Method for detecting dynamic salt content of orchard soil in saline-alkali soil based on multi-source sensor fusion

Abstract

The application discloses a salt and alkali soil orchard soil salt dynamic detection method based on multi-source sensor fusion, and belongs to the technical field of soil salt detection. The method comprises the following steps: acquiring visual conductivity data of multiple depth layers of an orchard target area based on a multi-frequency electromagnetic induction sensor array; synchronously collecting monitoring data of a soil moisture sensor and a temperature sensor of the target area; converting the visual conductivity data and the monitoring data into initial salt estimation values of different depth layers of the target area according to a pre-established multi-element nonlinear mapping relationship among soil salt, visual conductivity, soil moisture and temperature; performing space-time interpolation on the initial salt estimation values to obtain a three-dimensional salt distribution field; and extracting salt dynamic change characteristics based on the three-dimensional salt distribution field of multiple continuous time periods to obtain a salt dynamic detection result. The method can accurately invert soil salt of different depths and characterize the space-time dynamic change law of salt.

Description

Technical Field

[0001] This invention relates to the field of soil salinity detection technology, specifically a method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion. Background Technology

[0002] Rapid and accurate detection of soil salinity in saline-alkali orchards is crucial for guiding soil improvement and precision irrigation. Traditional methods for obtaining soil salinity information primarily rely on field sampling and laboratory chemical analysis, which are time-consuming and labor-intensive, making them unsuitable for large-scale, real-time monitoring. Electromagnetic induction technology, due to its non-contact and rapid measurement capabilities, has been introduced into soil salinity surveys, indirectly reflecting soil salinity content by measuring apparent conductivity. However, apparent conductivity is strongly modulated by factors such as soil moisture and temperature, and existing single electromagnetic induction detection methods cannot effectively decouple these coupling effects, resulting in low accuracy in salinity retrieval, especially with more significant errors under different seasons and irrigation conditions.

[0003] In terms of spatial characterization of salinity, existing technologies mostly employ regular grid sampling combined with ordinary kriging interpolation to generate two-dimensional planar salinity distribution maps. These methods neglect salinity variation in the vertical depth dimension, making it difficult to represent the three-dimensional distribution structure of salinity in the orchard root zone. The interpolation process typically relies solely on the spatial autocorrelation of sample points, failing to utilize auxiliary information closely related to salinity transport, such as soil texture and topography, resulting in significant distortion in locally complex areas. In the temporal dimension, current monitoring remains at a static comparison stage of separate mappings at different periods, failing to automatically extract evolution trends and change rates from continuous salinity field data, and thus unable to quantitatively characterize the dynamic migration and accumulation processes of salinity.

[0004] To address the above issues, it is necessary to solve how to utilize multi-source sensors in the complex environment of orchards to eliminate the coupling interference of moisture and temperature on apparent conductivity, thereby achieving accurate estimation of salt content at various depths; and how to automatically identify regions of significant salt variation from time-series distributions and quantify their dynamic evolution rate by constructing a three-dimensional spatiotemporal salt field, thus forming a complete dynamic salt detection scheme. Summary of the Invention

[0005] This invention provides a method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion. It aims to solve the problem of insufficient accuracy in soil salinity inversion by electromagnetic induction apparent conductivity due to the influence of moisture and temperature. At the same time, it solves the problem that existing methods cannot obtain the three-dimensional spatial distribution of soil salinity and automatically extract dynamic change features from continuous periodic data, so as to achieve three-dimensional, dynamic and high-precision detection of soil salinity in saline-alkali orchards.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion, comprising the following steps:

[0007] By using a multi-frequency electromagnetic induction sensor array pre-deployed in the orchard area, electromagnetic waves are emitted at multiple frequencies and secondary field signals are received to obtain apparent conductivity data of multiple depth layers in the target area.

[0008] Simultaneously collect monitoring data from soil moisture sensors and temperature sensors deployed in the orchard, and perform spatial interpolation to obtain soil moisture and temperature distribution that match the spatial location of the apparent conductivity data;

[0009] Based on the pre-established multivariate nonlinear mapping relationship between soil salinity and apparent conductivity, soil moisture, and temperature, the apparent conductivity data, soil moisture distribution, and temperature distribution are converted into initial salinity estimates for different depth layers in the target area.

[0010] Spatiotemporal interpolation is performed on the initial salinity estimate to generate a three-dimensional salinity distribution field covering the entire orchard area;

[0011] Based on the three-dimensional salinity distribution field over multiple consecutive time periods, the dynamic change characteristics of salinity are extracted to obtain the dynamic detection results of salinity.

[0012] Preferably, when acquiring apparent conductivity data using a multi-frequency electromagnetic induction sensor array, the array is arranged at predetermined intervals. Each sensor emits electromagnetic waves of multiple frequencies. The apparent conductivity value for the corresponding frequency is calculated based on the received secondary field signal. A weighted average of the apparent conductivity values ​​for the same depth layer is then performed to obtain the apparent conductivity data for that depth layer. The weighting coefficients are determined based on the sensor position and depth response function. This method, through multi-frequency integrated observation, solves the problem of uncertain depth detection with a single frequency, effectively improving the accuracy and resolution of conductivity data for different depth layers.

[0013] Preferably, when collecting soil moisture and temperature monitoring data simultaneously, multiple soil moisture probes and temperature probes are evenly deployed to collect soil volumetric water content and soil temperature in real time. After wireless transmission and aggregation, spatial interpolation is performed on the data from multiple probes at the same time to obtain soil moisture distribution and temperature distribution that match the spatial location of apparent conductivity data. This provides high-precision environmental parameters for salinity inversion and reduces the interference of moisture and temperature changes on conductivity measurement.

[0014] Preferably, the multivariate nonlinear mapping relationship is pre-established in the following manner: standard soil samples with different salinity gradients, different moisture contents, and different temperatures are prepared in the laboratory; the apparent conductivity values ​​of each standard sample are measured using a multi-frequency electromagnetic induction sensor; an orthogonal polynomial regression method is used to fit a functional expression between apparent conductivity, soil moisture, temperature, and soil salinity; and the coefficients of this functional expression are saved as mapping relationship parameters. This mapping relationship fully considers the nonlinear coupling effect of soil moisture and temperature on conductivity response, making salinity estimation more accurate.

[0015] Preferably, when converting apparent conductivity data, soil moisture distribution, and temperature distribution into initial salinity estimates, the apparent conductivity data of each depth layer, along with the corresponding soil moisture and temperature values, are substituted into the function expression to calculate the estimated soil salinity at that location. Median filtering is then applied to the estimated values ​​for all locations within the same depth layer to remove outliers, resulting in the initial salinity estimate. The window size for the median filter is determined based on the spacing between adjacent sensors within the same depth layer, preferably 1.2 times the number of grid points corresponding to the sensor spacing. The window is square, effectively suppressing single-point anomalies while preserving true spatial variation information in salinity.

[0016] Preferably, when obtaining a three-dimensional salinity distribution field by spatiotemporal interpolation of the initial salinity estimate, an improved Kriging interpolation method is used. This method interpolates the initial salinity estimate in three spatial dimensions and constructs a variogram model using soil texture type and topographic elevation as covariates. Weighting coefficients are determined based on the variogram model, and the salinity value for each grid node is calculated to generate a three-dimensional salinity distribution field covering the entire orchard area. This interpolation method integrates soil texture and topographic information, significantly improving the spatial estimation accuracy of salinity in non-homogeneous media and making the distribution field more consistent with the actual orchard conditions.

[0017] Preferably, the process of extracting the dynamic characteristics of salinity includes: calculating the difference field of the three-dimensional salinity distribution field between adjacent time periods; performing threshold segmentation on the difference field to identify regions with significant salinity changes; for each region with significant changes, statistically analyzing the time series of salinity changes within that region; fitting the periodic and trend components of the time series using an autoregressive moving average model; and using the slope of the trend component as the rate of dynamic salinity change. This method can filter out random fluctuations, separate periodic changes and long-term trends, and quantitatively assess the accumulation or desalination rate of soil salinity.

[0018] Preferably, when using an autoregressive moving average model for fitting, the time series of salinity changes is tested for stationarity. If it is a non-stationary series, first-order differencing is performed. The number of autoregressive terms and moving average terms are determined based on the autocorrelation function and partial autocorrelation function. The model parameters are estimated using the least squares method to obtain the fitted model. The stationarity test uses the ADF test. If the test statistic is greater than the critical value, the series is determined to be non-stationary, and differencing is performed until the series becomes stationary. This ensures the accuracy of the model fitting and the reliability of the prediction.

[0019] Preferably, the dynamic salinity detection results include: drawing a dynamic salinity change trend map based on the dynamic change rate of salinity in each significant change area and the three-dimensional salinity distribution field of the current period, marking areas with change rates exceeding a preset threshold as key areas of concern, and outputting a detection report containing the location, area, and change rate of key areas of concern, providing intuitive and operable decision support for the precise management of saline-alkali land in orchards.

[0020] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0021] A multi-frequency electromagnetic induction sensor array collects apparent conductivity data from multiple depth layers, simultaneously acquiring soil moisture and temperature monitoring data. An initial salinity estimate is calculated using a pre-established multivariate nonlinear mapping relationship between soil salinity and apparent conductivity, soil moisture, and temperature. This mapping relationship is obtained by fitting the functional expressions of apparent conductivity, moisture, temperature, and salinity using orthogonal polynomial regression to indoor standard soil samples under different salinity gradients, moisture contents, and temperatures. This fusion approach incorporates moisture and temperature as explicit variables into the salinity inversion model, directly eliminating the coupling effect of soil moisture changes and temperature fluctuations on the apparent conductivity signal. This allows the inverted salinity estimate to closely approximate the true soil salinity content, maintaining high interpretation accuracy under various field humidity and thermal conditions without relying on external correction or empirical coefficient adjustments.

[0022] The process involves generating a three-dimensional salinity distribution field through spatiotemporal interpolation of the initial salinity estimate, and extracting the dynamic change characteristics of salinity based on the three-dimensional salinity distribution field over multiple consecutive time periods. An improved Kriging interpolation method is used, with soil texture type and topographic elevation as covariates to construct a variogram model, extending the spatially discrete salinity estimate to a three-dimensional regular grid covering the entire orchard area. In the time series analysis, the difference field of the three-dimensional salinity distribution field between adjacent periods is calculated and threshold segmentation is performed to identify areas of significant salinity change. For each area of ​​significant change, an autoregressive moving average model is used to fit the periodic and trend components of the salinity change time series, and the slope of the trend component is used as the rate of dynamic salinity change. The three-dimensional interpolation strategy, which introduces soil texture and topographic elevation as covariates, effectively improves the spatial prediction continuity of salinity at soil interfaces and topographic transitions, and reduces local interpolation uncertainty. The combination of difference field threshold segmentation and autoregressive moving average model can automatically capture active zones of salinity change and decompose the evolution process into periodic fluctuations and directional trends. The output dynamic change rate can quantitatively characterize the intensity of salinity accumulation or leaching, providing a direct criterion for graded early warning and regional regulation. This changes the past mode of qualitatively judging the direction of salinity growth and decline by manually comparing distribution maps of different time phases. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0024] Figure 1 This is a flowchart of a method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion;

[0025] Figure 2 This is a flowchart of three-dimensional salinity co-kriging interpolation with the introduction of soil texture and topographic elevation covariates. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] See Figure 1This invention provides a method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion. It acquires apparent conductivity data at multiple depths within a target area of ​​the orchard using a multi-frequency electromagnetic induction sensor array, and simultaneously collects monitoring data from soil moisture and temperature sensors in the target area. Based on a pre-established multivariate nonlinear mapping relationship between soil salinity and apparent conductivity, soil moisture, and temperature, the apparent conductivity data and monitoring data are converted into initial salinity estimates at different depths within the target area. Spatiotemporal interpolation is performed on these initial salinity estimates to obtain a three-dimensional salinity distribution field. Based on the three-dimensional salinity distribution field over multiple consecutive time periods, dynamic salinity change characteristics are extracted to obtain the dynamic salinity detection results.

[0028] Example 1

[0029] In practice, a multi-frequency electromagnetic induction sensor array is deployed within the orchard area at predetermined intervals. These intervals are determined based on the area of ​​the target orchard region and the spatial variation characteristics of soil salinity. The multi-frequency electromagnetic induction sensors are evenly arranged along both row and column directions within the orchard area, with identical row and column spacing between adjacent sensors. The multi-frequency electromagnetic induction sensor consists of a transmitting coil, a receiving coil, a signal generation module, a signal acquisition module, and a control unit. The signal generation module generates excitation currents of multiple frequencies and feeds them into the transmitting coil, which radiates a primary electromagnetic field. The multiple frequencies emitted by the multi-frequency electromagnetic induction sensor cover low-frequency, mid-frequency, and high-frequency bands. The low-frequency band ranges from 1kHz to 3kHz, the mid-frequency band from 5kHz to 10kHz, and the high-frequency band from 12kHz to 18kHz. The penetration depth of the primary electromagnetic field in the soil varies depending on the frequency; the low-frequency electromagnetic wave has the largest detection depth, while the high-frequency electromagnetic wave has the smallest. An electromagnetic field induces eddy currents in the soil medium, which in turn excite a secondary magnetic field. The receiving coil then receives the secondary field signal, which contains in-phase and quadrature components. The signal acquisition module amplifies, filters, and performs analog-to-digital conversion on the secondary field signal.

[0030] The apparent conductivity values ​​at different frequencies are calculated based on the secondary field signal. The apparent conductivity value is determined by the ratio of the in-phase component of the secondary field signal to the primary magnetic field strength, using the following expression:

[0031] in, This indicates apparent conductivity, expressed in Siemens units per meter. Represents angular frequency. Equal to the frequency of the emitted electromagnetic wave multiplied by The unit is radians per second; Represents the permeability of free space. The value is Henry per meter; This indicates the distance between the transmitting coil and the receiving coil, in meters. The in-phase component of the secondary magnetic field strength at the receiving coil is represented by amperes per meter. This represents the intensity of the primary magnetic field generated by the transmitting coil at the location of the receiving coil, measured in amperes per meter. It is calculated using the number of turns, current, and geometric parameters of the transmitting coil.

[0032] After obtaining the apparent conductivity values ​​of each multi-frequency electromagnetic induction sensor at various transmission frequencies, the apparent conductivity values ​​corresponding to the same depth layer are weighted and averaged to obtain the apparent conductivity data of the depth layer. The depth layer is divided based on the skin depth of electromagnetic waves of different frequencies, and the skin depth expression is: ,in Indicates skin depth. Represents angular frequency. Represents the permeability of free space. The average electrical conductivity of the soil is represented by the depth layer. Based on skin depth, the area from the surface to the maximum detection depth is divided into multiple depth layers at equal intervals. The number of depth layers corresponds to the number of transmission frequencies, and each depth layer corresponds to the detection sensitivity center depth for a specific frequency. Weighting coefficients are determined based on the location and depth response function of the multi-frequency electromagnetic induction sensor. The depth response function describes the sensitivity distribution of a specific frequency electromagnetic wave to the apparent conductivity of each depth layer. The depth response function is obtained through forward modeling of layered media using Maxwell's equations. The relative position between the multi-frequency electromagnetic induction sensor and the depth layer, and the thickness of the depth layer, are used as inputs. The forward modeling outputs the sensitivity value of a specific frequency for each depth layer at each sensor location. The weighting coefficients are proportional to the sensitivity values. For each depth layer, the sensitivity values ​​of all multi-frequency electromagnetic induction sensors at that depth layer are used to construct a weight vector. This weight vector is normalized so that the sum of the weights equals one. The apparent conductivity values ​​of the frequencies corresponding to that depth layer calculated by each multi-frequency electromagnetic induction sensor are multiplied by their respective normalized weights and then summed to obtain the apparent conductivity data for that depth layer.

[0033] In some embodiments, the predetermined spacing of the multi-frequency electromagnetic induction sensor array is set to five meters. Each multi-frequency electromagnetic induction sensor emits electromagnetic waves at three frequencies: 1.5 kHz, 8 kHz, and 15 kHz. The corresponding detection center depth layers are 0.5 meters, 1 meter, and 1.5 meters, respectively. The thickness of each of the three depth layers is 0.5 meters. The depth response function is obtained by forward modeling a three-layer horizontal stratified geodetic model. During the forward modeling calculation, the resistivity of each layer is the average resistivity value of the orchard area.

[0034] Regarding the deployment and data acquisition of soil moisture and temperature sensors, multiple soil moisture and temperature probes are evenly distributed throughout the target area of ​​the orchard. The placement of these probes corresponds one-to-one with the spatial grid nodes of the multi-frequency electromagnetic induction sensor array. One set of soil moisture and temperature probes is installed at each grid node. The soil moisture probes are buried at the center depth of the corresponding depth layer, while the temperature probes are installed at the same location. The soil moisture probes measure soil volumetric water content based on time-domain or frequency-domain reflectometry, outputting a voltage value, which is converted into a percentage of soil volumetric water content through a calibration curve. The temperature probes, based on platinum resistance or thermistor principles, output a resistance value, which is converted into a temperature value through a voltage divider circuit and an analog-to-digital converter. Each soil moisture and temperature probe is equipped with a microcontroller unit and a wireless communication module. The microcontroller unit reads sensor values ​​at a set acquisition time interval, synchronized with the apparent conductivity data acquisition cycle of the multi-frequency electromagnetic induction sensor array. The wireless communication module uses the LoRa protocol for data transmission. Each soil moisture probe and temperature probe acts as a terminal node, sending the collected soil volumetric water content and soil temperature values ​​to the aggregation node in the form of data frames via the LoRa wireless link. The aggregation node connects to the host computer in the data acquisition center via a serial port and transmits the received data to the host computer.

[0035] The data acquisition center performs spatial interpolation on multiple soil moisture and temperature probe data collected simultaneously. The inverse distance-weighted interpolation method is used, and the spatial resolution of the interpolation target grid matches the spatial location of the apparent conductivity data. Specifically, the spatial grid nodes of the interpolated soil moisture and temperature distributions completely coincide with the apparent conductivity measurement points of the multi-frequency electromagnetic induction sensor array. The inverse distance-weighted interpolation method uses the reciprocal of the distance between the interpolation target point and the surrounding known soil moisture probe locations as the weight. The weighting function uses a squared distance decay, calculated as follows: ,in Indicates the first The weights of a known soil moisture probe to the interpolation target point. Indicates the first Spatial distance between a known soil moisture probe and the interpolation target point Indicates the distance decay index. The value is set to two. After normalizing all weights, they are multiplied by the corresponding soil volumetric water content value from the soil moisture probe and summed to obtain the soil volumetric water content value of the interpolation target point. The spatial interpolation process for temperature probe data is the same as that for soil moisture probe data. Soil moisture distribution and temperature distribution that match the spatial location of apparent conductivity data are generated through interpolation. The soil moisture distribution and temperature distribution are stored in two-dimensional matrix form, with the rows and columns of the matrix corresponding to the planar coordinate grid of the orchard area.

[0036] Example 2

[0037] In practice, the process of establishing a multivariate nonlinear mapping relationship between soil salinity and apparent conductivity, soil moisture, and temperature is as follows: Unspoiled soil samples were collected from the target saline-alkali orchard area. The soil was air-dried in the laboratory, then ground and passed through a 2mm mesh sieve to obtain the basic soil matrix. When preparing standard soil samples, three factors were set: soil salinity gradient, soil volumetric water content gradient, and soil temperature gradient. The soil salinity gradient was set at eight levels, ranging from a saline-free state to the highest salinity level ever recorded in the orchard. These eight levels were 0.5 g / kg, 1.0 g / kg, 2.0 g / kg, 4.0 g / kg, 6.0 g / kg, 8.0 g / kg, 10.0 g / kg, and 12.0 g / kg, respectively. This was achieved by adding a fixed amount of sodium chloride to the basic soil matrix and mixing thoroughly. The soil volumetric water content gradient was calculated using eight levels ranging from air-dried water content to field capacity, with levels of 0.05 cubic centimeters per cubic centimeter, 0.10 cubic centimeters per cubic centimeter, 0.15 cubic centimeters per cubic centimeter, 0.20 cubic centimeters per cubic centimeter, 0.25 cubic centimeters per cubic centimeter, 0.30 cubic centimeters per cubic centimeter, 0.35 cubic centimeters per cubic centimeter, and 0.40 cubic centimeters per cubic centimeter. This gradient was achieved by uniformly spraying a measured amount of deionized water onto the soil samples. The temperature gradient was calculated using eight levels ranging from the lowest spring soil temperature to the highest summer soil temperature in the orchard, with levels of 5°C, 10°C, 15°C, 20°C, 25°C, 30°C, 35°C, and 40°C. This gradient was achieved by maintaining the soil samples at a constant temperature in a constant-temperature chamber. A total of 512 standard soil samples were prepared using a full-factor combination approach for the eight levels of the three factors. Each standard soil sample was placed in a standard measuring container with an inner diameter of 20 cm and a height of 30 cm. During filling, the soil bulk density was controlled to be consistent with the actual soil bulk density of the target area in the orchard. The actual soil bulk density of the orchard was obtained by sampling and measuring using the ring cutter method.

[0038] The apparent conductivity of each standard soil sample was measured using a multi-frequency electromagnetic induction sensor. During measurement, the multi-frequency electromagnetic induction sensor was maintained at a fixed vertical distance of 5 cm from the surface of the soil sample in the standard measurement container, with the sensor's transmitting coil plane parallel to the soil surface. For each standard soil sample, the multi-frequency electromagnetic induction sensor sequentially emitted electromagnetic waves at three frequencies: 1.5 kHz, 8 kHz, and 15 kHz. The in-phase and quadrature components of the secondary field signal at each frequency were recorded, and the corresponding apparent conductivity value was calculated. Each standard soil sample was measured five times at each frequency, and the arithmetic mean of the five apparent conductivity values ​​was taken as the apparent conductivity value of that standard soil sample at that frequency. Simultaneously, the actual soil salinity, soil volumetric water content, and soil temperature values ​​corresponding to each standard soil sample were recorded. All data were combined into a calibration dataset. Each data record in the calibration dataset contains a set of soil salinity values, three apparent conductivity values ​​corresponding to the three frequencies, soil volumetric water content values, and soil temperature values ​​for a single standard soil sample.

[0039] Based on the calibrated dataset, an orthogonal polynomial regression method was used to fit the functional expression between apparent conductivity, soil moisture, temperature, and soil salinity. The functional expression is in the form of a quadratic polynomial containing linear, quadratic, and interaction terms:

[0040]

[0041] in, This indicates soil salinity, expressed in grams per kilogram. This represents the apparent conductivity value, in Siemens per meter, corresponding to the apparent conductivity data of a specific depth layer; This indicates the volumetric water content of the soil, expressed in cubic centimeters per cubic centimeter. This indicates soil temperature, expressed in degrees Celsius. The coefficient of the constant term, , , The coefficient of the linear term, , , The coefficient of the quadratic term, , , These are the interaction term coefficients; all coefficients are determined by an orthogonal polynomial regression process.

[0042] The implementation method of orthogonal polynomial regression is as follows: first, for the apparent conductivity variable... Variables in soil volumetric water content and soil temperature variables Centering was performed on each variable by subtracting its mean from the calibration dataset. Orthogonal polynomial basis functions were then constructed based on the centered variables using the Gram-Schmidt orthogonalization process, progressively generating mutually orthogonal basis functions from lower-order to higher-order terms. Soil salinity values ​​were then... Using orthogonal polynomial basis functions as the dependent variable and independent variables, the coefficients of each orthogonal polynomial basis function are estimated using the least squares method. The least squares estimation is achieved by solving a system of normal equations, which degenerates into a diagonal matrix due to the orthogonality of the orthogonal polynomial basis functions. The coefficients of each orthogonal polynomial basis function can then be solved independently. The obtained coefficients of the orthogonal polynomial basis functions are then transformed back to the coefficients in the original variable space through an inverse transformation. to The coefficients to The numerical values ​​are saved as parameters of the multivariate nonlinear mapping relationship to the parameter configuration file. The parameter configuration file is stored in XML format, and records the name of each coefficient and its corresponding floating-point value.

[0043] In some embodiments, apparent conductivity variable Using a weighted average apparent conductivity value corresponding to each depth layer, for a three-layer depth structure, a set of mapping parameters is established for each depth layer. The three sets of parameters correspond to depth layers of 0.5 meters, 1 meter, and 1.5 meters, respectively. The apparent conductivity value in each set of parameters... This is the apparent conductivity data obtained by weighted averaging of the frequency measurements corresponding to this depth layer. The parameter configuration file also stores the association between the depth layer identifier and the corresponding coefficient set.

[0044] After establishing the multivariate nonlinear mapping relationship, the apparent conductivity data of each depth layer in the target area, along with the corresponding soil moisture and temperature values ​​at spatial locations, are substituted into the function expression of the multivariate nonlinear mapping relationship to calculate the estimated soil salinity at each location. During the calculation, the soil moisture value at the corresponding spatial location is taken from a soil moisture distribution matrix that perfectly matches the spatial location of the apparent conductivity data after spatial interpolation, and the temperature value at the corresponding spatial location is taken from a temperature distribution matrix that perfectly matches the spatial location of the apparent conductivity data after spatial interpolation. Function calculations are performed point-by-point for all measurement points within the target area to generate a set of estimated soil salinity values ​​covering all depth layers and all planar locations.

[0045] Median filtering is applied to the soil salinity estimates at all locations within the same depth layer to remove outliers. The window size for median filtering is determined based on the spacing between adjacent multi-frequency electromagnetic induction sensors at the same depth layer. The method is as follows: multiply the spacing between adjacent multi-frequency electromagnetic induction sensors by a coefficient of 1.2 to obtain the spatial side length of the filtering window. Divide the spatial side length of the filtering window by the side length of the spatial interpolation grid to obtain the number of grid points corresponding to the filtering window. Round the result up. If the number of grid points obtained after rounding is even, increment the number by one to make it odd, ensuring that the filtering window has a unique central grid point. The filtering window is square, and the number of grid points on the side of the square is the odd number of grid points obtained above. The median filtering process is as follows: using each grid point on the depth layer plane as the center, extract the soil salinity estimates of all grid points within the square window. Arrange the extracted values ​​in ascending order, and take the median of the arranged sequence as the filtered soil salinity estimate for the central grid point. For grid points located at the edge of the depth layer plane, when the square window extends beyond the boundary, the grid points outside the boundary are excluded from the window, and only the values ​​of the actual grid points within the window are used for median calculation. After median filtering, the soil salinity estimates of all grid points constitute the initial salinity estimates for that depth layer.

[0046] Example 3

[0047] In specific implementation, please refer to Figure 2 An improved Kriging interpolation method was used to interpolate the initial salinity estimates in three spatial dimensions: two horizontal directions of the planar position coordinates and the vertical depth direction. The planar position coordinate system adopted the orchard construction coordinate system, with the origin located at the southwest corner of the orchard area. The east-west direction was the first horizontal coordinate axis, and the north-south direction was the second horizontal coordinate axis, with the unit of the coordinate axes being meters. The depth coordinates had the ground surface as the zero point, with the vertical direction downwards representing the direction of numerical increase, and the unit was consistent with the planar coordinate units.

[0048] Soil texture type and topographic elevation are introduced as covariates during the interpolation process. Soil texture types are divided into three categories: sandy soil, loam, and clay. Soil texture type data is derived from a soil survey map of the orchard area at a scale of 1:1000. The soil survey map is digitized, converting the soil texture type of each spatial location point into a numerical code: sandy soil is coded as code one, loam as code two, and clay as code three. The coded soil texture types form a matrix of coded values ​​that correspond one-to-one with the spatial grid nodes of the initial salinity estimate. Topographic elevation data is obtained through real-time dynamic differential positioning measurement or UAV aerial surveying. Real-time dynamic differential positioning measurement equipment collects ground point elevations along a pre-set grid line within the orchard. UAV aerial surveying acquires aerial images using an optical camera and a global navigation satellite system receiver. Aerial triangulation is used to generate a digital elevation model. The spatial resolution of the digital elevation model is consistent with the spatial grid nodes of the initial salinity estimate. The topographic elevation value at each grid node is the altitude value, and the topographic elevation value is used as a continuous variable in the interpolation calculation.

[0049] The process of constructing the variogram model is as follows: First, calculate the experimental variogram value of the initial salinity estimate. The formula for calculating the experimental variogram is:

[0050]

[0051] in, Represents the lag vector The corresponding experimental variogram value is dimensionless and its value is consistent with the square unit of the salinity estimate. This represents the spatial lag vector, which includes three components in the planar and depth directions. The magnitude of a vector represents spatial distance, measured in meters. This indicates that the spatial interval is approximately equal to the lag vector across all initial salinity estimate data points. Number of data pairs; Indicates the first The spatial location vector of each data point is represented by three coordinate components, with the unit being meters; Represents spatial position vector The initial salt content estimate at the location, in grams per kilogram; Indicates and Phase lag vector The initial salinity estimate at the spatial location is given in grams per kilogram. When calculating the experimental variability, the three-dimensional space is grouped according to different lag distances and directions, and the lag vectors are statistically analyzed separately. The sum of the squares of the differences between all corresponding data pairs, divided by twice the number of data pairs, yields a set of discrete experimental variation function values.

[0052] A spherical model is used to fit the discrete experimental variation function values. The spherical model is based on the nugget value. , base value and variable range Three parameters control the function relationship of the spherical model when the distance is less than the range. At that time When the distance is greater than or equal to the range At that time ,in The lag distance is used. The fitting process employs weighted least squares, with the number of data pairs at different lag distances used as the reference. As weights, the weighted sum of squared residuals between the calculated values ​​from the spherical model and the experimental variation function values ​​is minimized, and the nugget value is determined through iterative optimization. , base value and variable range The optimal values ​​are determined. The soil texture type covariate has its own variogram and a cross-variogram with the initial salinity estimate; the topographic elevation covariate also has its own variogram and a cross-variogram. The experimental variogram and cross-variogram were calculated for the soil texture type coding matrix and fitted using a spherical model; similarly, the experimental variogram and cross-variogram were calculated for the topographic elevation matrix and fitted using a spherical model.

[0053] The weighting coefficients are determined based on the variogram model, and a set of estimation equations including soil texture type covariates and topographic elevation covariates is constructed using the co-kriging method. The form of the estimation equations is as follows: for each interpolation location... and each known data point Establish the first system of equations and the second system of equations. The first system of equations is... ,in The number of known points is used to estimate the initial salinity. The weighting coefficients of the main variable, i.e., the initial salinity estimate. For and The value of the variation function of the main variable is the independent variable. The number of known points for the soil texture type covariate. These are the weighting coefficients for the soil texture type covariate. The cross-variance function is the numerical value between the soil texture type covariate and the initial salinity estimate. The number of known points for the topographic elevation covariate. The weighting coefficients for the topographic elevation covariate. This represents the numerical value of the cross-variance function between the topographic elevation covariate and the initial salinity estimate. As the first Lagrange multiplier, Points to be interpolated With known points The values ​​of the variograms of the main variables are obtained. The second and third equation sets correspond to the unbiased constraints of the soil texture type covariate and the topographic elevation covariate, respectively. Their forms are similar to the first equation set, with the cross-variogram values ​​on the right side of the equations, and the second and third Lagrange multipliers are introduced respectively. A symmetric linear equation system is constructed by simultaneously solving the equations corresponding to all known points. Solving the linear equation system yields the weight coefficients of all main variables. and the weighting coefficients of the two covariates and .

[0054] Using the weight coefficients obtained from the solution, calculate the interpolation location. The estimated salinity is calculated using the following formula: ,in For soil texture type covariate in The encoded value at that location, For topographic elevation covariates in The elevation value at the location is determined. The weighting coefficient solution and estimation process is repeated at each interpolated grid node to obtain a three-dimensional salinity distribution field covering the entire orchard area. The grid resolution of the three-dimensional salinity distribution field has a spacing in both planar directions that is half the spacing of the multi-frequency electromagnetic induction sensor, and a spacing in the depth direction that is consistent with the depth layer spacing. The three-dimensional salinity distribution field is stored in the form of a three-dimensional array. The first dimension of the array corresponds to the east-west coordinate index, the second dimension corresponds to the north-south coordinate index, and the third dimension corresponds to the depth coordinate index. Each array element is a data structure containing the east-west coordinate value, north-south coordinate value, depth coordinate value, and salinity estimate of the grid node.

[0055] Example 4

[0056] In practice, the difference field of the three-dimensional salinity distribution field between adjacent time periods is calculated. The time period is set to a fixed time interval, the length of which is determined according to the temporal resolution requirements of orchard salinity monitoring; the time interval can be set to one week or one month. The three-dimensional salinity distribution field corresponding to each time period is denoted as follows: ,in Represents a spatial position vector. Indicates the first The time marker for the first time period. The three-dimensional salinity distribution field corresponding to each time period is denoted as follows: Difference field The definition is as follows:

[0057]

[0058] in, Represents spatial position vector Time cycle arrive The change in salt content, expressed in grams per kilogram; The coordinate system is a three-dimensional spatial coordinate system, which includes two horizontal components of the planar position coordinates and one vertical component of the depth coordinates. The unit of each component is meters. and The identifier number for two adjacent time periods; Indicates spatial location Located in the The salinity value for each time period, expressed in grams per kilogram; Indicates spatial location At the The salinity value for each time period, expressed in grams per kilogram.

[0059] For difference field Perform threshold segmentation. Set a threshold for salinity change. , The threshold for salt change was determined based on the sensitivity requirements for monitoring salt content changes in orchards. The value can be taken as 0.3 grams per kilogram, which corresponds to the detectable change in salinity under moderate variability in orchard soil. The absolute value of the salinity change at each grid node in the difference field is then compared with the salinity change threshold. In comparison, if Then mark the grid node as a change point; if If the value is not a change point, then the grid node will not be marked as a change point. After traversing all grid nodes in the difference field, a binary label matrix consisting of change points is obtained.

[0060] Connectivity analysis is performed on the change points in the binarized label matrix. The analysis employs a three-dimensional six-neighbor connectivity criterion, where each change point is connected to six neighboring change points spatially adjacent in the east-west, north-south, and depth directions, with a distance of one grid interval. A seed-filling algorithm or a disjoint-set data structure algorithm is used to group all adjacent change points satisfying the connectivity relationship into the same connected component. Each connected component is assigned a unique identifier, a positive integer incrementing from one. For each connected component, the number of change points contained within it is counted, and the spatial extent of the connected component is calculated. The spatial extent is determined by the minimum and maximum coordinates of the connected component in the east-west, north-south, and depth directions. Spatially continuous connected components are identified as regions of significant salinity change. Each region of significant salinity change retains its unique identifier and is stored in association with the spatial extent information of that connected component.

[0061] For each region of significant salinity change, the time series of salinity changes within that region is statistically analyzed. The region of significant salinity change is identified by the identifier […]. The region was analyzed to obtain the changes in salinity over various time periods. The time markers for the changes in salinity between adjacent time periods are denoted as follows: , from Start numbering, the total number of time period pairs is Time series of salinity changes It consists of the regional average salinity change for each time period. The calculation method is as follows: for the first... For each adjacent time period pair, select the region of significant salinity change. The arithmetic mean of the absolute values ​​of the differences between all grid nodes. ,in The identifier is The spatial range of areas with significant salinity variations. Indicates the region Number of internal grid nodes In the first Each time period corresponds to the spatial position. The difference at each point. Arranged chronologically according to the time period, areas of significant salinity variation are obtained. Time series of salinity changes .

[0062] An autoregressive moving average model was used to fit the periodic and trend components of the time series of salinity changes. For the time series... A stationarity test is performed using the ADF test. The ADF test calculates the test statistic by establishing an autoregressive model of the time series. Test statistic This is the ratio of the estimated autoregressive coefficient to its standard error. The significance level is set to 0.05, and the critical value for the ADF test is the critical value at this significance level. The critical value is obtained by looking up a table. If the test statistic... If the value exceeds the critical value of the ADF test, the time series is determined to be non-stationary. For non-stationary series, a difference process is performed on the time series to obtain a new series. ,in , from arrive For first-order difference sequences Perform the ADF test again. If the sequence is still non-stationary, continue with second-order differencing until the differencing sequence becomes stationary. Record the differencing order as 1. If the original time series If a sequence is determined to be stationary directly through the ADF test, then the difference order is... If the sequence becomes stationary after first-order differencing, then the order of the differencing is... .

[0063] The number of autoregressive terms is determined based on the autocorrelation function and partial autocorrelation function of the stationary series. and moving average number of terms Calculate the sample autocorrelation function of the stationary sequence, and the autocorrelation function at different lag orders. value at The calculation formula is ,in Indicates the sequence after stabilization. Each element value The mean of the stationary sequence. Let be the length of the stationarized sequence. In the autocorrelation function plot, the maximum lag order where the autocorrelation function is significantly non-zero is denoted as . , As a moving average term The reference value is [value]. The partial autocorrelation function (PAC) of the stationary series is calculated recursively using the Yule-Walker equation. In the PAC plot, the largest lag order where the PAC is significantly non-zero is denoted as [value]. , As an autoregressive term The reference value. In and Several candidate combinations were selected from the vicinity, and the order was determined using the information criterion. The information criterion adopted was the AIC criterion. The combination of the number of autoregressive terms and the moving average term that minimized the AIC value was selected as the final number of autoregressive terms. and moving average number of terms .

[0064] Model parameters are estimated using the least squares method. ARIMA ( The expression for the model is:

[0065]

[0066] in, express Difference operator of order, difference order If it is a non-negative integer, when hour ; Indicates the first Time series values ​​of salt content change for each time period, in grams per kilogram; For time period indexing, Take a positive integer. The range is from 1 to the time series length. ; The number of autoregressive terms is a positive integer. These are the autoregressive coefficients, each of which is a real number and generally ranges from -1 to 1. The number of terms in the moving average is a non-negative integer. These are the moving average coefficients, each of which is a real number and generally ranges from -1 to 1. For the first The random error terms of each time period pair follow a normal distribution with a mean of zero and a constant variance. The differences between different time period pairs... Independent of each other The sequence constitutes a white noise sequence.

[0067] The least squares method is used to estimate model parameters by: dividing the differencing sequence... As the dependent variable, the lagged term and estimated residuals As independent variables, the autoregressive coefficients and moving average coefficients are solved through iterative optimization. In the initial iteration, the residual term is set to zero, and the autoregressive coefficients are estimated using least squares. The residual sequence is then calculated based on the autoregressive coefficients, and the moving average coefficients are estimated by back-substituting. This process is repeated until the coefficients converge. After obtaining the fitted model, the salinity change for the next time period is predicted based on the fitted model. During prediction, the predicted value of the difference sequence for the next period is calculated using the model's recursive formula, and then the predicted value of the original sequence is restored through inverse differencing.

[0068] The slope of the trend component is used as the rate of dynamic change in salinity. The trend component is obtained by denoting the fitted value sequence of the ARIMA model to the historical time series as follows: Indexed by time period As the independent variable, the fitted value Using [variable name] as the dependent variable, perform univariate linear regression. The linear regression model is: ,in For the intercept term, Let be the slope of the trend component. The least squares method is used to estimate it. and The formula for calculating the slope is: ,in This is the average value of the time period index. The average value of the fitted value sequence. The slope of the trend component. This refers to the rate of dynamic change in salinity in areas where salinity changes significantly, expressed in grams per kilogram per time period.

[0069] Example 5

[0070] In practice, a dynamic trend map of salinity is drawn based on the rate of change of salinity in each area of ​​significant salinity variation and the three-dimensional salinity distribution field of the current time period. A planar base map of the target area of ​​the orchard is obtained, which is a two-dimensional projection of the orchard construction coordinate system. The location of the fruit tree planting rows, the direction of irrigation pipes, and the distribution of drainage ditches within the orchard are marked on the planar base map. The three-dimensional salinity distribution field of the current time period is processed layer by layer along the depth direction to extract the planar salinity distribution at each depth layer. The planar salinity distribution of the current time period is superimposed on the planar base map. The planar salinity distribution is presented in the form of filled color blocks. The color mapping relationship of the filled color blocks adopts a gradient spectrum from blue to red. Blue corresponds to low salinity zones with salinity values ​​below one gram per kilogram, and red corresponds to high salinity zones with salinity values ​​above six grams per kilogram. Color values ​​for intermediate salinity value intervals are assigned using linear interpolation. For each area of ​​significant salinity change, a bold boundary line is used to delineate the area on the salinity dynamic change trend map. The line width of the bold boundary line is set to 2.0 mm, the color is a solid black line, and the line type is a solid line. Within each area of ​​significant salinity change, the numerical value of the salinity dynamic change rate and the sign of the direction of change are marked. The numerical value of the salinity dynamic change rate is accurate to two decimal places, and the unit of the salinity dynamic change rate is grams per kilogram per time period. In the sign of change direction, a positive character indicates a salt accumulation trend, and a negative character indicates a desalination trend. The sign of change direction is placed before the numerical value of the salinity dynamic change rate.

[0071] Areas showing significant salt changes where the absolute value of the rate of change in salt dynamics exceeds a preset threshold are marked as key areas of concern. The preset threshold is set based on the salt tolerance threshold of the main fruit trees in the orchard. This threshold is determined through potted salt stress gradient experiments in an artificial climate chamber, using the soil salinity in the root zone when the photosynthetic rate of the fruit tree leaves drops to 50% of the control treatment as the salt tolerance threshold. Taking the Red Fragrant Pear as the main variety in the saline-alkali land orchard as an example, the salt tolerance threshold for the Red Fragrant Pear was determined to be 1.8 grams per kilogram. The preset threshold is set at 0.6 grams per kilogram per time period. This is achieved by dividing the Red Fragrant Pear's salt tolerance threshold of 1.8 grams per kilogram by three, resulting in a warning level change rate of 0.6 grams per kilogram per time period. Setting this value to 0.6 ensures that the accumulated salt content does not exceed the salt tolerance threshold within three time periods. On the salt dynamic change trend map, areas with significant salt changes where the absolute value of the salt dynamic change rate exceeds 0.6 grams per kilogram per time period are identified as key areas of concern. Diagonal shading lines are added outside the thickened boundary lines of key areas of concern. The angle of the diagonal shading lines is 45 degrees, the spacing between the diagonal lines is 3 millimeters, and the color of the diagonal lines is red and semi-transparent.

[0072] The output includes a detection report containing the location, area, and salinity dynamic change rate of key concern areas. The report is organized using a combination of text descriptions and graphical representations. The text description lists the spatial coordinate range, area, salinity dynamic change rate, direction of change, and concern level for each key concern area. The spatial coordinate range is described by the smallest bounding rectangle of the key concern area in the planar coordinate system. This smallest bounding rectangle is defined by four parameters: the smallest east-west coordinate value, the largest east-west coordinate value, the smallest north-south coordinate value, and the largest north-south coordinate value. All four parameters are in meters and rounded to one decimal place. The depth range is described by the depth layer identifier intervals that the key concern area spans vertically. These intervals are indicated by the starting and ending depth layer numbers, which increase from one downwards from the surface. The area is obtained by calculating the area of ​​the outline polygon projected onto the planar surface of the key concern area, in square meters, rounded to the nearest integer. The salinity dynamic change rate is directly taken from the salinity dynamic change rate of the corresponding area with significant salinity changes, in grams per kilogram per time period, accurate to two decimal places. The direction of the trend is indicated by the sign of the dynamic rate of change in salinity. A positive rate indicates salt accumulation, while a negative rate indicates desalination. The level of concern is determined by the ratio of the absolute value of the dynamic rate of change in salinity to a preset threshold. There are two levels: high concern and medium concern. The ratio is calculated as follows: a ratio greater than or equal to 2.0 (i.e., an absolute rate of change of 1.2 grams per kilogram per time period) is considered high concern; a ratio between 1 and 2 (i.e., an absolute rate of change of 0.6 grams per kilogram per time period to 1.2 grams per kilogram per time period) is considered medium concern. The text description section also includes a summary statistics row, listing the total number and total area of ​​all key concern areas.

[0073] The graphical display includes a 3D salinity distribution field profile, a salinity change trend distribution map, and a map of key areas of concern. The 3D salinity distribution field profile is a vertical section taken along the main salinity drainage direction of the orchard. This main drainage direction is determined by analyzing the slope aspect data of the orchard's topographic elevation model. The slope aspect data was calculated using the slope aspect analysis tool in ArcGIS software, and the mode slope aspect was taken as the main drainage direction. The vertical profile displays the vertical distribution of salinity values ​​in the form of contour lines, with a salinity interval of 0.5 grams per kilogram. The contour lines are labeled with salinity values. The salinity change trend distribution map shows the planar distribution of all areas with significant salinity changes. The center position of each area with a significant salinity change is marked with a dotted symbol with an arrow. The direction of the arrow indicates the direction of the salinity change trend: an upward arrow indicates a salt accumulation trend, and a downward arrow indicates a desalination trend. The size of the arrow symbol is proportional to the absolute value of the salinity dynamic change rate. The key concern area distribution map is displayed as a separate layer, showing only the areas marked as key concern areas. These key concern areas are indicated by red fill, with their numbers labeled outside the red fill. These numbers correspond one-to-one with the numbers listed in the text description section. The inspection report is generated as an electronic document in PDF format. The PDF document displays the report generation date and time in the header and the saline-alkali land orchard name and document page numbers in the footer. The PDF document is stored in a designated path on the file server of the data acquisition center and can be opened and displayed using a PDF reader on a data display terminal. The data display terminal is an LCD monitor with a resolution of at least 1920 x 1080 pixels. Simultaneously, the PDF document is remotely pushed to mobile terminals via a push notification service based on the MQTT protocol. A dedicated viewing application is installed on the mobile terminal. Upon receiving the PDF document push, the dedicated viewing application sends a notification and caches the document for offline viewing.

[0074] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion, characterized in that, Includes the following steps: Apparent conductivity data of multiple depth layers in the target area of ​​the orchard were acquired based on a multi-frequency electromagnetic induction sensor array. Simultaneously collect monitoring data from soil moisture and temperature sensors in the target area; Based on the pre-established multivariate nonlinear mapping relationship between soil salinity and apparent conductivity, soil moisture, and temperature, the apparent conductivity data and monitoring data are converted into initial salinity estimates for different depth layers in the target area. Spatiotemporal interpolation is performed on the initial salinity estimate to obtain a three-dimensional salinity distribution field; Based on the three-dimensional salinity distribution field over multiple consecutive time periods, the dynamic change characteristics of salinity are extracted to obtain the dynamic detection results of salinity.

2. The method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion according to claim 1, characterized in that, The acquisition of apparent conductivity data at multiple depth layers in the target area of ​​the orchard based on a multi-frequency electromagnetic induction sensor array is specifically as follows: A multi-frequency electromagnetic induction sensor array is deployed in the orchard area at predetermined intervals. Each sensor emits electromagnetic waves of multiple frequencies and receives secondary field signals. The apparent conductivity values ​​corresponding to different frequencies are calculated based on the secondary field signals. The apparent conductivity values ​​corresponding to the same depth layer are weighted and averaged to obtain the apparent conductivity data of that depth layer. The weighting coefficients are determined according to the sensor position and depth response function.

3. The method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion according to claim 1, characterized in that, The simultaneous acquisition of monitoring data from soil moisture and temperature sensors in the target area specifically includes: Multiple soil moisture and temperature probes are evenly deployed in the target area of ​​the orchard. Each probe collects soil volumetric water content and soil temperature in real time. The data is then wirelessly transmitted to the data acquisition center. Spatial interpolation is performed on the data from multiple probes at the same time to obtain the soil moisture and temperature distribution that matches the spatial location of the apparent conductivity data.

4. The method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion according to claim 1, characterized in that, The aforementioned multivariate nonlinear mapping relationship between soil salinity and apparent conductivity, soil moisture, and temperature, established beforehand, specifically includes: The multivariate nonlinear mapping relationship is established as follows: standard soil samples with different salinity gradients, different moisture contents, and different temperatures are prepared in the laboratory. The apparent conductivity values ​​of each standard sample are measured using a multi-frequency electromagnetic induction sensor. An orthogonal polynomial regression method is used to fit the functional expression between apparent conductivity, soil moisture, temperature, and soil salinity. The coefficients of the functional expression are saved as mapping relationship parameters.

5. The method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion according to claim 1, characterized in that, The process of converting the apparent conductivity data and monitoring data into initial salinity estimates for different depth layers in the target area specifically involves: Substituting the apparent conductivity data of each depth layer, along with the corresponding soil moisture and temperature values ​​at the spatial location, into the function expression of the multivariate nonlinear mapping relationship, the soil salinity estimate at that location point is calculated. Then, median filtering is applied to the estimated values ​​of all locations at the same depth layer to remove outliers, resulting in the initial salinity estimate.

6. The method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion according to claim 1, characterized in that, The process of performing spatiotemporal interpolation on the initial salinity estimate to obtain a three-dimensional salinity distribution field is as follows: An improved Kriging interpolation method was used to interpolate the initial salinity estimate in three spatial dimensions. Soil texture type and topographic elevation were used as covariates during the interpolation process to construct a variation function model. The weighting coefficients were determined based on the variation function model, and the salinity value of each grid node was calculated to generate a three-dimensional salinity distribution field covering the entire orchard area.

7. The method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion according to claim 1, characterized in that, The extraction of dynamic salinity change characteristics based on a three-dimensional salinity distribution field with multiple consecutive time periods specifically includes: The difference field of the three-dimensional salinity distribution field between adjacent time periods is calculated, and the difference field is segmented by threshold to identify regions with significant salinity changes. For each region with significant changes, the time series of salinity changes in that region is statistically analyzed. The periodic component and trend component of the time series are fitted using an autoregressive moving average model, and the slope of the trend component is taken as the rate of dynamic change of salinity.

8. The method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion according to claim 7, characterized in that, The method of fitting the periodic and trend components of the time series using an autoregressive moving average model is as follows: The time series of salinity changes is tested for stationarity. If it is a non-stationary series, it is subjected to first-order differencing. The number of autoregressive terms and the number of moving average terms are determined based on the autocorrelation function and the partial autocorrelation function. The model parameters are estimated using the least squares method to obtain the fitted model. The amount of salt change in the next period is predicted based on the fitted model. The stationarity test uses the ADF test. If the test statistic is greater than the critical value, the series is determined to be non-stationary. In this case, the time series is differencing once and tested again until the series is stationary.

9. The method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion according to claim 1, characterized in that, The obtained dynamic salt content detection results are as follows: Based on the dynamic rate of salinity change in each significant change area and the three-dimensional salinity distribution field of the current period, a salinity dynamic change trend map is drawn, and areas with a change rate exceeding a preset threshold are marked as key areas of concern. A detection report containing the location, area, and rate of change of key areas of concern is output.

10. The method for dynamic detection of soil salinity in saline-alkali orchards based on multi-source sensor fusion according to claim 5, characterized in that, The window size for median filtering is determined based on the distance between adjacent sensors on the same depth layer, and is set to 1.2 times the number of grid points corresponding to the sensor distance, and the window is square.

Images