Method and device for determining deployment strategy of soil moisture sensor network

Abstract

The invention discloses a method and device for determining a deployment strategy of a soil moisture sensor network, and the method comprises the steps: obtaining the feature data of each pixel point based on the collected digital elevation data of a to-be-deployed region through employing the soil moisture data obtained through the inversion of microwave remote sensing data as reference data; clustering is carried out according to the feature data of all the pixel points, the positions of representative sample points are obtained, then modeling is carried out by using a Gaussian process regression model, evaluation and screening are carried out by combining the obtained positions of the representative sample points and the inverted soil moisture data, and therefore the optimal deployment positions and number of sensors are determined; the number and the positions of the sensors selected through the method are high in accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of soil moisture sensor network deployment, and in particular to a method and device for determining a soil moisture sensor network deployment strategy. Background Art

[0002] Soil moisture is a key variable that affects surface and atmospheric water cycles and energy balance. Ecological drought, water resources planning, soil management, underground pollutant transport, and precision agriculture all rely on spatial distribution information of soil moisture (i.e., soil moisture content). In recent years, advances in Internet of Things (IoT) technology have promoted the application of wireless sensor networks (WSNs) in intelligent soil sensing. However, soil moisture is highly heterogeneous in space, and its distribution is significantly affected by vegetation cover, soil texture heterogeneity, topography, and physical and chemical properties at the basin scale. Therefore, it is crucial to rationally select the deployment location of soil characteristic sensors to representatively observe and predict soil moisture at the basin scale and ensure that the collected data can accurately reflect the soil moisture characteristics of the entire region.

[0003] To optimize the layout of soil moisture sensor networks, existing methods usually rely on collecting data from sample points within a specific time period and using AP (Affinity Propagation) neighbor propagation clustering and other methods to consider the similarity and difference of soil moisture data. By calculating the similarity between soil moisture data in different geographical locations, similar sample points are classified into one category in order to reduce data redundancy and reduce the investment cost of sensors. However, such methods rely on deploying a large number of collection nodes in the study area to obtain soil moisture data, which not only has high labor costs but also a long data collection cycle. In addition, based on unsupervised classification methods (such as K-means clustering, hierarchical clustering, etc.), although soil areas with different characteristics can be identified, the optimal number of categories cannot be determined, and therefore the optimal number of sensors to be deployed cannot be determined.

[0004] At present, microwave remote sensing technology calculates soil moisture through optical and thermal data. The most widely used one is the Thermal-Optical TRAapezoid Model (TOTRAM) based on surface temperature and vegetation index. This model uses the "trapezoidal" distribution relationship between surface temperature and vegetation index to identify different soil moisture conditions, thereby realizing the estimation and inversion of soil moisture. Microwave remote sensing technology has been successfully applied to estimate surface soil moisture, and the soil moisture distribution information provided by microwave remote sensing data can be used to verify the simulation effect of the proposed deployment of sensors. However, for the layout of sensors, the existing soil moisture sensor optimization location method has the following shortcomings and cannot meet the needs of efficient, high-quality and representative deployment of field sensors:

[0005] (1) Although clustering methods based on the similarity of soil moisture data collected from sample points (such as AP clustering) can reduce redundant sensor data and improve data utilization efficiency, they cannot fully consider complex geographical and soil properties, resulting in inaccurate sensor deployment in some areas. At the same time, the cost of obtaining soil moisture data is high, and the labor and time costs are also high. Especially in remote areas at the watershed scale, soil moisture data is relatively scarce, and sensor points may not provide enough samples for accurate clustering analysis.

[0006] (2) Unsupervised classification based on watershed scale characteristics (such as slope, aspect, climate, solar radiation, soil texture, and vegetation cover type, etc.) cannot determine the appropriate number of clusters. If the number of clusters is not selected properly, it may lead to over-clustering or under-clustering, affecting the effectiveness of the analysis results, and further causing redundancy in the deployment of sites, reducing economic practicality. In addition, clustering results usually do not have clear labels to verify their correctness. In the absence of a reference dataset, the classification results may only be based on the intrinsic patterns of the data and lack ground truth for verification. Summary of the invention

[0007] In order to solve the above technical problems, an embodiment of the present invention provides a method and device for determining a soil moisture sensor network deployment strategy to solve the technical problem in the prior art that the location and number of sensor layouts in a watershed are not accurate enough due to the inability to fully consider complex geographical and soil properties.

[0008] A first aspect of an embodiment of the present invention provides a method for determining a soil moisture sensor network deployment strategy, the method comprising:

[0009] Extract remote sensing data of the area to be deployed to obtain the shortwave infrared transform reflectance and normalized vegetation index of the area to be deployed;

[0010] Based on the short-wave infrared transform reflectance and the normalized vegetation index, a pixel distribution relationship diagram is constructed, and the dry edge and the wet edge are obtained according to the pixel distribution relationship diagram. Based on the dry edge and the wet edge, the soil moisture is inverted to obtain the soil moisture content of the area to be deployed;

[0011] Collect digital elevation data of the area to be deployed, obtain feature data of each pixel point according to the digital elevation data, soil texture and vegetation cover type data, determine the initial number of cluster centers, obtain the initial number, and obtain the likelihood function under given model parameters based on the feature data of each pixel point, wherein the feature data include latitude, longitude, slope, aspect, altitude, potential relative radiation, soil texture and vegetation cover type;

[0012] According to the characteristic data of each pixel point, a prior distribution under given model parameters is obtained, the prior distribution and the likelihood function are multiplied to obtain the posterior distribution under given model parameters, the posterior distribution is estimated by numerical method to obtain an estimated value, the position of each cluster center is determined according to the estimated value, and according to the position of each cluster center, a nearest neighbor search is used to obtain a pixel point whose distance to each cluster center is less than a preset value, and the pixel point is used as a representative sample point, wherein the given model parameters include a mixing coefficient, a mean and a covariance matrix;

[0013] The Gaussian process regression model is used for modeling to obtain the covariance model. The characteristic data of each representative sample point is input into the covariance model for prediction to obtain the predicted soil moisture content of the representative sample point. The root mean square error between the predicted soil moisture content and the soil moisture content obtained by soil moisture inversion is calculated. If the root mean square error is still greater than the preset threshold, the process proceeds to the step of determining the initial number of cluster centers, the initial number of cluster centers is increased, and the updated initial number is obtained. The positions of each updated cluster center and the new root mean square error are continued to be calculated based on the updated initial number until the new root mean square error is less than the preset threshold, then the corresponding updated initial number is determined as the final number of cluster centers, the final number of cluster centers is determined to be the number of sensors to be deployed, and the positions of the updated cluster centers are the positions of the sensors to be deployed.

[0014] In a possible implementation manner of the first aspect, a dry edge and a wet edge are obtained according to a pixel distribution relationship diagram, and soil moisture inversion is performed based on the dry edge and the wet edge to obtain soil moisture content in the area to be deployed, including:

[0015] Based on the shortwave infrared transform reflectance and normalized vegetation index, the pixel distribution relationship diagram is obtained by linear regression calculation to obtain the dry edge and wet edge;

[0016] Based on the dry edge and wet edge, the soil moisture is inverted using the thermo-optical trapezoidal model to obtain the soil moisture content in the area to be deployed.

[0017] In a possible implementation manner of the first aspect, a calculation formula for the slope is:

[0018]

[0019] In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis, Z(x,y) is the elevation value, indicating the elevation at the pixel point (x,y), dx and dy are the vertical resolutions of the digital elevation data on the x-axis and y-axis respectively;

[0020] The calculation formula for slope aspect is:

[0021]

[0022] In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis;

[0023] The formula for calculating the potential relative radiation is:

[0024] PRR=I sc ·cos(α)·cos(θ)·cos(φ)

[0025] In the formula, I sc is the solar constant, α is the solar altitude angle, which represents the angle between the sun and the horizon, θ is the angle between the slope and the solar altitude angle, and φ is the angle between the slope direction and the solar azimuth.

[0026] In a possible implementation manner of the first aspect, obtaining a likelihood function under given model parameters based on feature data of each pixel point includes:

[0027] Based on the feature data of each pixel point, the feature space range of each pixel point is represented by the multivariate normal distribution, and the multivariate normal distribution probability density function is obtained, where the expression of the multivariate normal distribution probability density function is:

[0028]

[0029] In the formula, μ is the expected value, Σ is the covariance matrix, and x is the feature data of each pixel;

[0030] The probability density of the Gaussian mixture model of each pixel is obtained according to the multivariate normal distribution probability density function of each pixel, where the expression of the probability density of the Gaussian mixture model is:

[0031]

[0032] In the formula, μ is the expected value, Σ is the covariance matrix, π m is the mixing coefficient of the mth Gaussian component, x is the characteristic data of each pixel, μ m is the mean of the mth Gaussian component, Σ m is the covariance matrix of the mth Gaussian component, and p(x) is the probability density function of the Gaussian mixture model of each pixel;

[0033] According to the probability density of the Gaussian mixture model of each pixel, the likelihood function under given model parameters is obtained, where the likelihood function is:

[0034]

[0035] In the formula, p(x n |π,μ,Σ) represents the probability of the nth pixel under given model parameters, is the multivariate normal distribution of the mth Gaussian component, π m is the mixing coefficient of the mth Gaussian component, μ m is the mean of the mth Gaussian component, ∑ m is the covariance matrix of the mth Gaussian component, N is the number of pixels, and M is the number of categories in the Gaussian mixture model.

[0036] In a possible implementation of the first aspect, calculating a root mean square error between predicted soil moisture content and soil moisture content obtained by soil moisture inversion includes:

[0037] Calculate the root mean square error between the predicted soil moisture content and the soil moisture content obtained by soil moisture inversion. The root mean square error calculation formula is:

[0038]

[0039] In the formula, n represents the number of sample points, y i ′ is the predicted soil moisture content of the representative sample point, y i The soil moisture content obtained by soil moisture inversion.

[0040] In order to solve the same technical problem, a second aspect of an embodiment of the present invention provides a soil moisture sensor network deployment strategy determination device, the device comprising:

[0041] An extraction module is used to extract remote sensing data of the area to be deployed, and obtain the short-wave infrared transform reflectance and normalized vegetation index of the area to be deployed;

[0042] An inversion module is used to construct a pixel distribution relationship diagram based on shortwave infrared transform reflectance and normalized vegetation index, obtain dry edges and wet edges based on the pixel distribution relationship diagram, perform soil moisture inversion based on the dry edges and wet edges, and obtain soil moisture content in the area to be deployed;

[0043] The first calculation module is used to collect digital elevation data of the area to be deployed, obtain characteristic data of each pixel point according to the digital elevation data and the soil texture and vegetation cover type data, determine the initial number of cluster centers, obtain the initial number, and obtain the likelihood function under given model parameters based on the characteristic data of each pixel point, wherein the characteristic data includes latitude, longitude, slope, slope aspect, altitude, potential relative radiation, soil texture and vegetation cover type;

[0044] The second calculation module is used to obtain a prior distribution under given model parameters according to the feature data of each pixel point, multiply the prior distribution and the likelihood function to obtain a posterior distribution under given model parameters, estimate the posterior distribution using a numerical method to obtain an estimated value, determine the position of each cluster center according to the estimated value, and use the nearest neighbor search to obtain pixel points whose distances to each cluster center are less than a preset value according to the position of each cluster center, and use the pixel points as representative sample points, wherein the given model parameters include a mixing coefficient, a mean, and a covariance matrix;

[0045] The third calculation module is used to use the Gaussian process regression model to build a model, obtain a covariance model, input the characteristic data of each representative sample point into the covariance model for prediction, obtain the predicted soil moisture content of the representative sample point, calculate the root mean square error between the predicted soil moisture content and the soil moisture content obtained by soil moisture inversion, if the root mean square error is still greater than the preset threshold, then go to the step of determining the initial number of cluster centers, increase the initial number of cluster centers, obtain the updated initial number, and continue to calculate the positions of each updated cluster center and the new root mean square error based on the updated initial number, until the new root mean square error obtained is less than the preset threshold, determine the corresponding updated initial number as the final number of cluster centers, determine the final number of cluster centers as the number of sensors to be deployed, and the positions of the updated cluster centers as the positions of the sensors to be deployed.

[0046] In a possible implementation manner of the second aspect, the inversion module includes a linear regression calculation unit and a soil moisture inversion unit, wherein:

[0047] The linear regression calculation unit is used to obtain the dry edge and the wet edge by obtaining the pixel distribution relationship diagram based on the short-wave infrared transform reflectance and the normalized vegetation index through linear regression calculation;

[0048] The soil moisture inversion unit is used to invert soil moisture based on the dry edge and wet edge using the thermo-optical trapezoidal model to obtain the soil moisture content in the area to be deployed.

[0049] In a possible implementation manner of the second aspect, a calculation formula for the slope is:

[0050]

[0051] In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis, Z(x,y) is the elevation value, indicating the elevation at the pixel point (x,y), dx and dy are the vertical resolutions of the digital elevation data on the x-axis and y-axis respectively;

[0052] The calculation formula for slope aspect is:

[0053]

[0054] In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis;

[0055] The formula for calculating the potential relative radiation is:

[0056] PRR=I sc ·cos(α)·cos(θ)·cos(φ)

[0057] In the formula, I sc is the solar constant, α is the solar altitude angle, which represents the angle between the sun and the horizon, θ is the angle between the slope and the solar altitude angle, and φ is the angle between the slope direction and the solar azimuth.

[0058] A third aspect of an embodiment of the present invention provides a computer device, including:

[0059] Memory for storing computer programs;

[0060] A processor is used to implement the steps of the method for determining the soil moisture sensor network deployment strategy according to the first aspect when executing the computer program.

[0061] A fourth aspect of an embodiment of the present invention provides a storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the method for determining a soil moisture sensor network deployment strategy according to the first aspect are implemented.

[0062] The technical solution of the present invention has the following advantages:

[0063] The method for determining the network deployment strategy of soil moisture sensors provided by the embodiment of the present invention uses soil moisture data inverted from microwave remote sensing data as reference data, then obtains characteristic data of each pixel point based on the collected digital elevation data of the area to be deployed, clusters the characteristic data of each pixel point to obtain the position of the representative sample point, and then uses the Gaussian process regression model to perform modeling, combines the obtained position of the representative sample point with the inverted soil moisture data for evaluation and screening, thereby determining the optimal deployment position and number of sensors. The number and position of sensors selected by the above method have high accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0065] Figure 1 It is a flow chart of a method for determining a soil moisture sensor network deployment strategy in an embodiment of the present invention;

[0066] Figure 2 A pixel point distribution relationship diagram of a method for determining a soil moisture sensor network deployment strategy in an embodiment of the present invention;

[0067] Figure 3 A flow chart of determining a network deployment strategy of a method for determining a network deployment strategy for a soil moisture sensor in an embodiment of the present invention;

[0068] Figure 4 It is a structural block diagram of a device for determining a soil moisture sensor network deployment strategy in an embodiment of the present invention. DETAILED DESCRIPTION

[0069] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0070] In the description of the present invention, it should be noted that the terms "first", "second" and "third" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0071] The soil moisture sensor network deployment strategy determination method provided by the embodiment of the present invention is as follows: Figure 1 As shown, Figure 1 The flow chart of the method for determining the deployment strategy of the soil moisture sensor network includes steps S101 to S103, and each step is specifically as follows:

[0072] S101: extracting remote sensing data of the area to be deployed, and obtaining shortwave infrared transform reflectance and normalized vegetation index of the area to be deployed.

[0073] In this embodiment, the shortwave infrared transform reflectance STR and the normalized vegetation index NDVI of the area to be deployed are extracted through the band reflectance data of remote sensing satellites (such as Landsat, Sentinel, etc.), and the calculation formulas are:

[0074]

[0075] In the formula, R SWIR , R Red , R NIR They are the reflectance of shortwave infrared, red light and near infrared bands of remote sensing data respectively, STR is the shortwave infrared transformation reflectance, and NDVI is the normalized difference vegetation index.

[0076] S102: construct a pixel distribution relationship diagram based on the shortwave infrared transform reflectance and the normalized vegetation index, obtain a dry edge and a wet edge according to the pixel distribution relationship diagram, perform soil moisture inversion based on the dry edge and the wet edge, and obtain the soil moisture content of the area to be deployed.

[0077] In this embodiment, a relationship diagram between the shortwave infrared transform reflectance STR and the normalized difference vegetation index NDVI is drawn, and the spatial distribution of STR-NDVI forms a trapezoid, such as Figure 2 As shown, the point with a high STR value is the wet edge, and the point with a low STR value is the dry edge.

[0078] According to the Kubelka-Munk bidirectional radiation model, we can get:

[0079]

[0080] Where K and S are the light absorption coefficient and scattering coefficient of the layer, r is the transformed reflectivity, and R SWIR is the short-wave infrared reflectivity, and the pixels with the same NDVI value represent a fixed vegetation coverage ratio FVC, then:

[0081] K=K 0 +K water θ(1-FVC)+K water SM FVC

[0082] S=S 0 +S water θ(1-FVC)+S water SM FVC

[0083] In the formula, K 0 , S 0 , K water , S water Corresponding to the light absorption coefficient and scattering coefficient of bare soil and water, respectively, θ is the soil moisture content in the root zone, and SM is the surface soil moisture content, so:

[0084]

[0085] In the short-wave infrared (SWIR) wavelength range (Swater ≈0) then:

[0086]

[0087] In the formula, r is the transformed reflectivity, K water is the scattering coefficient of bare soil;

[0088] Then, the wet edge and dry edge corresponding to the pixel distribution relationship diagram are calculated by linear regression. According to the wet edge and dry edge, the thermal-optical trapezoidal model (OPTRAM) model is used to perform soil moisture (SM) inversion to obtain the soil moisture content of the area to be deployed.

[0089] In one embodiment, a dry edge and a wet edge are obtained according to a pixel point distribution relationship diagram, and soil moisture inversion is performed based on the dry edge and the wet edge to obtain soil moisture content in the area to be deployed, including:

[0090] Based on the shortwave infrared transform reflectance and normalized vegetation index, the pixel distribution relationship diagram is obtained by linear regression calculation to obtain the dry edge and wet edge;

[0091] Based on the dry edge and wet edge, the soil moisture is inverted using the thermo-optical trapezoidal model to obtain the soil moisture content in the area to be deployed.

[0092] In this embodiment, the moisture content SM and STR at different NDVI values ​​have a linear relationship, and the STR of the wet edge corresponding to the pixel point distribution relationship diagram can be calculated by linear regression. d and dry side STR w , and the calculation formulas are:

[0093] STR d =i d +s d NDVI

[0094] STR w =i w +s w NDVI

[0095] In the formula, i d 、i w is the intercept of the linear regression equation of the dry side and the wet side, s d 、s w is the slope of the equation for the linear regression of the dry and wet sides.

[0096] Based on the linear relationship between soil surface moisture content and shortwave infrared reflectance (STR), soil moisture content SM can be calculated by the following equation:

[0097]

[0098] Where STR is the true shortwave infrared transform reflectance under a certain NDVI condition.

[0099] In summary, the formula for inverting soil moisture (SM) using the thermal-optical trapezoidal model (OPTRAM) is:

[0100]

[0101] In the formula, i d 、i w is the intercept of the linear regression equation of the dry side and the wet side, s d 、s w is the slope of the equation for the linear regression of the dry and wet sides.

[0102] S103: Collect digital elevation data of the area to be deployed, obtain feature data of each pixel point according to the digital elevation data, soil texture and vegetation cover type data, determine the initial number of cluster centers, obtain the initial number, and obtain the likelihood function under given model parameters based on the feature data of each pixel point, wherein the feature data includes latitude, longitude, slope, aspect, altitude, potential relative radiation, soil texture and vegetation cover type.

[0103] In this embodiment, the raster data with a high resolution of 30m at the watershed scale is collected, mainly including digital elevation data (DEM). The eight types of watershed characteristics of soil pixels are further calculated based on the digital elevation data, soil texture and vegetation cover type data: latitude, longitude, slope, aspect, elevation, potential relative radiation, and soil texture and vegetation cover type data are collected at the same time. The eight types of features of each soil pixel are clustered using the Gaussian mixture model (GMM). First, the number of cluster centers is set. The range of the feature space observed by each sensor is represented by a multivariate normal distribution. The feature space of each feature sensor is (

[0104] In one embodiment, the slope is calculated as follows:

[0105]

[0106] In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis, Z(x,y) is the elevation value, indicating the elevation at the pixel point (x,y), dx and dy are the vertical resolutions of the digital elevation data on the x-axis and y-axis respectively;

[0107] The calculation formula for slope aspect is:

[0108]

[0109] In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis;

[0110] The formula for calculating the potential relative radiation is:

[0111] PRR=I sc ·cos(α)·cos(θ)·cos(φ)

[0112] In the formula, I sc is the solar constant, α is the solar altitude angle, which represents the angle between the sun and the horizon, θ is the angle between the slope and the solar altitude angle, and φ is the angle between the slope direction and the solar azimuth.

[0113] In this embodiment, the slope S and the slope direction can be calculated based on high-resolution digital elevation data, and the formulas are:

[0114]

[0115]

[0116] In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis, Z(x,y) is the elevation value, indicating the elevation at the pixel point (x,y), dx and dy are the vertical resolutions of the digital elevation data on the x-axis and y-axis respectively;

[0117] The calculation formula for slope aspect is:

[0118]

[0119] In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis;

[0120] The formula for calculating the potential relative radiation is:

[0121] PRR=I sc ·cos(α)·cos(θ)·cos(φ)

[0122] In the formula, I sc is the solar constant, α is the solar altitude angle, which represents the angle between the sun and the horizon, θ is the angle between the slope and the solar altitude angle, and φ is the angle between the slope direction and the solar azimuth.

[0123] In one embodiment, based on the feature data of each pixel point, a likelihood function under given model parameters is obtained, including:

[0124] Based on the feature data of each pixel point, the feature space range of each pixel point is represented by the multivariate normal distribution, and the multivariate normal distribution probability density function is obtained, where the expression of the multivariate normal distribution probability density function is:

[0125]

[0126] In the formula, μ is the expected value, ∑ is the covariance matrix, and x is the feature data of each pixel;

[0127] The probability density of the Gaussian mixture model of each pixel is obtained according to the multivariate normal distribution probability density function of each pixel, where the expression of the probability density of the Gaussian mixture model is:

[0128]

[0129] In the formula, μ is the expected value, ∑ is the covariance matrix, and π m is the mixing coefficient of the mth Gaussian component, x is the characteristic data of each pixel, μ m is the mean of the mth Gaussian component, ∑ m is the covariance matrix of the mth Gaussian component, and p(x) is the probability density function of the Gaussian mixture model of each pixel;

[0130] According to the probability density of the Gaussian mixture model of each pixel, the likelihood function under given model parameters is obtained, where the likelihood function is:

[0131]

[0132] In the formula, p(x n |π,μ,Σ) is the probability of the nth pixel under given model parameters, is the multivariate normal distribution of the mth Gaussian component, π m is the mixing coefficient of the mth Gaussian component, μ m is the mean of the mth Gaussian component, ∑ m is the covariance matrix of the mth Gaussian component, N is the number of pixels, and M is the number of categories in the Gaussian mixture model.

[0133] In this embodiment, the number of cluster centers is a hyperparameter set by the model, representing the number of basin feature clusters and the number of preliminarily selected sensors. Each sensor observes the feature space ( The range of is represented by the multivariate normal distribution, and the expression of the probability density function of the multivariate normal distribution is:

[0134]

[0135] In the formula, μ is the expected value, Σ is the covariance matrix, and x is the feature data of each pixel:

[0136] Then, after obtaining the multivariate normal distribution probability density function of each pixel point, the probability density of the Gaussian mixture model of each pixel point is obtained, where the expression of the probability density of the Gaussian mixture model is:

[0137]

[0138] In the formula, μ is the expected value, Σ is the covariance matrix, π m is the mixing coefficient of the mth Gaussian component, x is the characteristic data of each pixel, μ m is the mean of the mth Gaussian component, Σ m is the covariance matrix of the mth Gaussian component, and p(x) is the probability density function of the Gaussian mixture model of each pixel;

[0139] Then, according to the likelihood function, the probability of the data of the known sample points appearing is calculated as:

[0140]

[0141] In the formula, p(x n |π,μ,Σ) is the probability of the nth pixel under given model parameters, is the multivariate normal distribution of the mth Gaussian component, π m is the mixing coefficient of the mth Gaussian component, μ m is the mean of the mth Gaussian component, ∑ m is the covariance matrix of the mth Gaussian component, N is the number of pixels, and M is the number of categories in the Gaussian mixture model.

[0142] S104: Based on the feature data of each pixel point, a prior distribution under given model parameters is obtained, the prior distribution and the likelihood function are multiplied to obtain the posterior distribution under given model parameters, the posterior distribution is estimated using a numerical method to obtain an estimated value, the position of each cluster center is determined according to the estimated value, and based on the position of each cluster center, a nearest neighbor search is used to obtain pixel points whose distance to each cluster center is less than a preset value, and the pixel points are used as representative sample points, wherein the given model parameters include a mixing coefficient, a mean, and a covariance matrix.

[0143] In this embodiment, a given model parameter (such as π m , μ m ,Σ m ), given the data to be clustered, according to Bayes' theorem, the posterior distribution is the product of the likelihood function and the prior distribution, thus obtaining the posterior distribution of the model parameters, that is:

[0144]

[0145] In the formula, μ is the expected value, ∑ is the covariance matrix, and x is the feature data of each pixel point.

[0146] During the calculation process, numerical methods are used to estimate the posterior distribution and determine the cluster centers.

[0147] For each cluster center μ m , using the nearest neighbor search, find the point (x nearest ), as the representative sample point, the calculation formula is:

[0148]

[0149] Among them, ∥x i -q∥isx i The distance from the query point q is measured using the Euclidean distance metric. In this case, D represents an 8-dimensional vector, x nearest Representative sample points.

[0150] S105: Use the Gaussian process regression model to build a model and obtain a covariance model. Input the characteristic data of each representative sample point into the covariance model for prediction to obtain the predicted soil moisture content of the representative sample point. Calculate the root mean square error between the predicted soil moisture content and the soil moisture content obtained by soil moisture inversion. If the root mean square error is still greater than the preset threshold, go to the step of determining the initial number of cluster centers, increase the initial number of cluster centers, and obtain an updated initial number. Based on the updated initial number, continue to calculate the positions of each updated cluster center and the new root mean square error, until the new root mean square error is less than the preset threshold, determine the corresponding updated initial number as the final number of cluster centers, determine the final number of cluster centers as the number of sensors to be deployed, and the positions of the updated cluster centers as the positions of the sensors to be deployed.

[0151] In this embodiment, the soil moisture data of the primary representative sample points are used to estimate the soil moisture distribution (i.e., soil moisture content) of the entire watershed through the Gaussian process (GP). Specifically, a covariance-based model is established using the Gaussian process, using the known input-output relationship and the training data of the sample point locations. The prediction can be obtained by inputting the soil moisture and uncertainty of the new location. Using the watershed characteristic variables slope, aspect, elevation, potential relative radiation Used as the dependent variable in the estimation, the formula is:

[0152] Y~GP(m(x),k(x,x′))

[0153]

[0154] Where Y represents the soil moisture predicted by the covariance model, the Gaussian process GP is determined by a mean function m(x) and a covariance function k(x, x′), σ 2 is the variance of all representative samples, and l is the length scale of all representative samples.

[0155] The error in the Gaussian process model is estimated by the root mean square error (RMSE) between the predicted soil moisture and the inverted soil moisture data of the area to be deployed. If the minimum root mean square error is not found, the number of sensors is increased in sequence, and the process of S103-S104 is repeated to calculate the root mean square error (RMSE) after increasing the number of representative sample points. The number of representative sample points corresponding to the minimum root mean square error is the most reasonable number of sensors. At this time, the distribution position is determined by the nearest neighbor search to determine the latitude and longitude coordinates of the representative sample points, which is the optimal deployment position.

[0156] In one embodiment, calculating the root mean square error between the predicted soil moisture content and the soil moisture content obtained by performing soil moisture inversion includes:

[0157] Calculate the root mean square error between the predicted soil moisture content and the soil moisture content obtained by soil moisture inversion. The root mean square error calculation formula is:

[0158]

[0159] In the formula, n represents the number of sample points, y i ′ is the predicted soil moisture content of the representative sample point, y i The soil moisture content obtained by soil moisture inversion.

[0160] In this embodiment, the root mean square error between the predicted soil moisture content and the soil moisture content obtained by soil moisture inversion is estimated, and the calculation formula of the root mean square error is:

[0161]

[0162] In the formula, n represents the number of sample points, y i ′ is the predicted soil moisture content of the representative sample point, y i The soil moisture content obtained by soil moisture inversion.

[0163] The method for determining the deployment strategy of the soil moisture sensor network proposed in the present invention adopts the soil moisture inverted by the OPTRAM model based on remote sensing information, and there is no need to arrange experimental sites in the basin in advance, saving a lot of manpower costs, and there is no need to measure sample soil moisture data for a period of time, saving time costs. In addition, clustering is performed based on multi-dimensional basin scale characteristic data, and representative soil moisture conditions at the basin scale are preferably deployed; soil moisture data inverted by microwave remote sensing is introduced to verify the effect of the machine learning model (Gaussian process regression), and then the optimal number of sensors is evaluated.

[0164] By using an unsupervised clustering algorithm (Gaussian mixture model) to screen representative samples and then using a supervised machine learning model (Gaussian process regression) for simulation verification, the accuracy of sensor location and quantity is improved.

[0165] The soil moisture sensor network deployment strategy determination device provided by the embodiment of the present invention is as follows: Figure 4 As shown, Figure 4 Determine the device block diagram for the soil moisture sensor network deployment strategy, including:

[0166] An extraction module 401 is used to extract remote sensing data of the area to be deployed, and obtain the short-wave infrared transform reflectance and normalized vegetation index of the area to be deployed;

[0167] An inversion module 402 is used to construct a pixel distribution relationship diagram based on the shortwave infrared transform reflectance and the normalized vegetation index, obtain a dry edge and a wet edge according to the pixel distribution relationship diagram, perform soil moisture inversion based on the dry edge and the wet edge, and obtain the soil moisture content of the area to be deployed;

[0168] The first calculation module 403 is used to collect digital elevation data of the area to be deployed, obtain feature data of each pixel point according to the digital elevation data, soil texture and vegetation cover type data, determine the initial number of cluster centers, obtain the initial number, and obtain the likelihood function under given model parameters based on the feature data of each pixel point, wherein the feature data includes latitude, longitude, slope, aspect, altitude, potential relative radiation, soil texture and vegetation cover type;

[0169] The second calculation module 404 is used to obtain a prior distribution under given model parameters according to the feature data of each pixel point, multiply the prior distribution and the likelihood function to obtain a posterior distribution under the given model parameters, estimate the posterior distribution using a numerical method to obtain an estimated value, determine the position of each cluster center according to the estimated value, and use the nearest neighbor search to obtain a pixel point whose distance from each cluster center is less than a preset value according to the position of each cluster center, and use the pixel point as a representative sample point, wherein the given model parameters include a mixing coefficient, a mean, and a covariance matrix;

[0170] The third calculation module 405 is used to use the Gaussian process regression model to build a model, obtain a covariance model, input the characteristic data of each representative sample point into the covariance model for prediction, obtain the predicted soil moisture content of the representative sample point, calculate the root mean square error between the predicted soil moisture content and the soil moisture content obtained by soil moisture inversion, and if the root mean square error is still greater than the preset threshold, then go to the step of determining the initial number of cluster centers, increase the initial number of cluster centers, obtain an updated initial number, and continue to calculate the positions of each updated cluster center and the new root mean square error based on the updated initial number, until the new root mean square error obtained is less than the preset threshold, determine the corresponding updated initial number as the final number of cluster centers, determine the final number of cluster centers as the number of sensors to be deployed, and the positions of the updated cluster centers as the positions of the sensors to be deployed.

[0171] In one embodiment, the inversion module 402 includes a linear regression calculation unit and a soil moisture inversion unit.

[0172] in,

[0173] The linear regression calculation unit is used to obtain the dry edge and the wet edge by obtaining the pixel distribution relationship diagram based on the short-wave infrared transform reflectance and the normalized vegetation index through linear regression calculation;

[0174] The soil moisture inversion unit is used to invert soil moisture based on the dry edge and wet edge using the thermo-optical trapezoidal model to obtain the soil moisture content in the area to be deployed.

[0175] In one embodiment, the slope is calculated as follows:

[0176]

[0177] In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis, Z(x,y) is the elevation value, indicating the elevation at the pixel point (x,y), dx and dy are the vertical resolutions of the digital elevation data on the x-axis and y-axis respectively;

[0178] The calculation formula for slope aspect is:

[0179]

[0180] In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis;

[0181] The formula for calculating the potential relative radiation is:

[0182] PRR=I sc ·cos(α)·cos(θ)·cos(φ)

[0183] In the formula, I sc is the solar constant, α is the solar altitude angle, which represents the angle between the sun and the horizon, θ is the angle between the slope and the solar altitude angle, and φ is the angle between the slope direction and the solar azimuth.

[0184] The specific implementation of the device for determining a soil moisture sensor network deployment strategy is basically the same as the specific implementation of the method for determining a soil moisture sensor network deployment strategy described above, and will not be repeated here.

[0185] In one embodiment of the present application, a computer device is provided, which includes a memory and a processor, wherein a computer program is stored in the memory, and the above steps are implemented when the processor executes the computer program; the computer device provided in this embodiment has an implementation principle and technical effects similar to those of the above method embodiments, and will not be repeated here.

[0186] In one embodiment of the present application, a computer-readable storage medium is provided, on which a computer program is stored, and the above steps are implemented when the computer program is executed by a processor; the computer-readable storage medium provided in this embodiment has an implementation principle and technical effects similar to those of the above method embodiment, and will not be repeated here.

[0187] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0188] The specific embodiments described above further illustrate the purpose, technical solutions and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for determining a soil moisture sensor network deployment strategy, characterized in that: include: Extracting remote sensing data of the area to be deployed to obtain shortwave infrared transform reflectance and normalized vegetation index of the area to be deployed; A pixel point distribution relationship diagram is constructed based on the shortwave infrared transform reflectivity and the normalized difference vegetation index, a dry edge and a wet edge are obtained according to the pixel point distribution relationship diagram, and soil moisture is inverted based on the dry edge and the wet edge to obtain soil moisture content of the area to be deployed; Collect digital elevation data of the area to be deployed, obtain feature data of each pixel point according to the digital elevation data, soil texture and vegetation cover type data, determine the initial number of cluster centers, obtain the initial number, and obtain the likelihood function under given model parameters based on the feature data of each pixel point, wherein the feature data includes latitude, longitude, slope, aspect, altitude, potential relative radiation, soil texture and vegetation cover type; According to the characteristic data of each pixel point, a prior distribution under given model parameters is obtained, the prior distribution and the likelihood function are multiplied to obtain a posterior distribution under given model parameters, the posterior distribution is estimated by a numerical method to obtain an estimated value, the position of each cluster center is determined according to the estimated value, and according to the position of each cluster center, a nearest neighbor search is used to obtain a pixel point whose distance to each cluster center is less than a preset value, and the pixel point is used as a representative sample point, wherein the given model parameters include a mixing coefficient, a mean, and a covariance matrix; The Gaussian process regression model is used for modeling to obtain a covariance model, and the characteristic data of each representative sample point is input into the covariance model for prediction to obtain the predicted soil moisture content of the representative sample point, and the root mean square error between the predicted soil moisture content and the soil moisture content obtained by soil moisture inversion is calculated. If the root mean square error is still greater than the preset threshold, the step of determining the initial number of cluster centers is performed, and the initial number of cluster centers is increased to obtain an updated initial number. The positions of each updated cluster center and the new root mean square error are continued to be calculated based on the updated initial number until the new root mean square error is less than the preset threshold, and the corresponding updated initial number is determined as the final number of cluster centers, and the final number of cluster centers is determined as the number of sensors to be deployed, and the positions of the updated cluster centers are the positions of sensors to be deployed.

2. The method for determining a soil moisture sensor network deployment strategy according to claim 1, wherein: The obtaining of the dry edge and the wet edge according to the pixel point distribution relationship diagram, performing soil moisture inversion based on the dry edge and the wet edge, and obtaining the soil moisture content of the area to be deployed includes: Based on the shortwave infrared transform reflectance and the normalized vegetation index, the pixel point distribution relationship diagram is obtained by linear regression calculation to obtain the dry edge and the wet edge; Based on the dry edge and the wet edge, the soil moisture is inverted using the thermo-optical trapezoidal model to obtain the soil moisture content of the area to be deployed.

3. The method for determining a soil moisture sensor network deployment strategy according to claim 1, wherein: The calculation formula of the slope is: In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis, Z(x,y) is the elevation value, indicating the elevation at the pixel point (x,y), dx and dy are the vertical resolutions of the digital elevation data on the x-axis and y-axis respectively; The calculation formula of the slope aspect is: In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis; The calculation formula for the potential relative radiation is: PRR=I sc ·cos(α)·cos(θ)·cos(φ) In the formula, I sc is the solar constant, α is the solar altitude angle, which represents the angle between the sun and the horizon, θ is the angle between the slope and the solar altitude angle, and φ is the angle between the slope direction and the solar azimuth.

4. The method for determining a soil moisture sensor network deployment strategy according to claim 1, wherein: The obtaining of a likelihood function under given model parameters based on the feature data of each pixel point includes: Based on the feature data of each pixel point, the feature space range of each pixel point is represented by a multivariate normal distribution to obtain a multivariate normal distribution probability density function, wherein the expression of the multivariate normal distribution probability density function is: Wherein, μ is the expected value, Σ is the covariance matrix, and x is the feature data of each pixel point; The probability density of the Gaussian mixture model of each pixel point is obtained according to the multivariate normal distribution probability density function of each pixel point, wherein the expression of the probability density of the Gaussian mixture model is: In the formula, μ is the expected value, Σ is the covariance matrix, π m is the mixing coefficient of the mth Gaussian component, x is the characteristic data of each pixel, μ m is the mean of the mth Gaussian component, ∑ m is the covariance matrix of the mth Gaussian component, and p(x) is the probability density function of the Gaussian mixture model of each of the pixels; According to the probability density of the Gaussian mixture model of each pixel point, the likelihood function under given model parameters is obtained, wherein the likelihood function is: In the formula, p(x n |π,μ,Σ) is the probability of the nth pixel under given model parameters, is the multivariate normal distribution of the mth Gaussian component, π m is the mixing coefficient of the mth Gaussian component, μ m is the mean of the mth Gaussian component, ∑ m is the covariance matrix of the mth Gaussian component, N is the number of pixels, and M is the number of categories in the Gaussian mixture model.

5. The method for determining a soil moisture sensor network deployment strategy according to claim 1, wherein: The calculating the root mean square error of the soil moisture content of the representative sample point and the soil moisture content distribution data comprises: The root mean square error between the predicted soil moisture content and the soil moisture content obtained by performing soil moisture inversion is calculated. The calculation formula of the root mean square error is: In the formula, n represents the number of sample points, y i ′ is the predicted soil moisture content of the representative sample point, y i The soil moisture content obtained by soil moisture inversion.

6. A soil moisture sensor network deployment strategy determination device, characterized in that: include: An extraction module is used to extract remote sensing data of the area to be deployed, and obtain the short-wave infrared transform reflectivity and normalized vegetation index of the area to be deployed; An inversion module is used to construct a pixel point distribution relationship diagram based on the shortwave infrared transform reflectivity and the normalized vegetation index, obtain a dry edge and a wet edge according to the pixel point distribution relationship diagram, perform soil moisture inversion based on the dry edge and the wet edge, and obtain soil moisture content in the area to be deployed; The first calculation module is used to collect digital elevation data of the area to be deployed, obtain characteristic data of each pixel point according to the digital elevation data, soil texture and vegetation cover type, determine the initial number of cluster centers, obtain the initial number, and obtain the likelihood function under given model parameters based on the characteristic data of each pixel point, wherein the characteristic data includes latitude, longitude, slope, aspect, altitude, potential relative radiation, soil texture and vegetation cover type; A second calculation module is used to obtain a prior distribution under given model parameters according to the feature data of each pixel point, multiply the prior distribution and the likelihood function to obtain a posterior distribution under given model parameters, estimate the posterior distribution using a numerical method to obtain an estimated value, determine the position of each cluster center according to the estimated value, and use a nearest neighbor search to obtain a pixel point whose distance to each cluster center is less than a preset value according to the position of each cluster center, and use the pixel point as a representative sample point, wherein the given model parameters include a mixing coefficient, a mean, and a covariance matrix; The third calculation module is used to use the Gaussian process regression model to build a model to obtain a covariance model, input the characteristic data of each representative sample point into the covariance model for prediction, obtain the predicted soil moisture content of the representative sample point, calculate the root mean square error between the predicted soil moisture content and the soil moisture content obtained by soil moisture inversion, if the root mean square error is still greater than the preset threshold, then go to the step of determining the initial number of cluster centers, increase the initial number of cluster centers, obtain the updated initial number, continue to calculate the position of each updated cluster center and the new root mean square error based on the updated initial number, until the new root mean square error obtained is less than the preset threshold, determine the corresponding updated initial number as the final number of cluster centers, determine the final number of cluster centers as the number of sensors to be deployed, and the position of the updated cluster center is the position of the sensor to be deployed.

7. The method for determining a soil moisture sensor network deployment strategy according to claim 6, wherein: The inversion module includes a linear regression calculation unit and a soil moisture inversion unit, wherein: The linear regression calculation unit is used to obtain the pixel point distribution relationship diagram by linear regression calculation based on the short-wave infrared transform reflectance and the normalized vegetation index to obtain the dry edge and the wet edge; The soil moisture inversion unit is used to invert soil moisture based on the dry edge and the wet edge using a thermo-optical trapezoidal model to obtain soil moisture content in the area to be deployed.

8. The method for determining a soil moisture sensor network deployment strategy according to claim 6, wherein: The slope calculation formula is: In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis, Z(x,y) is the elevation value, indicating the elevation at the pixel point (x,y), dx and dy are the vertical resolutions of the digital elevation data on the x-axis and y-axis respectively; The calculation formula of the slope aspect is: In the formula, is the elevation change rate of the pixel point on the x-axis, is the elevation change rate of the pixel point on the y-axis; The calculation formula for the potential relative radiation is: PRR=I sc ·cos(α)·cos(θ)·cos(φ) In the formula, I sc is the solar constant, α is the solar altitude angle, which represents the angle between the sun and the horizon, θ is the angle between the slope and the solar altitude angle, and φ is the angle between the slope direction and the solar azimuth.

9. A computer device, characterized in that: include: Memory for storing computer programs; A processor, configured to implement the steps of the method for determining a soil moisture sensor network deployment strategy as described in any one of claims 1 to 5 when executing the computer program.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for determining a soil moisture sensor network deployment strategy according to any one of claims 1 to 5 are implemented.

Images