| In recent years,ground-penetrating radar(GPR)has been widely used for the accurate,non-destructive,and efficient detection of spatiotemporal soil moisture distribution on the Earth’s surface.Existing methods for GPR-based soil moisture detection primarily rely on the time-domain properties of signals.These methods estimate the soil dielectric constant through time-domain signal analysis and establish corresponding empirical models to estimate soil moisture content.They are advantageous due to their simple principles and easy implementation.However,the accuracy of soil moisture inversion using these methods is influenced by the obtained dielectric constant and the related empirical models,leading to potential secondary transmission errors.Electromagnetic waves experience attenuation and absorption during their propagation in subsurface media,resulting in signal energy decay and frequency shift phenomena.Power spectrum estimation methods describe the power density distribution of signals at different frequencies.By establishing a relationship model between the power spectrum and soil moisture content,it is possible to achieve soil moisture inversion.This approach offers intuitive soil moisture detection and higher inversion accuracy.Therefore,it is of significant importance to study a soil moisture detection method based on power spectrum estimation using ground-penetrating radar,aiming to achieve precise and rapid detection of soil moisture content,which has implications in environmental ecology,agricultural production,land reclamation,and other fields.This paper employs research methods such as theoretical analysis,numerical simulation,physical experiments,and correlation analysis to investigate the applicability of power spectrum estimation methods for ground-penetrating radar signals.Suitable power spectrum estimation methods for ground-penetrating radar signals are selected through the analysis.Subsequently,numerical simulation and physical experiments are conducted to explore the response relationship between ground-penetrating radar power spectrum attributes and soil moisture content.Attribute optimization methods are utilized to select power spectrum parameters that have high correlation with soil moisture content and are independent of each other for the inversion of soil moisture content.Finally,a soil moisture prediction method is proposed,which combines the power spectrum attribute parameters of ground-penetrating radar with a backpropagation neural network(BPNN).The main achievements are as follows:(1)Classical and modern power spectrum estimation methods are employed to calculate the power spectrum of ground-penetrating radar signals,and their performance is analyzed.The results demonstrate that the radar signal power spectrum obtained using the improved autoregressive-covariance algorithm(AR-COV)yields the best results.(2)The relationship between soil moisture content and ground-penetrating radar power spectrum is investigated through numerical simulation and physical experiments.It is found that an increase in soil moisture content causes a shift of the power spectrum energy band towards lower frequencies,an increase in the energy proportion in the low-frequency band,a decrease in the energy proportion in the high-frequency band,and an enhanced aggregation of energy distribution.The extracted attribute parameters from the power spectrum are correlated with soil moisture content.The results show that the attribute parameters representing frequency exhibit a linear negative correlation with moisture content,while the attribute parameters representing energy exhibit a logarithmic negative correlation.Furthermore,the attribute parameters representing the complexity of power spectrum distribution in frequency bands exhibit a linear negative correlation with moisture content.(3)An attribute optimization algorithm was utilized to extract power spectrum attribute parameters for 400 MHz and 900 MHz antennas,including main frequency,central frequency,centroid frequency,edge frequency,band energy,frequency standard deviation,and energy proportion within different frequency bands.Based on these parameters,a method combining the power spectrum attribute parameters of ground-penetrating radar with a backpropagation neural network(BPNN)was proposed for the identification and prediction of soil moisture state.Experimental results demonstrate that the accuracy rate of soil moisture state identification using this method exceeds 90%.For soil moisture prediction,the average absolute errors between the predicted and actual moisture content for the 400 MHz and 900 MHz antennas are 1.25% and 1.21%,respectively,with root mean square errors(RMSE)of 0.015 and 0.016.(4)In field measurements,the soil moisture content was inverted using the correlation between power spectrum attribute parameters and soil moisture content.The relative errors between the 400 MHz and 900 MHz antennas and the actual soil moisture content in both exploration sites and open fields were within a mean range of 21%.The method combining power spectrum attribute parameters with BPNN for soil moisture state inversion resulted in one misjudgment out of 16 soil moisture states in terms of soil water abundance determination.Regarding soil moisture prediction,the relative errors between the 400 MHz and 900 MHz antennas and the actual moisture content in both exploration sites and open fields were within a range of 11%. |
