Ploughing soil layer position detection method and device based on ground penetrating radar and Markov random field

Abstract

The invention discloses a cultivated land soil layer position detection method and device based on a ground penetrating radar and a Markov random field, cultivated land B-SCAN data is obtained through the ground penetrating radar, preprocessing and layer position seed point selection are carried out, and a Markov random field model is constructed based on an undirected graph model and an initial layer position tracking mark field; calculating an optimal horizon point by integrating the posterior probability energy field of the horizon point and the horizon envelope feature advantage item; and finally, calculating a horizon depth value according to the soil dielectric constant, and evaluating whether the thicknesses of the plough layer and the effective soil layer reach the standard or not. The method has higher precision and stability in a complex soil environment, can realize efficient and accurate soil layer tracking without relying on large-scale high-quality data, is suitable for rapid processing of large-scale cultivated land soil data, meets soil layer thickness measurement and evaluation requirements in land reclamation and high-standard farmland construction, and has a wide application prospect. And an innovative technical solution is provided for farmland management and construction.

Description

Technical Field

[0001] This application belongs to the technical field of ground penetrating radar echo data interpretation, and particularly relates to a method and device for automatically detecting horizons of cultivated land soil echo data. Background Art

[0002] Land consolidation refers to the arrangement, reclamation, and development of unused or unreasonably utilized land to increase the effective cultivated land area and improve the quality of cultivated land, which is the basis for the construction of high-standard farmland. However, due to the great difficulty in controlling the quality of land consolidation projects, there are potential risks such as low acceptance standards, lack of acceptance means, and fraud. To address these issues, studying efficient and accurate land consolidation project quality acceptance technologies has become an important means to ensure the quality of newly reclaimed cultivated land construction.

[0003] The detection of cultivated land soil horizons is an important link in ensuring the quality of land consolidation and high-standard farmland construction. The key lies in accurately measuring the thicknesses of the plow layer and the anti-seepage layer, which together constitute the effective soil layer thickness. Existing detection technologies include the profile method, the boring method, geophysical methods, etc. Ground penetrating radar belongs to one of the geophysical methods and has the characteristics of high efficiency, non-destructiveness, and continuity. In recent years, with the rapid development of ground penetrating radar technology, this technology has gradually been applied to the field of soil detection. When using ground penetrating radar to detect the thickness of cultivated land soil horizons, automatic horizon tracking of ground penetrating radar data is a crucial step, but there are often the following problems: (1) The nearby rocks look very similar to the considered soil horizons, and the tracking algorithm is easily jumped to the wrong horizon; (2) Due to the structural deformation in the underground environment of cultivated land, the horizons may be discontinuous; (3) The appearance of a horizon may vary at different positions, and nearby horizons may bifurcate or disappear.

[0004] Currently, most existing horizon tracking methods focus on the field of signal correlation analysis, and there is less research on ground penetrating radar soil echo data. For example, the correlation-based sliding window algorithm (CN111562620A, CN109886989A, CN118795468A). Such methods calculate the correlation between waveforms to quickly find suitable peak targets. The tracking results have good continuity and fast calculation speed. However, this method is easily affected by local noise, and it is difficult to distinguish the rock signals near the horizons. The tracking algorithm is easily jumped to the wrong soil layer and cannot be competent for the soil environment with variable reflection signals. Although some methods combined with deep learning have improved intelligence, they often require high-quality data drive to adapt to the situation of seasonal changes and regional differences in soil properties, and there are many difficulties in the application of actual products. Therefore, in the process of interpreting cultivated land soil data, there is an urgent need for a stable, efficient, and widely applicable automatic horizon detection algorithm to accurately calculate the thicknesses of the plow layer and the anti-seepage layer and provide an important basis for scientifically evaluating the quality of cultivated land. Summary of the Invention

[0005] The purpose of the present invention is to provide a method and device for detecting the horizons of cultivated land soil layers based on ground penetrating radar and Markov random field in view of the deficiencies of the prior art, aiming to solve the problem in the prior art that the thicknesses of the tillage layer and the anti-seepage layer cannot be calculated quickly, non-destructively and accurately.

[0006] The principle of ground penetrating radar is that when high-frequency electromagnetic waves propagate underground and encounter interfaces with different dielectric constants, reflections will occur. By receiving these reflected waves and analyzing their travel time, amplitude and waveform, the structure and morphological size of underground media can be inferred. The transmitting antenna of the ground penetrating radar emits high-frequency broadband electromagnetic wave pulses underground, and the receiving antenna receives the reflected waves from the underground media interfaces. The echo data obtained at each detection point is called A-SCAN data, and the echoes of multiple detection points are superimposed to form B-SCAN data.

[0007] To achieve the above objectives, the objectives of the present invention are realized through the following technical solutions: In the first aspect, the present invention provides a method for detecting the horizons of cultivated land soil layers based on ground penetrating radar and Markov random field, and the method includes the following steps:

[0008] Step 1: Use ground penetrating radar to obtain B-SCAN data of the underground soil layer of the cultivated land on the arranged sampling line, and then perform preprocessing;

[0009] Step 2: Select horizon seed points, construct an undirected graph model on the B-SCAN data according to the same-polarity principle, calculate the shortest path, and determine the initial horizon tracking marker field;

[0010] Step 3: Establish a Markov random field model MRF according to the undirected graph model and the initial horizon tracking marker field, and calculate the posterior probability energy field of the horizon points;

[0011] Step 4: Determine the dominant term according to the horizon envelope characteristics, construct an energy objective function for the horizon points by integrating the posterior probability energy field and the dominant term, and optimize to calculate the optimal horizon points;

[0012] Step 5: Calculate the horizon depth value according to the soil dielectric constant;

[0013] Step 6: Obtain data of different survey lines on the sampling line, repeat Steps 1 to 5 until all data is processed, and evaluate whether the thicknesses of the cultivated layer and the effective soil layer of the cultivated land meet the standards.

[0014] Further, in Step 1, the sampling line is a number of radar survey lines arranged according to the size and shape of the cultivated land. The preprocessing is to preprocess the B-SCAN data of the underground soil layer of the cultivated land in the order of removing the DC component, band-pass filtering, removing background noise, initial maximum phase calibration, zero-point adjustment and linear gain.

[0015] Further, in step two, the process of constructing the undirected graph model is to expand new nodes forward and backward along the horizontal position of the survey line from the horizon seed points according to the same-polarity principle, and connect them with edges. Among them, each node represents a data point, and the edge represents the adjacent relationship between these points. The starting point is the horizon seed point, and the end point is the point with the strongest envelope feature in the last column along the expansion direction. The undirected graph is represented as ; where represents the node set, , represents the nth node; represents the edge set, , represents the edge connecting node and node ; The expansion according to the same-polarity principle means that the positive and negative polarities of the amplitude of the new node are the same as those of the horizon seed point, and each time new nodes are expanded on the A-SCAN data, starting from the nearest position in sequence; The edges in the undirected graph model use the amplitude correlation of adjacent nodes as weights.

[0016] Further, the initial horizon tracking marker field marks the horizon points and non-horizon points with different numbers, and the marked positions are the nearest extreme value positions of the waveforms in the corresponding periods.

[0017] Further, in step three, the Markov random field model needs to define the dependence relationship between nodes and construct an energy function:

[0018]

[0019]

[0020]

[0021] Among them, the prior term , is the feature of node , is the corresponding horizon point marking label, is the probability that node is given a certain mark; The feature is a multi-dimensional vector composed of the envelope feature, cosine phase feature, frequency domain feature and entropy feature of the node, and the is calculated by the multivariate Gaussian probability distribution, refers to the parameters of the multivariate Gaussian distribution , , is the mean vector, is the covariance matrix; The neighborhood constraint term , is the strength parameter of the smoothing term, It is the scale parameter in the distance metric, which is used to control the influence range of the distance. is the Euclidean distance between adjacent node features; The value is determined based on the mean and standard deviation of the Euclidean distances of all adjacent nodes.

[0022] Furthermore, the Markov random field model is solved according to the expectation maximization algorithm EM. The EM algorithm includes an E step and an M step. In the E step, the posterior probability of the node label is calculated according to the observed data and the current model parameters. In the M step, the model parameters are updated. ;if The absolute error before and after using the EM algorithm is less than Or the number of iterations is greater than N times, the calculation is completed, and the maximum a posteriori probability energy field of the layer site is obtained. The maximum a posteriori probability energy field of the layer site refers to the probability matrix that each node belongs to the layer site.

[0023] Further, in step 4, the advantage item is ,in, It is a node The envelope characteristic intensity conversion value at is the weight coefficient; two-dimensional window is used to calculate the nodes The instantaneous amplitude of these regions is sorted from large to small and the first several points are taken to obtain , corresponding to , the remaining nodes , b is a constant greater than 0; the objective function is , with the help of graph theory model, the shortest path of the layer marking field energy is calculated, and this path is the final layer marking result.

[0024] Furthermore, in step five, the effective soil layer thickness refers to the total thickness of the soil above the parent material layer that can be used by crops, including the tillage layer and the impermeable layer; when there is a barrier layer, it is the thickness of the soil layer above the barrier layer, and the layer in paddy field farmland includes the tillage layer, the impermeable layer and the infiltration layer; the method for evaluating whether the thickness of the tillage layer and the effective soil layer of the cultivated land meets the standard is to determine the confidence interval of the thickness of the tillage layer and the effective soil layer of the cultivated land respectively. , ,in, is the mean, is the standard deviation, is the sample size, for Distribution critical value; if In error If the relevant standards or specifications are met, the thickness of the cultivated layer and the thickness of the effective soil layer of the cultivated land will meet the standards.

[0025] In a second aspect, the present invention also provides a device for detecting the horizons of cultivated land soil layers based on ground penetrating radar and Markov random field, including a memory and one or more processors. Executable code is stored in the memory, and when the processor executes the executable code, a method for detecting the horizons of cultivated land soil layers based on ground penetrating radar and Markov random field as described is implemented.

[0026] In a third aspect, the present invention also provides a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, the method for detecting the horizons of cultivated land soil layers based on ground penetrating radar and Markov random field as described is implemented.

[0027] Compared with the prior art, the present invention has at least the following beneficial effects:

[0028] Based on further analysis and research of the problems in the prior art, the present invention realizes that the current methods of applying correlation and deep learning in other media are difficult to effectively track the horizons of cultivated land soil layers. Through a method for detecting the thickness of the cultivated layer and the effective soil layer of cultivated land based on Markov random field provided by the present invention, an undirected graph model is constructed in the B-SCAN data of cultivated land, and a posterior probability energy field model is established in combination with the Markov random field (MRF), which can effectively cope with the complex and diverse reflection signals in the cultivated land soil environment.

[0029] Compared with the traditional correlation-based method, the method of the present invention can more accurately distinguish the signals near the soil horizons and avoid tracking and jumping to the wrong horizons. By introducing horizon envelope features and dominant terms to construct a horizon point energy objective function, a variety of feature information is fused into the optimization objective, effectively improving the robustness and anti-noise performance of the algorithm. Even in the case of low signal quality or significant changes in soil properties, a high tracking accuracy can be ensured. In addition, the method of the present invention does not rely on high-quality data-driven, but is based on physical characteristics and a stable horizon tracking algorithm, avoiding the dependence of deep learning methods on large-scale and high-quality labeled data, and is applicable to scenarios where the data acquisition cost is high and the spatio-temporal variation of soil properties is significant in practical applications. Through the optimization calculation of the graph theory model and Markov random field, the initial horizon tracking marker field can be quickly determined and the optimal horizon points can be optimized. The tracking result has good continuity and high calculation efficiency, and can meet the rapid processing requirements of large-scale cultivated land soil data.

[0030] The present invention can quickly, non-destructively and accurately detect the cultivated layer and the anti-seepage layer, provides a method for evaluating whether the thickness of the cultivated layer and the effective soil layer of cultivated land meets the standard, can well meet the measurement and evaluation requirements of the thickness of the cultivated layer and the effective soil layer in land improvement and high-standard farmland construction, and provides an innovative technical solution for farmland management. Description of the Drawings

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0032] Figure 1 The flowchart of a method for detecting the soil layer position of cultivated land based on ground penetrating radar and Markov random field provided by an embodiment of the present application.

[0033] Figure 2 The schematic diagram of the survey line of cultivated land provided by an embodiment of the present application (top view).

[0034] Figure 3 The schematic diagram of constructing an undirected graph theory model provided by an embodiment of the present application, where (a) is the schematic diagram of the B-SCAN waveform and (b) is the B-SCAN grayscale image.

[0035] Figure 4 The schematic diagram of the initial layer position marking field provided by an embodiment of the present application.

[0036] Figure 5 The schematic diagram of constructing layer position features provided by an embodiment of the present application.

[0037] Figure 6 The schematic diagram of the energy superposition of the Markov random field provided by an embodiment of the present application.

[0038] Figures 7 - 12 The layer position tracking result diagrams of 6 groups of tillage layers and anti-seepage layers provided by the embodiments of the present application.

[0039] Figure 13 The structural diagram of a device for detecting the soil layer position of cultivated land based on ground penetrating radar and Markov random field of the present application. Detailed implementation manners

[0040] To more clearly illustrate the technical problems to be solved, the technical solutions and the beneficial effects of the present application, the following will further elaborate on the present application in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0041] The following will be further detailed through specific implementation manners:

[0042] In an embodiment of the present invention, a method for detecting the soil layer position of cultivated land based on ground penetrating radar and Markov random field is provided. As Figure 1 shown, it includes the following steps:

[0043] Step 1: Use ground penetrating radar to obtain the B-SCAN data of the underground soil layer of cultivated land on the arranged sampling line and perform preprocessing. The arrangement method of the sampling line in this embodiment is as Figure 2 shown. The survey line includes at least 3 transverse lines and 3 longitudinal lines, in a "well" shape, with a certain distance between each survey line, such as 2.5 m. Then, the obtained B-SCAN data of the underground soil layer of cultivated land is preprocessed in the order of removing the DC component, band-pass filtering, removing background noise, initial maximum phase calibration, zero-point adjustment, and linear gain.

[0044] Step 2: Select layer seed points, construct an undirected graph model on the B-SCAN data, calculate the shortest path, and determine the initial layer tracking marker field.

[0045] Taking the B-SCAN data of survey line 1 in this embodiment as an example, Figure 3 it is a schematic diagram of the corresponding seed points, where Figure 3 in (a) is the waveform diagram of the B-SCAN data, Figure 3 in (b) is Figure 3 the two-dimensional grayscale diagram converted from the waveform diagram in (a) of Figure 3 in (b). Each column in (b) represents a waveform data, and each row represents a data sampling point. As Figure 3 shown in (a), select the layer seed point position [104, 164]. 104 refers to the position of the seed point at the 104th sampling point, and 164 refers to the seed point on the 164th trace. Then, construct an undirected graph model according to the same-polarity principle. The undirected graph model means expanding new nodes from the layer seed point to the forward and backward directions of the horizontal position (X-axis) of the survey line and connecting them with edges; among them, each node represents a data point, and the edge represents the adjacent relationship between these points. The starting point is the layer seed point, and the end point is the point with the strongest envelope feature in the last column along the expansion direction. The undirected graph , , , where represents the node set, represents the edge set, represents the node and the edge connecting node , represents the nth node.

[0046] The expansion according to the same-polarity principle means that the positive and negative polarities of the amplitude of the new node are the same as those of the layer seed point, and each time new nodes are expanded on the A-SCAN data. In this example starting from the position closest to the previous node, search sequentially in the vertical direction. In the undirected graph model, the edges use the correlation between adjacent nodes as weights, and the correlation refers to the similarity of nodes. In this embodiment, the amplitude features of the node regions are used to calculate the correlation, which is extracted with a one-dimensional window, and the window size is column vectors. The correlation calculation formula for two adjacent nodes is as follows:

[0047]

[0048] where and are the amplitudes of the th column and the th column waveforms of the th adjacent nodes, is the window length, which is taken as 14 here.

[0049] In this embodiment, two undirected graph models are constructed from the seed point forward and backward. The weight of the edge takes the reciprocal of the correlation coefficient, and the Dijkstra algorithm is used to solve the shortest path, and the results are merged.

[0050] The Dijkstra algorithm gradually constructs the shortest path of the entire graph by gradually selecting the shortest path nodes that have not been visited currently and updating the shortest paths of their neighbors. The node positions in the shortest path are recorded as layer points, the layer points are marked as 2, the non-layer points are marked as 1, and the marked positions are the nearest extreme positions of the periodic waveforms where they are located. Finally, the initial layer position marking field is obtained. As shown in the matrix of Figure 4 , the positions of the pentagrams are the peak layer points, marked as 2, and the non-layer points are marked as 1. According to the initial layer position marking field, the initial parameters of the model are calculated. The model parameters in this embodiment are the mean vector and the covariance matrix .

[0051] Step 3: Establish a Markov Random Field (MRF) model based on the undirected graph model and the initial layer position tracking marking field, and calculate the posterior probability energy field of the layer points. The Markov Random Field model needs to define the dependence relationship between nodes and introduce an energy function to describe the change of node states. The energy function is constructed as follows:

[0052]

[0053] ,

[0054] where the prior term , is the feature of node , is the corresponding layer site marking label, is a node According to the feature Assign a label The probability of. As Figure 5 shown, Contains the instantaneous amplitude, cosine phase, frequency domain feature, and entropy feature of the node Together, they form a 4D vector, Then it is calculated according to the multivariate Gaussian distribution.

[0055]

[0056] Among them, Refers to the parameters of the multivariate Gaussian distribution 、 , Is the covariance matrix, Is the mean vector, Is the dimension.

[0057] Neighborhood constraint term , Is the intensity parameter of the smoothing term, Is the scale parameter in the distance metric, used to control the influence range of the distance; Is the Euclidean distance of the features of adjacent nodes. If The labels are the same, The value is 0. If The labels are different, then . According to the mean and standard deviation of the Euclidean distances of all adjacent nodes, it can be further optimized through experiments. In this embodiment Take 0.5, Take 0.8.

[0058] In the Bayesian framework, the Markov random field model is solved using the Iterative Conditional Mode (ICM) and Expectation-Maximization (EM) algorithms to predict the label field. ICM calculates the conditional probability of each node according to the initial parameters of the model, updates the label of the node, and makes the energy function locally minimum, which can be expressed as:

[0059]

[0060] Among them, Is the node 's neighborhood. The meaning of the formula is to calculate the local energy under different labels. The label of the node Is updated to the label that makes the energy function locally minimum 。

[0061] When using the EM algorithm to solve, in the E-step, according to the observed data and the current model parameters, calculate the posterior probability of the node label. The posterior probability , where ; Also use the multivariate Gaussian function model for calculation; represents the node labeled as , represents the sum of the posterior probabilities that the nodes are respectively labeled as 1 and 2; represents the conditional probability that it is the node , represents the set of node labels after removing the element new set, The calculation formula is as follows:

[0062]

[0063] where is the neighborhood of the node , that is, the connected nodes.

[0064] In the M-step, according to the conditional probability of the E-step, update the model parameters , which can be expressed as , is the number of iterations, and the parameter includes the covariance matrix , the mean vector , and the update formula is as follows:

[0065] ,

[0066] where is the total number of nodes.

[0067] The present invention uses ICM conditional iteration according to the new parameters to obtain the posterior probability that each node belongs to the layer site, and then calculates , if the absolute error before and after using the EM algorithm is less than (the preferred value is ) or the number of iterations is greater than N (the preferred value is 40), the calculation is completed. According to the final result, the present invention calls the posterior probability of all nodes the maximum posterior probability energy field.

[0068] Step 4: Determine the dominant terms based on the horizon envelope features, construct the horizon point energy objective function by integrating the posterior probability energy field and the dominant terms, and optimize to calculate the optimal horizon points, as Figure 6 shown. The dominant terms are , where is the converted value of the envelope feature intensity at node , and is the weight coefficient. Calculate the average instantaneous amplitude of the node area according to a two-dimensional window of size . Sort the instantaneous amplitudes in these areas from large to small and take the first several points, denoted as . For these nodes, , and for the remaining nodes, , with b taking the value of 3. The objective function is . In this embodiment, the weight coefficient and the intensity parameter of the smoothing term in the objective function are determined by the grid search method, , . Evaluate the model performance for each combination of candidate values in the validation set, and finally determine the optimal parameters , . Then, with the help of the previously constructed graph theory model, solve it using the Dijkstra algorithm to obtain the set of optimal horizon points .

[0069] Step 5: Calculate the horizon depth value according to the soil dielectric constant. Repeat the above steps to process all the data and evaluate whether the thicknesses of the tillage layer and the effective soil layer meet the standards. This embodiment can be used for newly reclaimed arable land for planting aquatic plants. The horizons include the tillage layer and the anti-seepage layer. The effective soil layer thickness refers to the sum of the thicknesses of the tillage layer and the anti-seepage layer. Each B-SCAN data tracks 2 horizons, and the horizon tracking results are as Figures 7 - 12 shown. There are a total of 6 groups of graphs. The upper side of each group of graphs is the horizon tracking result graph, and the lower side is the absolute deviation of the corresponding horizon points. After obtaining the horizon points in Step 4, calculate the horizon point depth value , the dielectric constant of the tillage layer , the dielectric constant of the anti-seepage layer , is the time at the horizon point position in the ground penetrating radar B-SCAN data, and is the speed of light.

[0070] In order to better reflect the beneficial effects of the present application, this embodiment provides the absolute average deviation of the tillage layer and the impermeable layer in these 6 groups of B-SCAN data, and compares it with the traditional layer tracking method based on waveform correlation. The calculation results of the absolute average deviation of the tillage layer, the impermeable layer and all layers of the present application are better than the traditional layer tracking method based on waveform correlation; the layer tracking method based on waveform correlation starts from the seed point, calculates the correlation between the subsequent waveform and the reference waveform column by column, marks the position with the highest correlation as the layer position of the current column, and updates the reference waveform.

[0071] The absolute mean deviation , is the actual average layer thickness of the tillage layer and the anti-seepage layer, which is measured by cutting open the cultivated soil layer. It is the average thickness calculated based on B-SCAN data.

[0072] The method for evaluating whether the thickness of the tillage layer and the effective soil layer meets the standard is to determine the confidence interval of the thickness of the tillage layer and the effective soil layer respectively. , ,like ,in, is the mean, is the standard deviation, is the sample size, and an average of 200 groups of layer results are sampled in each B-SCAN data, forming a total of 1200 groups of samples. for Distribution critical value. If the relevant standards or specifications are met, such as the requirements in GB / T 30600-2022 "General Rules for High-standard Farmland Construction", the thickness of the cultivated layer and the effective soil layer of the cultivated land meet the standards. In this embodiment, the thickness of the cultivated layer should be greater than 20 cm, and the thickness of the effective layer should be greater than 60 cm. Take 8cm. After calculation, we get the 95% confidence interval of the tillage layer. , the plow bottom layer is based on a 95% confidence interval , unit cm, , , it can be basically determined that the thickness of the cultivated layer and the effective soil layer of the newly reclaimed land meet the requirements for high-standard farmland construction.

[0073] Corresponding to the aforementioned embodiment of a method for detecting cultivated land soil layers based on ground penetrating radar and Markov random fields, the present invention also provides an embodiment of a device for detecting cultivated land soil layers based on ground penetrating radar and Markov random fields.

[0074] See also Figure 13, an apparatus for detecting the soil layer position of cultivated land based on ground penetrating radar and Markov random field provided by an embodiment of the present invention includes a memory and one or more processors. An executable code is stored in the memory. When the processor executes the executable code, it is used to implement a method for detecting the soil layer position of cultivated land based on ground penetrating radar and Markov random field in the above embodiment.

[0075] An embodiment of an apparatus for detecting the soil layer position of cultivated land based on ground penetrating radar and Markov random field provided by the present invention can be applied to any device with data processing capabilities. The any device with data processing capabilities can be a device or apparatus such as a computer. The apparatus embodiment can be implemented by software, or by hardware or a combination of software and hardware. Taking software implementation as an example, as a logically meaningful apparatus, it is formed by the processor of any device with data processing capabilities reading the corresponding computer program instructions in the non-volatile memory into the memory for running. From a hardware level, as Figure 13 shown, it is a hardware structure diagram of any device with data processing capabilities where an apparatus for detecting the soil layer position of cultivated land based on ground penetrating radar and Markov random field provided by the present invention is located. In addition to Figure 13 the processor, memory, network interface, and non-volatile memory shown, usually according to the actual functions of any device with data processing capabilities where the apparatus in the embodiment is located, other hardware may also be included, which will not be elaborated here.

[0076] For the implementation processes of the functions and roles of each unit in the above apparatus, please refer to the implementation processes of the corresponding steps in the above method for details, which will not be elaborated here.

[0077] For the apparatus embodiment, since it basically corresponds to the method embodiment, the relevant parts can refer to the partial description of the method embodiment. The apparatus embodiments described above are only illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the present invention solution. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0078] An embodiment of the present invention also provides a computer-readable storage medium, on which a program is stored. When the program is executed by a processor, it implements a method for detecting the soil layer position of cultivated land based on ground penetrating radar and Markov random field in the above embodiment.

[0079] The computer-readable storage medium may be an internal storage unit of any device with data processing capabilities described in any of the foregoing embodiments, such as a hard disk or memory. The computer-readable storage medium may also be an external storage device of any device with data processing capabilities, such as a plug-in hard disk, a Smart Media Card (SMC), an SD card, a Flash Card, etc. equipped on the device. Further, the computer-readable storage medium may also include both an internal storage unit and an external storage device of any device with data processing capabilities. The computer-readable storage medium is used to store the computer program and other programs and data required by any device with data processing capabilities, and may also be used to temporarily store the data that has been output or will be output.

[0080] The present invention also provides a computer program product, including a computer program, which when executed by a processor, implements the method for detecting the soil layer position of cultivated land based on ground penetrating radar and Markov random field described above.

[0081] The above description only represents a preferred embodiment of the present application, and does not limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for detecting soil layer positions of cultivated land based on ground penetrating radar and Markov random field, characterized in that: The method comprises the following steps: Step 1: Use ground penetrating radar to obtain B-SCAN data of the underground soil layer of the cultivated land on the arranged sampling line, and then perform preprocessing; Step 2: Select the layer seed point, build an undirected graph model on the B-SCAN data according to the same polarity principle, calculate the shortest path, and determine the initial layer tracking mark field; Step 3: Establish a Markov random field model MRF based on the undirected graph model and the initial layer tracking mark field, and calculate the posterior probability energy field of the layer position; Step 4: Determine the dominant term according to the layer envelope characteristics, build the layer site energy objective function by integrating the posterior probability energy field and the dominant term, and optimize and calculate the best layer site; Step 5: Calculate the layer depth value according to the soil dielectric constant; Step 6. Obtain data from different survey lines on the sampling line, repeat steps 1 to 5 until all data are processed, and evaluate whether the thickness of the cultivated land cultivation layer and effective soil layer meets the standards.

2. The method for detecting soil layer of cultivated land based on ground penetrating radar and Markov random field according to claim 1, characterized in that: In step one, the sampling lines are a number of radar survey lines arranged according to the size and shape of the cultivated land; the preprocessing is to preprocess the B-SCAN data of the underground soil layer of the cultivated land in the order of removing DC components, bandpass filtering, removing background noise, initial maximum phase calibration, zero point adjustment and linear gain.

3. The method for detecting soil layer of cultivated land based on ground penetrating radar and Markov random field according to claim 1, characterized in that: In step 2, the process of constructing the undirected graph model is to expand new nodes forward and backward from the horizon seed point according to the same polarity principle to the horizontal position of the survey line, and connect them with edges, where each node represents a data point, and the edge represents the adjacent relationship between these points. The starting point is the horizon seed point, and the end point is the point with the strongest envelope feature in the last column along the expansion direction. The undirected graph is represented as ;in represents a set of nodes, , Represents the nth node; represents the edge set, , Represents a node and nodes The expansion according to the same polarity principle means that the amplitude of the new node is consistent with the layer seed point, and each expansion on the A-SCAN data New nodes are searched from the nearest position in sequence; the edges in the undirected graph model use the amplitude correlation of adjacent nodes as weights.

4. The method for detecting soil layer of cultivated land based on ground penetrating radar and Markov random field according to claim 1, characterized in that: The initial layer tracking mark field marks the layer position and the non-layer position with different numbers, and the mark position is the nearest extreme value position of the waveform of the cycle.

5. The method for detecting soil layer of cultivated land based on ground penetrating radar and Markov random field according to claim 1, characterized in that: In step 3, the Markov random field model needs to define the dependencies between nodes and construct an energy function: ； ； ； Among them, the prior , It is a node Features, Label the corresponding layer sites, It is a node The probability of assigning a certain mark; the feature is a multidimensional vector consisting of the envelope feature, cosine phase feature, frequency domain feature and entropy feature containing the node, It is calculated by multivariate Gaussian probability distribution, Refers to the parameters of the multivariate Gaussian distribution , , is the mean vector, is the covariance matrix; neighborhood constraint , is the strength parameter of the smoothing term, It is the scale parameter in the distance metric, which is used to control the influence range of the distance. is the Euclidean distance between adjacent node features; The value is determined based on the mean and standard deviation of the Euclidean distances of all adjacent nodes.

6. The method for detecting soil layer of cultivated land based on ground penetrating radar and Markov random field according to claim 5, characterized in that: The Markov random field model is solved according to the expectation maximization algorithm EM. The EM algorithm includes an E step and an M step. In the E step, the posterior probability of the node label is calculated according to the observed data and the current model parameters. In the M step, the model parameters are updated. ;if The absolute error before and after using the EM algorithm is less than Or the number of iterations is greater than N times, the calculation is completed, and the maximum a posteriori probability energy field of the layer site is obtained. The maximum a posteriori probability energy field of the layer site refers to the probability matrix that each node belongs to the layer site.

7. The method for detecting soil layer of cultivated land based on ground penetrating radar and Markov random field according to claim 1, characterized in that: In step 4, the advantage term is ,in, It is a node The envelope characteristic intensity conversion value at is the weight coefficient; two-dimensional window is used to calculate the nodes The instantaneous amplitude of these regions is sorted from large to small and the first several points are taken to obtain , corresponding to , the remaining nodes , b is a constant greater than 0; the objective function is , with the help of graph theory model, the shortest path of the layer marking field energy is calculated, and this path is the final layer marking result.

8. The method for detecting soil layer of cultivated land based on ground penetrating radar and Markov random field according to claim 1, characterized in that: In step 5, the effective soil layer thickness refers to the total thickness of the soil above the parent material layer that can be used by crops, including the tillage layer and the impermeable layer; when there is a barrier layer, it is the thickness of the soil layer above the barrier layer, and the layer in paddy field farmland includes the tillage layer, the impermeable layer and the infiltration layer; the method for evaluating whether the thickness of the tillage layer and the effective soil layer of the cultivated land meets the standard is to determine the confidence interval of the thickness of the tillage layer and the effective soil layer of the cultivated land respectively. , ,in, is the mean, is the standard deviation, is the sample size, for Distribution critical value; if In error If the relevant standards or specifications are met, the thickness of the cultivated layer and the thickness of the effective soil layer of the cultivated land will meet the standards.

9. A device for detecting soil layer positions of cultivated land based on ground penetrating radar and Markov random field, comprising a memory and one or more processors, wherein the memory stores executable code, characterized in that: When the processor executes the executable code, a method for detecting cultivated land soil layers based on ground penetrating radar and Markov random fields as described in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, a method for detecting cultivated land soil layers based on ground penetrating radar and Markov random fields as described in any one of claims 1 to 8 is implemented.

Images