A method for inverting soil moisture content of windbreak and sand-fixing forests based on ground penetrating radar

Abstract

The present application provides a method for inverting the soil moisture content of a windbreak and sand-fixing forest based on ground-penetrating radar, which belongs to the technical field of soil environment monitoring. The method comprises: determining in advance a detection area with a large number of coarse roots within a sample plot by technical means, and using ground-penetrating radar equipment to perform detailed detection in the detection area with a large number of coarse roots to invert the soil moisture content of the windbreak and sand-fixing forest. The combined use of an unmanned vehicle and a ground-penetrating radar equipment can greatly improve the detection efficiency. The present application further proposes a technical solution for using a drone to predict the number of coarse roots.

Description

Technical Field

[0001] The present application relates to the technical field of soil environment monitoring, and in particular to a method for inverting the soil moisture content of windbreak and sand fixation forests based on ground penetrating radar. Background Art

[0002] Water is the most important factor affecting plant growth and development in arid and semi-arid regions. Soil moisture is a crucial factor influencing physiological processes such as tree growth, transpiration, and photosynthetic carbon assimilation. Therefore, accurately quantifying the changing characteristics of soil moisture in the rhizosphere of trees is crucial for understanding the mechanisms of tree growth and survival in arid and semi-arid regions.

[0003] Traditional methods for measuring soil moisture include oven drying, neutron detection, and TDR. These methods typically measure soil moisture at specific locations, making it difficult to reflect continuous spatial variations in soil moisture. These methods are time-consuming, labor-intensive, and destructive to the soil. Remote sensing-based soil moisture inversion has overcome the technical barriers to dynamic soil moisture monitoring at the regional scale, but it can only invert surface soil moisture and has low accuracy. Ground-penetrating radar (GPR), a non-destructive detection technology, can rapidly measure soil moisture at the stand scale. Compared to other soil moisture detection methods, GPR offers advantages such as spatial continuity, high resolution, deep detection depth, and repeatability. Allred et al. combined GPR with TDR to study the horizontal distribution of soil moisture in the sand layer of a Canary Island golf course in the United States. Cui Fan et al. proposed a method for determining the moisture content of sandy loam soils using power spectrum, based on GPR technology. Currently, most studies on GPR-derived soil moisture have been conducted on unvegetated land, and few reports have examined the spatial distribution of soil moisture in forests using GPR. In addition, my country's windbreak and sand-fixing forests are widely distributed, and it is very necessary to understand the soil moisture conditions of their growth environment in a timely manner. However, the current detection methods are very time-consuming and labor-intensive, and are not automated. Technical personnel are unable to grasp the soil moisture content of the growth environment of windbreak and sand-fixing forests in a timely manner, which is not conducive to the formulation of relevant policies and adjustments to technical means for maintaining windbreak and sand-fixing forests.

[0004] This application takes the Larix gmelinii plantation in Zhanggutai area as the research object and designs a new method for inverting the soil moisture content of windbreak and sand fixation forests based on ground penetrating radar. Through this method, it is easy to quickly construct a distribution model of forest stand soil moisture content based on the inversion of ground penetrating radar root point reflection wave velocity. Summary of the Invention

[0005] This application aims to address the aforementioned issues in the prior art by providing a method for inverting soil moisture content in windbreak and sand-fixing forests using ground-penetrating radar. This method improves the efficiency and accuracy of soil moisture detection in windbreak and sand-fixing forests, essentially automates the entire monitoring process, and comprehensively improves all aspects of soil moisture monitoring in windbreak and sand-fixing forests.

[0006] In order to achieve the above objectives, this application adopts the following technical solutions:

[0007] A method for inverting soil moisture content of windbreak and sand fixation forests based on ground penetrating radar, characterized in that:

[0008] (1) Set up a 10m×10m sample plot in the forest. Before measurement, a map of the test area must be generated by computer.

[0009] (2) Clear the surface obstacles such as weeds, dead branches, pine cones and stones within the measurement point grid on site to reduce the impact of surface cover on detection;

[0010] (3) The unmanned vehicle tows the ground penetrating radar device equipped with a GPS receiver along the survey line according to the pre-generated map of the detection area, dragging the antenna at a constant speed to collect data. The first survey line starts from the y-axis boundary and moves from the origin along the y-axis at intervals of 0.25 m. The interval between the survey lines is 0.25 m, and a total of 40 survey lines are used.

[0011] (4) The sample is divided into 100 square areas of 1 square meter. The square area of ​​1 square meter is represented by Aij, where i = 0, 1, 2, ..., 9; j = 0, 1, 2, ..., 9. The raw data obtained from the measurement of each survey line are only used to remove the background clutter of the profile image by the mean filtering method. Then, the root point hyperbola signal recognition algorithm is used to calculate the coarse root distribution map in the profile image in the yh plane, where h represents the depth. The coarse root distribution maps corresponding to the four survey lines at x = i, x = i + 0.25, x = i + 0.50, and x = i + 0.75 are projected and superimposed in the yh plane.

[0012] (5) Count the number of coarse root projections of the four survey lines in the cross-sectional image in the yh plane in each 1 square meter square area Aij, and select the square area Aij with the number of coarse root projections ranking in the top 25%;

[0013] (6) The unmanned vehicle will tow the ground penetrating radar equipment to re-measure the selected square area Aij with a line interval and step interval of 0.1m;

[0014] (7) Preprocessing the radar image data obtained in step (6);

[0015] (8) Using the hyperbola semi-automatic recognition algorithm, identify the complete and clear root points of the hyperbola signal in the radar image, and calculate the reflection wave velocity V of the root point based on the random Hough transform and iterative method soil ;

[0016] (9) The top of the root hyperbola signal corresponds to the upper surface position of the thick root. Combined with the root reflection wave velocity, the root depth h can be determined.根点 , the formula is as follows: 根点 =V soil ×t / 2, where t is the round-trip time of electromagnetic wave propagation; then, based on the horizontal position of the root point on each slice, its two-dimensional coordinates (x, y) can be determined, thereby obtaining the three-dimensional spatial coordinates (x, y, h) of the root point;

[0017] (10) The reflected wave velocity V soil Substitute into the formula: In the equation ( ), c ​​is the propagation speed of electromagnetic waves in a vacuum. The dielectric constant ε of the soil between the ground and the root point is calculated, and then the volumetric water content of the soil between the ground and each root point is calculated according to the Topp formula.

[0018] In some technical solutions, the GPS receiver of the ground-penetrating radar equipment can monitor the location information of the ground-penetrating radar equipment in real time. When the ground-penetrating radar equipment reaches a designated measuring point, it will send an instruction to the unmanned vehicle to stop moving and the ground-penetrating radar equipment will start working.

[0019] In some technical solutions, step (5) can also be: further divide the 100 square areas with an area of ​​1 square meter into 25 equally divided larger square areas, and each of the 25 larger areas obtained by the secondary division contains 4 square areas Aij of 1 square meter. The number of coarse root projections in the 4 square areas Aij in the larger square area is compared, and the square area Aij in each larger square area with the largest number of coarse root projections Aij will be selected.

[0020] Preferably, the selected secondary measurement areas Aij do not have interconnected areas. If the secondary measurement areas Aij do have interconnected areas, then the Aij with the second largest number of coarse root projections can be selected from the four square areas Aij within the larger square area as the secondary measurement area.

[0021] Preferably, the pre-processing step in step (7) includes: system temperature drift error correction, background removal and signal gain.

[0022] In some technical solutions, soil water storage PSWS is calculated by combining soil volumetric water content and root depth. gpr , the formula is as follows: PSWS gpr =ASM×h 根点 ×1000,h 根点 is the root point depth, 1000 is the conversion factor from m to mm, and ASM is the soil volumetric water content.

[0023] In some technical solutions, inverse distance weighted interpolation is performed on the spatial scattered data set of soil water storage to reconstruct the spatial distribution characteristics of soil water storage and obtain the two-dimensional distribution characteristics of soil water storage at different depths.

[0024] In some technical solutions, a drone equipped with a camera is used to capture crown images of trees at a certain location in a shelterbelt plot. The crown images provide information including leaf color, leaf density, leaf size, branch density, branch color, and branch size. The images are then combined with the number of root points obtained by a ground-penetrating radar device at that location for analysis using a convolutional neural network to obtain the corresponding crown characteristics of trees at locations with a larger number of root points.

[0025] In some technical solutions, drones take photos of trees in the sample plot and then use computer analysis to determine the coordinates of areas with more thick roots.

[0026] Preferably, the map of the detection area includes at least the coordinates of the trees in the forest to be measured and the coordinates of the measuring points of each measuring line.

[0027] This application has the following advantages:

[0028] (1) This application uses an unmanned vehicle equipped with a laser radar and satellite navigation equipment. The ground-penetrating radar equipment and the unmanned vehicle can exchange data through wireless communication, and through pre-made maps, the efficiency of ground-penetrating radar measurement is greatly improved, achieving a high degree of automation;

[0029] (2) This application also proposes an optimized method for secondary selection of measurement areas, which can improve the speed of data processing while also ensuring that the maximum amount of soil moisture information is obtained;

[0030] (3) The present application further proposes to use a drone equipped with a camera to take a crown image of a tree at a certain position in a shelterbelt sample plot, and combine it with the number of root points obtained by a ground-penetrating radar device at that position through a convolutional neural network (CNN) for analysis, thereby obtaining the corresponding crown characteristics of the trees at a position with a large number of root points. Then, the soil moisture content of the windbreak and sand-fixing forest can be detected at a high speed by collaboratively operating drone photography and unmanned vehicle-towing ground-penetrating radar equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 Shown is a schematic structural diagram of a ground penetrating radar device in the prior art;

[0032] Figure 2 Shown is a schematic diagram of the principle of detecting thick roots using ground penetrating radar equipment and a hyperbolic signal diagram of the root point;

[0033] Figure 3 Shown is a flow chart of the steps for preprocessing ground penetrating radar image data;

[0034] Figure 4 Shown is the root point hyperbola signal recognition intention of this application;

[0035] Figure 5 The figure shows the layout of the survey line of the ground penetrating radar equipment;

[0036] Figure 6 Shown is a schematic diagram of the division of the forest area to be measured;

[0037] Figure 7 The figure shows the superposition distribution diagram of the root points of the test area within the range of (i, i+1) meters in the x-axis direction on the yh plane;

[0038] Figure 8 The figure shows a schematic diagram of secondary division of the already divided forest land to be measured;

[0039] Figure 9 The figure shows a schematic diagram of a secondary measurement area selected from the divided area after the divided forest land to be measured is divided again;

[0040] Figure 10 Shown is a schematic diagram and flow chart of the spatial distribution of soil moisture inverted by ground penetrating radar equipment. DETAILED DESCRIPTION

[0041] The present application will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are intended only to illustrate the present application and are not intended to limit the present application. It should also be noted that, for ease of description, only portions relevant to the present application are shown in the accompanying drawings, not all of them.

[0042] This application was carried out in Zhanggutai Town, Zhangwu County, Liaoning Province, on the southern edge of the Horqin Desert. The ground-penetrating radar equipment used in this application is mainly composed of a digital transceiver antenna, a radar host, a GPS receiver and an intelligent ranging wheel. The unmanned vehicle used is mainly composed of a vehicle body structure, a laser radar, and a satellite navigation device. The ground-penetrating radar equipment and the unmanned vehicle can communicate wirelessly for data exchange; the drone used in this application is equipped with a camera device, which can take pictures of trees to obtain relevant information above the tree surface.

[0043] Figure 1 The figure shows the structure of the ground penetrating radar device in the prior art. Figure 1The GPR equipment shown in the figure was adapted to be towed by an unmanned vehicle. The vehicle was connected to the GPR equipment, and wireless communication between the GPR equipment and the unmanned vehicle enabled data exchange. During the test, the vehicle towed the GPR equipment along the survey line, performing scanning measurements. The GPR equipment's GPS receiver monitored its location in real time. When the GPR equipment reached a designated measuring point, it sent a command to the vehicle to stop and resume operation. The GPR equipment's movement and measurement were fully automated, significantly improving inspection efficiency and accuracy. Before measurement, a computer-generated map of the inspection area was required. The map included at least the coordinates of the trees in the forest to be measured and the coordinates of each measuring point along the survey line. The unmanned vehicle's LiDAR detected obstacles in the surrounding environment during its movement, enabling the vehicle to effectively avoid them. If the coordinates of a measuring point conflicted with the location of an obstacle (such as a rock or tree trunk that could not be moved), the vehicle would automatically skip that point and proceed to the next one.

[0044] Figure 2 The figure shows a schematic diagram of the principle of detecting thick roots by a ground-penetrating radar device and the hyperbolic signal of the root point. The basic principle of ground-penetrating radar is that electromagnetic waves will reflect differently at the interface of two materials with different electromagnetic properties, and the target object can be detected by processing and analyzing the reflected signals. Generally, the water content of thick roots of trees (diameter ≥ 5mm) and soil media is different, so there are differences in electromagnetic properties, and an electrical interface is formed between the two. During the movement of the ground-penetrating radar, the transmitting antenna (T) transmits high-frequency pulse electromagnetic waves into the ground. A part of the electromagnetic waves is reflected on the root surface and then received by the ground receiving antenna (R) ( Figure 2 As shown in the schematic diagram a), a recognizable hyperbolic signal is formed on the ground penetrating radar image ( Figure 2 (See diagram b in the figure). The signal velocity is negatively correlated with the dielectric constant of the soil above the reflection point. That is, the higher the soil moisture content, the greater the soil dielectric constant, and the slower the electromagnetic wave propagation speed in the soil. In low-conductivity, non-magnetic, and non-saline soils, the propagation velocity of the ground penetrating radar reflected wave can be approximated as:

[0045]

[0046] Where c is the propagation speed of electromagnetic waves in vacuum (0.3 m·ns -1 ), ε is the soil dielectric constant, V soil is the propagation speed of radar electromagnetic waves in the soil.

[0047] The propagation speed of the reflected wave in the soil is obtained, and then according to the soil volume water content θ soil The empirical model between and the soil dielectric constant ε can be used to obtain the soil volume water content:

[0048] θ soil =-0.053+0.0293ε-0.00055ε 2 +0.0000043ε 3 (2)

[0049] Usually, software is needed to preprocess the GPR image data. The preprocessing mainly includes the following steps:

[0050] 1) System temperature drift error correction (Dewow): eliminate the drift of the radar wave signal's first arrival time point;

[0051] 2) Background average subtraction: Use the mean filter method to remove background clutter in the radar profile image;

[0052] 3) Signal gain (Spreading and exponential gain filter, SEC2): Enhances the loss of radar signal due to energy attenuation during downward propagation.

[0053] For details, please refer to Figure 3 Shown are the preprocessing steps for ground penetrating radar image data: (a) raw data, (b) system temperature drift error correction, (c) background removal, and (d) signal gain.

[0054] Figure 4 The following figure shows the intention of the root point hyperbola signal recognition of this application; using the hyperbola semi-automatic recognition algorithm, the root point of the hyperbola signal with complete and clear characteristics in the radar image is identified, and the reflected wave velocity of the root point is calculated based on the random Hough transform and iterative method;

[0055] The top of the root point hyperbola signal corresponds to the upper surface position of the thick root. Combined with the root point reflection wave velocity, the root point depth can be determined using the following formula:

[0056] h 根点 =V soil ×t / 2 (3)

[0057] Where h 根点 is the root depth (m), V soil The speed of electromagnetic wave propagation in soil medium (m / ns), t is the round-trip time of electromagnetic wave propagation (ns). Then, according to the horizontal position of the root point on each slice, its two-dimensional coordinates (x, y) can be determined, thereby obtaining the three-dimensional spatial coordinates (x, y, h) of the root point, where h is the depth information. Then, the reflected wave velocity V soilSubstitute formula (1) to calculate the dielectric constant ε of the soil between the ground and the root point, and then calculate the average soil moisture, ASM, m between the ground and each root point according to Topp formula (2). 3 ·m -3 ), assign it to the corresponding root point pixel, and obtain the spatial scattered data set of the soil volume water content inversion value.

[0058] Due to the large amount of data, preprocessing of ground-penetrating radar image data will take a long time and affect the entire measurement cycle. Therefore, the existing technology does not set the spacing of measurement points too densely, which leads to reduced accuracy when fitting the hyperbolic signal. In addition, the water content of tree roots (diameter ≥ 5mm) and soil media is different, which means that the more tree roots there are, the more effective information the ground-penetrating radar equipment can measure. The measurement positions of some trees with fewer roots are meaningless. In response to the above problems, this application proposes an optimized method for selecting the measurement area, which can improve the speed of data processing while also ensuring that the maximum amount of soil moisture information is obtained.

[0059] like Figure 5 As shown in the figure, this experiment was conducted on a 43-year-old sandy Pinus sylvestris plantation (450 trees / ha -1 ) was conducted in a forest plot with a 10m x 10m plot. Starting from the origin in the map, the plot was scanned using the parallel line method, with a line interval of 0.25m and a step interval of 0.25m. A total of 40 lines were measured, and the line at x = 10 was not measured.

[0060] Data collection steps: First, a map of the forest to be measured is obtained in the laboratory through surveying (the map should include at least the coordinates of the trees in the forest to be measured and the coordinates of the measuring points on each survey line). Surface obstacles such as weeds, dead branches, pine cones, and stones in the survey point grid are cleared on site to reduce the impact of surface cover on detection. The unmanned vehicle tows the ground-penetrating radar equipment along the survey line and drags the antenna at a constant speed to collect data. The first survey line starts from the y-axis boundary and moves from the origin along the y-axis at intervals of 0.25m. After completing the current survey line, each subsequent survey line is translated along the x-axis at intervals of 0.25m.

[0061] The raw data from each survey line is filtered using a mean filter to remove background clutter from the profile image. The root point hyperbola signal recognition algorithm is then used to calculate a schematic diagram of the coarse root distribution within the profile image in the yh plane. Because the actual root depth and size are not of interest, only the number of roots is of interest, the data processing is simple and the calculation can be completed quickly.

[0062] Figure 6The figure shows a schematic diagram of the division of a 10m×10m sample plot; the sample plot is divided into 100 square areas with an area of ​​1 square meter. The square area of ​​1 square meter is represented by Aij, where i = 0, 1, 2,…, 9; j = 0, 1, 2,…, 9.

[0063] Figure 7 The figure shows the schematic diagram of the superimposed distribution of the root points in the test area within the range of (i, i+1) meters in the x-axis direction on the yh plane. Specifically, the original data obtained from the measurement of the survey lines at x=i, x=i+0.25, x=i+0.50, and x=i+0.75 are used only with the mean filtering method to remove the background clutter of the profile image, and the schematic diagram of the coarse root distribution in the profile image in the yh plane of each survey line is calculated. The above four schematic diagrams of the coarse root distribution are projected and superimposed on the yh plane.

[0064] Will Figure 6 and Figure 7 The number of coarse root projections in the yh plane profile image of each 1-square-meter square area Aij can be statistically analyzed. The greater the number of coarse root projections, the more effective information the ground-penetrating radar device can measure. Such areas can be selected for focused measurement and analysis, and the resulting data can better characterize the soil moisture content nearby. For example, by comparing the number of coarse root projections within each square area Aij, the square areas Aij with the top 25% of coarse root projections are re-measured with a line interval and step interval of 0.1 m. Since only a small area (one-quarter of the original area) is measured, and the ground-penetrating radar device is automatically towed by an unmanned vehicle, the measurement can be completed quickly and the effective information obtained will be richer. The selection of line interval and step interval can be determined according to actual needs. In theory, the step interval for measuring the selected square area Aij should be closer than the step interval of the initial measurement.

[0065] However, if only the square area Aij with the largest number of coarse root projections ranked in the top 25% is re-measured with a measurement line interval of 0.1, the measurement may be completely concentrated in certain root-dense areas, and may not well reflect the moisture content of the entire forest to be measured. Therefore, the applicant further proposes a preferred solution: Figure 8 Shown is the Figure 6 Schematic diagram of the secondary division of the forest land to be measured that has been divided; Figure 6The 100 square areas of 1 square meter obtained are further divided into 25 equally divided larger square areas. Each of the 25 larger areas obtained by the secondary division contains 4 square areas Aij of 1 square meter. The number of coarse root projections in the 4 square areas Aij in the larger square area is compared. Among the square areas Aij in each larger square area, the one with the largest number of coarse root projections Aij will be selected; the selected square area Aij is as follows Figure 9 As shown, the selected secondary measurement areas will be re-measured with a measurement line interval of 0.1m. More preferably, the selected secondary measurement areas Aij do not have interconnected areas (i.e., the secondary measurement areas Aij do not have overlapping sides). If the secondary measurement areas Aij do have interconnected areas, then the Aij with the second largest number of coarse root projections can be selected from the four square areas Aij within the larger square area as the secondary measurement area. In this way, multiple areas that are as evenly distributed as possible can be measured, making the measurement results more representative and more accurate. The discussion here does not take into account the situation where shelterbelt trees may occupy the entire area of ​​certain Aij, because the diameter of trees in the shelterbelts currently planted in China is generally between 30-70cm, and the 10m×10m sample plot and 1 square meter square area Aij in this application are only exemplary, and a larger numerical range can also be selected.

[0066] like Figure 9 As shown, the area A11 with the largest number of coarse root projections is selected from the larger square area composed of areas A00, A10, A01 and A11, and A12, which is adjacent to and connected to A11, will not be selected even if it is also the area with the largest number of coarse root projections in the larger square area composed of areas A02, A12, A03 and A13. Similarly, A21, which is adjacent to and connected to A11, will no longer be selected. That is, if Aij in the larger square area to be selected is connected to the already selected area to be tested Aij, the connected area Aij will be abandoned, and the areas with the second largest number of coarse roots will be selected in turn until the appropriate area to be tested Aij is selected.

[0067] For the 25 secondary measurement areas Aij obtained, the unmanned vehicle will tow the ground penetrating radar equipment to conduct secondary measurements on the above areas to obtain radar image data;

[0068] The obtained radar image data is preprocessed. The preprocessing steps include: system temperature drift error correction, background removal and signal gain;

[0069] The hyperbola semi-automatic recognition algorithm is used to identify the complete and clear root points of the hyperbola signal in the radar image, and the reflection wave velocity V of the root point is calculated based on random Hough transform and iterative method. soil;

[0070] The top of the root point hyperbola signal corresponds to the upper surface position of the thick root. Combined with the root point reflection wave velocity, the root point depth h can be determined. 根点 , refer to formula (3);

[0071] Then, according to the horizontal position of the root point on each slice, its two-dimensional coordinates (x, y) can be determined, thereby obtaining the three-dimensional spatial coordinates (x, y, h) of the root point;

[0072] The reflected wave velocity V soil Substitute formula (1) to calculate the dielectric constant ε of the soil between the ground and the root point, and then calculate the average soil moisture, ASM, m between the ground and each root point according to Topp formula (2). 3 ·m -3 );

[0073] like Figure 10 As shown in Figure 1, in order to obtain the spatial distribution characteristics of soil moisture in different soil layers, it is necessary to convert the soil volumetric water content into a parameter related to depth and then perform spatial interpolation. Therefore, the soil water storage (PSWS) is calculated by combining the soil volumetric water content and the root depth. The formula is as follows:

[0074] PSWS gpr =ASM×h 根点 ×1000 (4)

[0075] Where h 根点 (m) is the root depth, PSWS gpr (mm) is the inversion value of the soil profile water storage between the ground and the root point, and 1000 is the conversion factor from m to mm. gpr The spatial coordinates (x, y, h) of the corresponding root points are assigned to obtain a scattered data set of soil water storage inversion values.

[0076] In Matlab, we performed inverse distance weighted interpolation (IDW) on the spatially scattered data set of soil water storage to reconstruct the spatial distribution characteristics of soil water storage. This yielded a two-dimensional distribution of soil water storage at different depths (e.g., H1 and H2). By subtracting the soil water storage at different depths, we obtained the two-dimensional distribution of the average soil volumetric water content (ISM) at different soil layers using the following formula:

[0077]

[0078] or

[0079]

[0080] Where P(H1), P(H2) and P(H3) are the two-dimensional distributions of soil water storage at soil depths H1, H2 and H3, respectively; ISM(H2-H1) and ISM(H3-H2) are the volumetric soil water contents of the (H2-H1) and (H3-H2) soil layers, respectively. Figure 10 ISM2 is ISM(H2-H1) and ISM3 is ISM(H3-H2).

[0081] Although the use of unmanned vehicle-driven ground-penetrating radar equipment to count coarse roots can effectively improve the efficiency of detection and realize automated detection compared with the existing technology, through repeated experiments and verification, the present application has also creatively discovered that it is possible to use a drone equipped with a camera to capture the crown image of a tree at a certain location in the shelterbelt sample plot. The crown image can provide information such as leaf color, leaf density, leaf size, branch density, branch color, branch size, etc., and combine this with the number of root points obtained by the ground-penetrating radar equipment at that location through a convolutional neural network (CNN) for analysis, thereby obtaining the corresponding crown characteristics of trees at locations with a large number of root points. Of course, this requires learning from a large number of samples to achieve this. In future work, drone photography and unmanned vehicle-driven ground-penetrating radar equipment can be used to work together to quickly complete the detection of soil moisture content in windbreak and sand fixation forests. After the drone photographs the trees in the sample plot, the coordinates of the area with more coarse roots are obtained through computer analysis. The computer then determines the scope of the detection area and draws a map. The unmanned vehicle-driven ground-penetrating radar equipment detects within the specified area on the map. Drone photography can draw conclusions about the thickness of roots more quickly. Its error may be larger than that of initial exploration using ground-penetrating radar equipment, but its efficiency will be higher.

[0082] The above description is merely a preferred embodiment of the present application and is not intended to limit the present application. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art may still improve the technical solutions described in the aforementioned embodiments or replace some of the technical features therein with equivalents. Any modifications or equivalent replacements made within the spirit and principles of the present application shall be included within the scope of protection of the present application.

Claims

1. A method for inverting soil moisture content of windbreak and sand fixation forests based on ground penetrating radar, characterized in that: (1) Set up a 10m×10m sample plot in the forest. Before measurement, a map of the test area must be generated by computer. (2) Clear the weeds, dead branches, pine cones and stone surface obstacles within the measurement point grid on site to reduce the impact of surface cover on detection; (3) The unmanned vehicle tows the ground penetrating radar device equipped with a GPS receiver along the survey line according to the pre-generated map of the detection area, dragging the antenna at a constant speed to collect data. The first survey line starts from the y-axis boundary and moves from the origin along the y-axis at intervals of 0.25 m. The interval between the survey lines is 0.25 m, and a total of 40 survey lines are used. (4) The sample is divided into 100 square areas of 1 square meter. The square area of ​​1 square meter is represented by Aij, where i = 0, 1, 2, ..., 9; j = 0, 1, 2, ..., 9. The raw data obtained from the measurement of each survey line are only used to remove the background clutter of the profile image by the mean filtering method. Then, the root point hyperbola signal recognition algorithm is used to calculate the coarse root distribution map in the profile image in the yh plane, where h represents the depth. The coarse root distribution maps corresponding to the four survey lines at x = i, x = i + 0.25, x = i + 0.50, and x = i + 0.75 are projected and superimposed in the yh plane. (5) Counting the number of coarse root projections of the four survey lines in the cross-sectional image of the yh plane in each 1 square meter square area Aij, and selecting the square area Aij with the number of coarse root projections ranking in the top 25%; Step (5) further includes: further dividing the 100 square areas with an area of ​​1 square meter into 25 equally divided larger square areas, each of the 25 larger areas obtained by the secondary division contains 4 1 square meter square areas Aij, comparing the number of coarse root projections in the 4 square areas Aij in the larger square area, and selecting the square area Aij with the largest number of coarse root projections in each larger square area Aij; the selected secondary measurement test areas Aij do not have interconnected areas, and if the secondary measurement test areas Aij do have interconnected areas, then selecting the Aij with the second largest number of coarse root projections in the 4 square areas Aij in the larger square area as the secondary measurement test area; (6) The unmanned vehicle tows the ground penetrating radar equipment to re-measure the selected square area Aij with a line interval and step interval of 0.1m; (7) Preprocessing the radar image data obtained in step (6); (8) Using the hyperbola semi-automatic recognition algorithm, identify the complete and clear root points of the hyperbola signal in the radar image, and calculate the reflection wave velocity V of the root point based on the random Hough transform and iterative method soil ; (9) The top of the root hyperbola signal corresponds to the upper surface position of the thick root. Combined with the root reflection wave velocity, the root depth h can be determined. 根点 , the formula is as follows: 根点 =V soil ×t / 2, where t is the round-trip time of electromagnetic wave propagation; then, based on the horizontal position of the root point on each slice, its two-dimensional coordinates (x, y) can be determined, thereby obtaining the three-dimensional spatial coordinates (x, y, h) of the root point; (10) The reflected wave velocity V soil Substitute into the formula: In the equation ( ), c ​​is the propagation speed of electromagnetic waves in a vacuum. The dielectric constant ε of the soil between the ground and the root point is calculated, and then the volumetric water content of the soil between the ground and each root point is calculated according to the Topp formula.

2. The method according to claim 1, wherein The GPS receiver of the ground-penetrating radar equipment can monitor the location information of the ground-penetrating radar equipment in real time. When the ground-penetrating radar equipment reaches a designated measuring point, it will send a command to the unmanned vehicle to stop moving and the ground-penetrating radar equipment will start working.

3. The method according to claim 1, wherein The pre-processing steps in step (7) include: system temperature drift error correction, background removal and signal gain.

4. The method according to claim 1, wherein Calculate soil water storage PSWS by combining soil volumetric water content and root depth gpr , the formula is as follows: PSWS gpr =ASM×h 根点 ×1000,h 根点 is the root point depth, 1000 is the conversion factor from m to mm, and ASM is the soil volumetric water content.

5. The method according to claim 4, wherein The spatial distribution characteristics of soil water storage were reconstructed by performing inverse distance weighted interpolation on the spatial scattered data set of soil water storage, and the two-dimensional distribution characteristics of soil water storage at different depths were obtained.

6. The method according to claim 1, wherein Further including: A drone equipped with a camera was used to capture crown images of trees at a certain location in the shelterbelt plot. The crown images provided information including leaf color, leaf density, leaf size, branch density, branch color, and branch size. These images were then combined with the number of root points obtained at that location by ground-penetrating radar equipment through a convolutional neural network for analysis, thereby obtaining the corresponding crown characteristics of trees at locations with a large number of root points.

7. The method according to claim 6, wherein The drone takes photos of the trees in the sample plot and then uses computer analysis to obtain the coordinates of the area with more thick roots.

8. The method according to claim 1, wherein The map of the detection area includes at least the coordinates of the trees in the forest to be measured and the coordinates of the measuring points of each measuring line.

Images