 10%,respectively.This method exhibits high precision in soil moisture detection,with small prediction errors,thus providing valuable insights for rapid soil moisture assessment.