Research On The Identification And Modeling Of Coastal Saline Soil Profile Characteristics Based On High-frequency Ground Penetrating Rada

The coastal saline soil in the Yellow River Delta is an important reserve land resource in China,and its improvement and utilization are of great significance to ensure national food security and promote the sustainable development of regional agriculture.The soil profile configuration affects the water-salt migration and soil physicochemical properties of coastal saline soil.The study of soil profile characteristics is the basic work for the improvement and utilization of saline soil.This paper starts with typical plots in the coastal saline soil area of the Yellow River Delta,uses the ground penetrating radar with a center frequency of 250 MHz to measure the profile method,the common center point method and the grid method,and uses U-Net,edge detection and other technical means based on the python platform.B-scan image analysis and processing,automatic completion of layered signal extraction of coastal saline soil profile,estimation of electromagnetic wave propagation speed and soil layer thickness,soil 2D slice and 3D model construction,etc.,revealing different salinization of coastal saline soil.The cross-sectional layering characteristics of GPR images of typical plots at low temperature will lay a foundation for quantitative inversion of soil properties and real-time monitoring of soil quality in the future.The main conclusions of this study are as follows:(1)In this paper,a technical system for automated processing and analysis of GPR signals is constructed.The system mainly consists of eight steps: channel synthesis,data annotation,image enhancement pre-processing,improved U-Net method processing,edge detection,soil layer identification and electromagnetic wave velocity estimation,and performs well in soil profile feature identification and soil layer thickness estimation.(2)The improved U-Net method is better than the prototype U-Net method in terms of the image,the AUC(Area Under the operation characteristic Curve)and the accuracy rate,and it is more effective in extracting the reflected wave signal of the soil layer and can distinguish the soil layer.It has stronger force and higher accuracy and can effectively suppress the over-fitting phenomenon during the 200-generation training process;compared with the processing results of the Line View software that comes with GPR,the improved U-Net method can identify the signal band of the soil layer and at the same time.To achieve a good filtering effect,compared with the Line View method,the number of soil layer reflected wave signals extracted is similar to the location,and more soil layer reflected wave signals can be extracted by overcoming the disadvantage of signal discontinuity.(3)The size of the Gaussian convolution kernel is 3 × 3 when the soil layer is extracted,and the hysteresis threshold is the minimum value of 5 and the maximum value of 25.The plots with different soil salinity have different soil edge characteristics: the lower the soil salinity content,the more continuous the signal can still be detected when the antenna spacing is farther.Wheat and corn can still detect obvious soil layer signals at an antenna distance of 3.9m,while there are many narrow signal bands at the edge of the soil layer in the bare ground,yellow greens,and reed plots.The higher the surface soil salinity,the stronger the attenuation of the air and ground wave signals,and the fewer and discontinuous velocity extraction points.The lower the soil salinity content of the plot,the more continuous the air wave,ground wave and reflected wave signal,the smaller the velocity fluctuation,and the more significant the lower soil stratification signal.(4)The estimation results of soil layer thickness in this paper are consistent with the actual situation of soil profile.In the reed plot,thatch plot and cotton plot,the estimation results are equivalent to or better than those processed by the Line View method,and in the bare slab plot and corn plot,the estimation results can accurately identify the electromagnetic wave signals of deeper soil layers.(5)This set of soil layer identification technology system can be used for three-dimensional modeling.In the slice model,the improved U-Net method has a certain processing ability for the FO method measurement data of deeper soil,and the soil layer below 50 ns can be compared with the FO method.The higher the salt content of the plot,the less information of the soil layer edge is extracted;the projection of the soil model in the X and Y directions can more clearly show the soil layer boundary of the soil profile.