Study On Soil Moisture Measurement Using Ground Penetrating Radar Based On Support Vector Machine

In recent years, the application of ground penetrating radar(GPR) on soil properties becamemore and more. However, the previous studies were focused on the relationship between the dielectric constant rate based on GPR and soil water content, and the related studies were mostly carried out in the laboratory. Therefore, the Shunyi District Zhaofeng industrial district was selected as an experimental areato conduct the application of GPR on the measurement of soil water content in this paper, and three 100 m×100 m plots were designed. Five ground penetrating radar survey lines were layouted in each plot, and 5 sample points were uniformly in each survey line and the soil water content and soil particle size distribution were determined. TheRANDA7 and MATLAB2012 b were used to conduct image processing, data analysis and extraction. Then, the data obtained from GPR survey and laboratory measurement would be import into the support vector machine to establish the classification model. The study found that:i) The method of cross correlation analysis was used to remove the interference of radar image information, which was helpful to study the soil moisture condition under the natural condition, and improved the relationship between the data precision and the subsequent classification.ii) According to waveform location and cross correlation analysis, using MATLAB2012 b to carry out the amplitude data fixed point, compared to the direct amplitude extraction, its accuracy was greatly improved;iii) Soil water content, amplitude, and cross correlation analysis results could be used to build models in LIBSVM. The results showed that the accuracy rate was only 33.3%using the default RBF parameterin the training process. Though parameter selection, using parameters(-c,1,-h,-g,0.2) the complexity of the boundary function could bestrengthened and the accuracy rate could increase to 78.3%;iv) The results showed that the accuracy of soil water content in the surface layer was higher than that in the bottom layer. If the soil water content at the depth of 60-80 cm were removed in the test data, the accuracy rate would increase to 81.5%.v) Support vector machine model could predict the relationship between ground radar amplitude data and soil moisture content, and though regression analysis, its MAE and RMSE values were 0.0486 and 0.0664, respectively.