Method and system for constructing rice growth model based on salt stress

Abstract

The present invention provides a method and system for constructing a rice growth model based on salt stress, relating to the technical field of growth model construction. The method first sets up monitoring points in a planting sample area and obtains soil conductivity data at different depths using multi-depth soil conductivity sensors and the Kriging interpolation method. Root density, multispectral remote sensing imagery, and leaf area index are collected to calculate the canopy salt stress index. The sodium adsorption ratio is determined by measuring relevant ion concentrations, and a salt stress coefficient is established by combining soil conductivity and irrigation volume. The canopy and soil stress indicators are integrated to generate a comprehensive stress index, and a salt feedback term is introduced to characterize crop compensation capacity. Finally, a hybrid machine learning model is constructed using a convolutional neural network and a long-short-term memory network to achieve dynamic prediction of rice growth. The present invention achieves multidimensional salt stress quantification in the soil-crop system, providing decision support for salinization control in precision agriculture.

Description

Technical Field

[0001] The present invention relates to the technical field of growth model construction, and in particular to a method and system for constructing a rice growth model based on salt stress. Background Art

[0002] Rice, a major global food crop, is often significantly affected by salt stress when growing in salinized soils. Excessive soil salinity can hinder root water absorption, cause ion toxicity, and lead to nutrient imbalance, which in turn inhibits rice photosynthesis and biomass accumulation. Traditional rice growth models are mostly based on physiological and ecological parameters under ideal environments, which are difficult to accurately reflect the dynamic growth process under salinization conditions, especially the complex relationship between the spatiotemporal variation of soil salinity and crop response. Therefore, there is an urgent need for an accurate model that can quantify the impact of salt stress and predict rice growth.

[0003] Existing technologies often rely on single-dimensional data, such as soil electrical conductivity or remotely sensed vegetation indices, lacking a coordinated soil-root-canopy analysis. Furthermore, the coupling mechanisms of salt dynamics in the soil, adaptive root growth, and canopy photosynthetic efficiency are not fully reflected in growth models. Due to the spatiotemporal heterogeneity of salt stress, traditional static models struggle to accurately simulate its cumulative effects, resulting in significant deviations between predicted results and actual growth conditions.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0005] The object of the present invention is to provide a method and system for constructing a rice growth model based on salt stress, so as to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A method for constructing a rice growth model based on salt stress, comprising the following steps:

[0008] Step 1: Determine the planting sample area and set up a monitoring point in the center of the planting sample area. During the monitoring period, collect soil electrical conductivity at different depths at the monitoring point and record the irrigation amount of the sample planting area where the monitoring point is located. Determine the soil electrical conductivity at different depths through Kriging interpolation.

[0009] Step 2: Collect rice samples at the monitoring points and measure the root density of the rice samples; obtain multispectral remote sensing images of the planting sample area and analyze the reflectance of different bands in the multispectral remote sensing images to obtain the salt stress index of the canopy response of the planting sample area and the leaf area index of the planting sample area;

[0010] Step 3: Measure the concentration of relevant ions at the monitoring points to obtain the sodium adsorption ratio of the planting sample area, and analyze the salt stress coefficient of the planting sample area in combination with soil conductivity and irrigation amount;

[0011] Step 4: Combine the canopy-response salt stress index and the soil-driven salt stress coefficient for joint calibration to determine the comprehensive stress index for the planting sample area; determine the salt feedback term based on the relationship between soil electrical conductivity and the salt activation threshold;

[0012] Step 5: The convolutional neural network model was used to analyze the soil electrical conductivity and root density of the planting sample area, and the long short-term memory network model was used to analyze the sodium adsorption ratio, stress comprehensive index, and leaf area index of the planting sample area to construct a rice growth model of the hybrid machine learning model.

[0013] Furthermore, determining the soil electrical conductivity at different depths specifically includes:

[0014] The rice planting area is evenly divided, and a certain area is randomly selected as the planting sample area. The center point of the planting sample area is determined as the monitoring point. Sensors are arranged at different depths in the soil at the monitoring point along the vertical direction to collect soil conductivity data. Kriging interpolation is performed on the conductivity data at discrete depths. The formula is expressed as:

[0015] ;

[0016] in, represents the soil conductivity at a depth of z meters at time t, Indicates the depth of the i-th sensor The conductivity value measured at time t, represents the depth of the i-th sensor position, i is the sensor index, n is the total number of sensors, and i∈[1,n], is the spatial weight coefficient, which represents the weight of the i-th sensor to the target depth z, where z represents the depth variable and t is the time variable within the monitoring period;

[0017] The soil conductivity at different depths of the monitoring point was integrated, and the ratio of the integral to the total depth was used as the soil conductivity at the monitoring point, which was also used as the soil conductivity of the sample planting area.

[0018] Furthermore, collecting the root density specifically includes:

[0019] Randomly select several rice samples at the monitoring point, dig a 10cm×10cm×50cm soil column with the sample rice as the center, and measure the root length of the sample rice in the laboratory to obtain the root density of the sample rice. The formula is as follows:

[0020] ;

[0021] Among them, RLD(o) is the root density of the oth sample rice, is the root length of the oth sample rice, V(o) is the volume of the soil column measured in the laboratory for the oth sample rice, and o is the sample rice index;

[0022] The mean root density of all rice samples at the monitoring point was taken as the root density of the planting sample area where the monitoring point was located.

[0023] Furthermore, obtaining the salt stress index and leaf area index specifically includes:

[0024] Obtain multispectral remote sensing images of the planting sample area, extract the red edge band reflectance, red band reflectance, and near-infrared band reflectance from the multispectral remote sensing images of the area to generate the salt stress index. The calculation formula is:

[0025] ;

[0026] Among them, SRI(t) represents the salt stress index of the canopy response at time t, R735(t) represents the red edge band reflectance at time t, R660(t) represents the red band reflectance at time t, and R790(t) represents the near infrared band reflectance at time t. is the weight coefficient;

[0027] The quantum sensor was placed horizontally 1m above the rice canopy at the monitoring point in the planting sample area. PAR was incident from above the canopy. A linear photosynthetically active radiation array was placed below the canopy to measure the canopy-projected photosynthetically active radiation to obtain the leaf area index of the planting sample area. The calculation formula is:

[0028] ;

[0029] Where LAI(t) represents the leaf area index at time t, represents the canopy-projected photosynthetically active radiation at time t, represents the incident PAR above the canopy at time t, k is the extinction coefficient, is the solar zenith angle.

[0030] Furthermore, determining the salt stress coefficient specifically includes:

[0031] The relevant ion concentrations include sodium ion concentration, calcium ion concentration, and magnesium ion concentration. A pore water sampler is used to collect the soil solution of the root active layer at the monitoring point, and the sodium ion concentration, calcium ion concentration, and magnesium ion concentration are measured to calculate the sodium adsorption ratio. The calculation formula is:

[0032] ;

[0033] in, is the sodium adsorption ratio at time t, is the sodium ion concentration at time t, is the calcium ion concentration at time t, is the magnesium ion concentration at time t;

[0034] The salt stress coefficient in the soil driving process was analyzed based on soil electrical conductivity and irrigation amount, and the formula is expressed as:

[0035] ;

[0036] in, represents the soil-driven salt stress coefficient at time t, is the initial salt stress coefficient, express Soil conductivity of the sample planting area at each moment, for The amount of irrigation at any given time, is a time variable, and , It represents the time interval from the start of the monitoring period to time t. is the salt accumulation coefficient, is the irrigation leaching coefficient.

[0037] Furthermore, determining the comprehensive stress index specifically includes:

[0038] The comprehensive stress index is calculated by combining the salt stress index of the canopy response and the salt stress coefficient driven by the soil:

[0039] ;

[0040] Among them, CCI represents the comprehensive index of coercion at time t, Root sensitivity index, SRI The salt stress index representing the canopy response at time t.

[0041] Furthermore, determining the salt feedback item specifically includes:

[0042] The calculation formula for determining the salt feedback term is:

[0043] ;

[0044] in, represents the salt feedback term at time t, For maximum compensation potential, is the attenuation coefficient, express Soil conductivity at all times, represents the salt activation threshold, represents the positive function, only when When points accumulate, is a time variable, and .

[0045] Furthermore, constructing the rice growth model specifically includes:

[0046] The mean bioaccumulation of sample rice at the monitoring point during the monitoring period was recorded. The soil conductivity and root density data of the planting sample area were processed using a convolutional neural network. The sodium adsorption ratio, stress comprehensive index and leaf area index data of the planting sample area were processed using a long short-term memory network. The outputs of the convolutional neural network and the long short-term memory network were weightedly fused through an attention mechanism. The weights of the attention mechanism were calculated using a softmax function to output the mean bioaccumulation and salt feedback item at the corresponding moment in the monitoring period.

[0047] The present invention further provides a system for constructing a rice growth model based on salt stress, wherein the system for constructing a rice growth model based on salt stress is used to implement the above-mentioned method for constructing a rice growth model based on salt stress, comprising:

[0048] The soil dynamic monitoring module is used to determine the planting sample area and set a monitoring point in the center of the planting sample area. During the monitoring period, the soil conductivity at different depths at the monitoring point is collected and the irrigation amount of the sample planting area where the monitoring point is located is recorded. The soil conductivity at different depths is determined by the Kriging interpolation method;

[0049] The vegetation remote sensing analysis module is used to collect rice samples at monitoring points and measure the root density of the sample rice; obtain multispectral remote sensing images of the planting sample area, analyze the reflectance of different bands in the multispectral remote sensing images to obtain the salt stress index of the canopy response of the planting sample area, and simultaneously obtain the leaf area index of the planting sample area;

[0050] Soil ion stress quantification module, used to measure the relevant ion concentrations at the monitoring points to obtain the sodium adsorption ratio of the planting sample area, and analyze the salt stress coefficient of the planting sample area in combination with soil conductivity and irrigation amount;

[0051] The salt stress comprehensive assessment module is used to combine the canopy-response salt stress index and the soil-driven salt stress coefficient for joint calibration to determine the comprehensive stress index of the planting sample area; and to determine the salt feedback term based on the relationship between soil conductivity and the salt activation threshold;

[0052] The hybrid growth model construction module is used to analyze the soil conductivity and root density of the planting sample area using a convolutional neural network model, and to analyze the sodium adsorption ratio, stress comprehensive index and leaf area index of the planting sample area using a long short-term memory network model to construct a rice growth model of the hybrid machine learning model.

[0053] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0054] The present invention realizes the accurate simulation and prediction of the rice growth process under saline environment through multi-source data fusion and mechanism-machine learning hybrid modeling. Its beneficial effects are mainly reflected in: this method innovatively combines the dynamic migration of soil salt, root distribution characteristics and canopy spectral response to construct a growth model that comprehensively reflects the impact of salt stress, overcoming the limitation of traditional models that only consider a single factor; by introducing a piecewise growth function and a salt feedback mechanism, it can more accurately simulate the growth response characteristics of rice under different stress levels, significantly improving the growth prediction accuracy in saline environment; the CNN-LSTM hybrid network architecture is used in combination with the attention mechanism to effectively process the complex spatiotemporal heterogeneity data in the soil-plant system, providing a reliable decision-making support tool for the precise cultivation and management of rice in saline-alkali land, which has important theoretical value and practical significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 Schematic diagram of the overall method flow of the present invention;

[0056] Figure 2 is the relationship diagram between salt stress coefficient and comprehensive stress index;

[0057] Figure 3 is the relationship diagram between salt stress index and comprehensive stress index;

[0058] Figure 4 Schematic diagram of the system structure of the present invention. DETAILED DESCRIPTION

[0059] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0060] It should be noted that, unless otherwise defined, the technical or scientific terms used in the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative position relationships. When the absolute position of the object being described changes, the relative position relationship may also change accordingly.

[0061] Example:

[0062] See also Figure 1 , the present invention provides a technical solution:

[0063] A method for constructing a rice growth model based on salt stress, comprising the following steps:

[0064] Step 1: Determine the planting sample area and set up a monitoring point in the center of the planting sample area. During the monitoring period, collect soil electrical conductivity at different depths at the monitoring point and record the irrigation amount of the sample planting area where the monitoring point is located. Determine the soil electrical conductivity at different depths through Kriging interpolation.

[0065] In this embodiment, determining the soil conductivity at different depths specifically includes:

[0066] The rice planting area is evenly divided according to its size. For example, for a 10-hectare rice planting area, a 50m×50m grid is used for division, with a total of 40 unit areas. A certain area is randomly determined to separately divide out an area for subsequent processing. For example, a random number generator is used to extract sample areas from the numbered areas. Experts can also conduct a preliminary assessment of each grid area to avoid using grids at the edge or grids adjacent to drainage outlets as planting sample areas. The center point of the planting sample area is determined as the monitoring point. In the soil at the monitoring point, soil conductivity sensors are arranged at different depths along the vertical direction. For example, five depth layers are set, namely 0-10cm, 10-20cm, 20-30cm, 30-40cm, and 40-50cm.

[0067] Conductivity data is collected by sensors, and Kriging interpolation is performed on the conductivity data at discrete depths. The formula is expressed as:

[0068] ;

[0069] in, represents the soil conductivity at a depth of z meters at time t, Indicates the depth of the i-th sensor The conductivity value measured at time t, represents the depth of the i-th sensor position, i is the sensor index, n is the total number of sensors, and i∈[1,n], is the spatial weight coefficient, which represents the weight of the i-th sensor to the target depth z, calculated by the Kriging algorithm, where z represents the depth variable and t is the time variable within the monitoring period;

[0070] In this embodiment, it is necessary to reconstruct the soil conductivity profile at different depths at the monitoring point continuously. Due to the limited number of sensors actually deployed, only conductivity data at discrete depths can be obtained. To obtain the conductivity at any depth, the Kriging interpolation method is required. The Kriging interpolation method uses the measured values ​​of the sensor positions at each known depth to infer the conductivity at the target depth. The contribution of each known point to the target point is determined by the spatial weight coefficient.

[0071] In this embodiment, n depth sensors are arranged vertically at the monitoring point to record soil conductivity at different depths. Kriging interpolation is used to infer the conductivity at a specific depth z. In this embodiment, when calculating the spatial weight coefficient, the soil conductivity data at all sensor depths are used to statistically analyze the conductivity differences at different depth intervals, and a semi-variance map is drawn. Based on the actual data distribution, the existing Gaussian model is selected as the variation function model, and the relationship between depth interval and semi-variance is fitted. The variation function model is used to reflect the spatial autocorrelation characteristics of soil conductivity in the vertical profile.

[0072] For the target depth z, a linear equation system containing n unknown weight coefficients is constructed. Each row of the equation system corresponds to a known sensor depth. The spatial correlation between this depth and all other known depths is given by the variogram. Through the unbiased constraint equation, the sum of all weight coefficients is guaranteed to be 1. The spatial correlation between the target depth z and each known sensor depth, that is, the variogram value, is calculated. The above correlation data is substituted into the Kriging equation system and solved using linear algebra methods to obtain the spatial weight coefficient of each known point for the target depth z.

[0073] The soil conductivity at different depths of the monitoring point was integrated, and the ratio of the integral to the total depth was used as the soil conductivity at the monitoring point, which was also used as the soil conductivity of the sample planting area.

[0074] This part uses gridded uniform sampling and Kriging spatial interpolation methods, combined with real-time monitoring data of soil electrical conductivity at multiple depths, to construct a dynamic model of coupled irrigation, transpiration and vertical salt migration, achieving high-precision quantification of the spatiotemporal distribution of soil salt in rice fields. This provides reliable soil environmental parameter input for subsequent rice salt stress assessment, significantly improving the accuracy of salt migration mechanism analysis and the timeliness of salt damage warning.

[0075] Step 2: Collect rice samples at the monitoring points and measure the root density of the rice samples; obtain multispectral remote sensing images of the planting sample area and analyze the reflectance of different bands in the multispectral remote sensing images to obtain the salt stress index of the canopy response of the planting sample area and the leaf area index of the planting sample area;

[0076] In this embodiment, collecting the root density specifically includes:

[0077] Randomly determine several rice samples at the monitoring point. With the sample rice as the center, dig a 10 cm × 10 cm × 50 cm soil column (i.e., a soil column 10 cm long, 10 cm wide, and 50 cm high). Measure the root length of the sample rice in the laboratory. For example, use the WinRHIZO Pro 2023 root analysis system to scan, set the scanning resolution to 1200dpi, use the high-precision length measurement mode, and focus on length analysis to obtain the root length of the sample rice. Then, use the root length to obtain the root density of the sample rice. The formula is as follows:

[0078] ;

[0079] Among them, RLD(o) is the root density of the oth sample rice, is the root length of the oth sample rice, V(o) is the volume of the soil column measured in the laboratory for the oth sample rice, and o is the sample rice index;

[0080] The mean root density of all rice samples at the monitoring point was taken as the root density of the planting sample area where the monitoring point was located.

[0081] In this embodiment, obtaining the salt stress index and the leaf area index specifically includes:

[0082] Obtain multispectral remote sensing images of the planting sample area, extract the red edge band reflectance (band at 735±10nm), red band reflectance (band at 660±10nm), and near infrared band reflectance (band at 790±10nm) from the multispectral remote sensing images of the area to generate the salt stress index. The calculation formula is:

[0083] ;

[0084] Among them, SRI(t) represents the salt stress index of the canopy response at time t, R735(t) represents the red edge band reflectance at time t, R660(t) represents the red band reflectance at time t, and R790(t) represents the near infrared band reflectance at time t. is the weight coefficient; the weight coefficient is fitted through field test data, and the general value range is [0.15, 0.25].

[0085] The salt stress index, as a dependent variable, comprehensively reflects the physiological response of the rice canopy to salt stress. Its numerical changes directly indicate the severity of salt damage to the crop. This index transforms invisible salt stress into a quantifiable spectral signal by integrating the red-edge band (735nm)'s sensitivity to chlorophyll, the red band (660nm)'s absorption characteristics for photosynthetically active substances, and the near-infrared band (790nm)'s response to leaf cell structure. Its technical benefit is that it enables large-scale, non-destructive salt damage monitoring, overcoming the time-consuming and labor-intensive limitations of traditional chemical analysis and providing real-time data support for salinization early warning in precision agriculture. An increase in the salt stress index reflects physiological disturbances and structural damage to the rice canopy under salt stress. Specifically, decreased red-edge reflectance indicates chlorophyll degradation, abnormal red band reflectance reveals photosynthetic dysfunction, and changes in the near-infrared band reflect cellular dehydration. In the growth model constructed in this paper, a continued increase in the salt stress index indicates an intensified coupling effect between canopy spectral characteristics and soil salt stress. A decrease in the salt stress index indicates a recovery in the rice canopy's physiological state or the expression of salt tolerance, as evidenced by a stabilization of the ratio of the red edge to the red band and a rebound in near-infrared reflectance, indicating a gradual recovery in leaf photosynthetic activity and cellular structure.

[0086] As the core independent variable, the red-edge blue shift phenomenon caused by chlorophyll degradation under salt stress was used. The decrease in its value directly indicates the damage of photosynthetic function; the ratio of the near-infrared to red bands The lambda weighting is introduced to reflect the impact of salt-induced cellular dehydration on leaf structure. The combination of these two independent variables captures the characteristics of salt damage from both physiological and structural dimensions. Their synergistic effect enables the model to distinguish between different stress stages, such as mild dehydration and severe metabolic disorders.

[0087] The difference between the red edge and the red band (R735-R660) in the formula is positively correlated with SRI. When salt damage intensifies, the reduction of chlorophyll will lead to a decrease in the red edge reflectivity and an increase in the absorption of the red band, which will reduce the difference and thus reduce the SRI. The ratio of the near infrared to the red band (R790 / R660) is positively correlated with the SRI through the positive weight coefficient. It is positively correlated with the SRI because salt-induced leaf wilting reduces near-infrared scattering, leading to a decrease in the ratio and a weakening of the SRI. This bidirectional response mechanism results in a monotonically decreasing relationship between the index and salt stress, ensuring that a decrease in the value always corresponds to an increase in stress.

[0088] The quantum sensor was placed horizontally 1m above the rice canopy at the monitoring point in the planting sample area. PAR was incident from above the canopy. A linear photosynthetically active radiation array or a hemispherical canopy analyzer was placed below the canopy. The photosynthetically active radiation projected by the canopy was measured 3 times per hour and the average was taken to obtain the leaf area index of the planting sample area. The calculation formula is:

[0089] ;

[0090] Where LAI(t) represents the leaf area index at time t, represents the canopy-projected photosynthetically active radiation at time t, represents the incident PAR above the canopy at time t, and k is the extinction coefficient, which ranges from [0.5 to 0.7]. The specific value can be determined by referring to the typical value of the extinction coefficient in relevant literature based on the rice variety and planting pattern. is the solar zenith angle, in radians.

[0091] The leaf area index is the dependent variable, which characterizes the total area of ​​rice leaves per unit surface area, and directly reflects the photosynthetic capacity of the crop population and the density of the canopy structure. The ratio of canopy transmitted photosynthetically active radiation to incident radiation is the core independent variable, and its logarithmic transformation value is directly related to the probability of intercepting light by leaves. The extinction coefficient is used as a parameter of leaf spatial distribution to correct the shadow effect caused by leaf overlap; the cosine term of the solar zenith angle corrects the effect of solar altitude on the light path length. The combination of these three independent variables constructs a physical model of LAI from the three dimensions of light energy transmission efficiency, canopy structure and environmental geometry to ensure the universality of calculations under different growth periods and cultivation modes. PAR penetration ( ) is negatively correlated with LAI, because increased leaf density will enhance light interception and lead to reduced transmittance; the extinction coefficient k is negatively correlated with the calculated LAI value, and the upright leaf canopy (with smaller k) requires a higher actual leaf area to achieve the same transmittance; an increase in the solar zenith angle θ (a decrease in cosθ) will amplify the calculated LAI value due to the extension of the optical path.

[0092] This section achieves the coordinated quantification of rice underground root structure and aboveground canopy physiological responses by combining destructive root sampling with non-destructive remote sensing monitoring technology. The fusion of high-precision root scanning and multispectral salt stress index not only establishes a correlation model of soil salt-root morphology-canopy spectral characteristics, but also verifies the actual impact of salt stress on photosynthetic capacity through dynamic leaf area index measurement, thereby providing reliable input with both temporal and spatial resolution for the coupling of physiological parameters of subsequent growth models, significantly improving the systematicity and accuracy of salt stress assessment.

[0093] Step 3: Measure the concentration of relevant ions at the monitoring points to obtain the sodium adsorption ratio of the planting sample area, and analyze the salt stress coefficient of the planting sample area in combination with soil conductivity and irrigation amount;

[0094] In this embodiment, determining the salt stress coefficient specifically includes:

[0095] The relevant ion concentrations include sodium ion concentration, calcium ion concentration, and magnesium ion concentration. A Rhizon SMS pore water sampler (pore size 0.1 μm) was used to collect soil solution in the root active layer at the monitoring point. The samples were collected 24 hours after irrigation to avoid the dilution effect of irrigation water. Three repeated samples were collected each time, and the average of the measured data was used in subsequent calculations. The sodium ion concentration, calcium ion concentration, and magnesium ion concentration were measured by atomic absorption spectroscopy or ion chromatography to calculate the sodium adsorption ratio. The calculation formula is:

[0096] ;

[0097] in, is the sodium adsorption ratio at time t, is the sodium ion concentration at time t, is the calcium ion concentration at time t, is the magnesium ion concentration at time t.

[0098] The sodium adsorption ratio (SAR) quantifies the competitive advantage of sodium ions over divalent cations (calcium and magnesium) in soil solution, and its value directly reflects the risk of soil sodicity. By characterizing the potential impact of sodium ions on the dispersion capacity of soil colloids, this indicator can predict the potential for soil structural deterioration, such as compaction or decreased permeability. Its technical benefit lies in simplifying complex soil ion interactions into a single, comparable parameter, providing a scientific basis for irrigation water quality assessment and soil conditioner application in saline-alkali land improvement. As the numerator, an increase in sodium ion concentration directly enhances the exchange between sodium ions and soil colloids. As the denominator, calcium and magnesium ion concentrations, introduced in the form of square roots, reflect the competitive inhibition of sodium ion adsorption by divalent cations. The SAR value is positively correlated with sodium ion concentration, with an increase in sodium ion concentration leading to a linear increase. It is negatively correlated with calcium and magnesium ion concentrations, with an increase in calcium and magnesium ion concentration leading to a gradual decrease in the SAR value.

[0099] The salt stress coefficient in the soil driving process was analyzed based on soil electrical conductivity and irrigation amount, and the formula is expressed as:

[0100] ;

[0101] in, represents the soil-driven salt stress coefficient at time t, is the initial salt stress coefficient, express Soil conductivity of the sample planting area at each moment, for The irrigation amount at the moment refers to the irrigation amount of the sample planting area, which can be obtained by recording the real-time irrigation water amount through the flow meter, water meter or irrigation system of the farmland. is a time variable, and The monitoring period is 90-120 days, which is not less than a complete rice growth period. 0 represents the starting time of the monitoring period. It represents the time interval from the start of the monitoring period to time t. is the salt accumulation coefficient, and its value range is [0.05, 0.15]. is the irrigation leaching coefficient, with a range of [0.1, 0.3]. The initial salt stress base can be determined through a soil background survey before planting, such as the saturated extraction method.

[0102] The salt stress coefficient, as the dependent variable, dynamically characterizes the cumulative stress intensity of soil salt dynamics on the rice root system. Its numerical integration combines the salt input reflected by soil electrical conductivity with the salt leaching effect dominated by irrigation volume, and quantifies the temporal aggravation or alleviation trend of salt damage through time integration. This coefficient converts discrete salt monitoring data into a continuous stress intensity index by integrating the time-varying relationship between soil electrical conductivity and irrigation volume. Its technical effect is to overcome the limitation that a single measurement cannot reflect the dynamic process of salt, and provide a quantitative basis for precise irrigation decision-making. The increase in the salt stress coefficient reflects the worsening trend of continuous accumulation of soil salt, indicating that the salt input rate exceeds the leaching capacity, resulting in increased osmotic stress and ion toxicity in the root zone. The decrease in the salt stress coefficient indicates the effective removal of salt from the root zone by irrigation leaching or natural precipitation, which manifests as a benign dynamic dominated by salt output.

[0103] Soil electrical conductivity (EC) serves as a positive driving factor, its integral value reflecting the continuous accumulation of salt in the root zone. The coefficient k1 adjusts the salt adsorption capacity of soils of different textures. Irrigation rate serves as a negative regulating factor, its integral value reflecting the effectiveness of leaching on salt removal. The coefficient k2 reflects the leaching efficiency of irrigation water quality (e.g., freshwater / brackish water). The SSI is positively correlated with soil electrical conductivity (EC). An increase in EC increases the SSI growth rate by a factor of k1. It is negatively correlated with irrigation rate (I), but is nonlinearly regulated by k2, so as I increases, the SSI decreases.

[0104] By dynamically integrating multi-source data such as soil electrical conductivity, irrigation volume, and sodium adsorption ratio, a quantitative model of the salt stress coefficient was constructed, achieving a technological leap from static salt monitoring to dynamic stress assessment. The core effect of this step is to convert discrete soil salt indicators into continuously traceable stress cumulative effects, which not only reflects the dynamic migration law of salt in the time dimension, but also reveals the potential harm of salt composition to the physical and chemical properties of the soil through the sodium adsorption ratio. Specifically, the SSI model quantifies the net effect of salt input and leaching output through integral operations, so that the assessment of salt stress is no longer limited to instantaneous concentrations, but covers the cumulative effects during the crop growth cycle. At the same time, the introduction of SAR further distinguishes the differences in the harm of sodium salt and other salts, and provides ion-specific guidance for the precise improvement of saline-alkali land. This multi-dimensional salt damage assessment system has significantly improved the targeted water management and soil improvement in salt-tolerant rice cultivation.

[0105] Step 4: Combine the canopy-response salt stress index and the soil-driven salt stress coefficient for joint calibration to determine the comprehensive stress index for the planting sample area; determine the salt feedback term based on the relationship between soil electrical conductivity and the salt activation threshold;

[0106] In this embodiment, determining the stress comprehensive index specifically includes:

[0107] The comprehensive stress index is calculated by combining the salt stress index of the canopy response and the salt stress coefficient driven by the soil:

[0108] ;

[0109] Among them, CCI represents the comprehensive index of coercion at time t, Root sensitivity index, SRI The salt stress index represents the canopy response at time t. The root sensitivity index can be determined by setting different salt gradients in a controlled experiment, measuring the canopy spectral response and root biomass changes, and fitting the optimal weight using the least squares method.

[0110] The stress index, as the dependent variable, is a composite indicator derived from a weighted fusion of canopy spectral response and soil salinity dynamics. Its numerical changes directly represent the overall physiological response of rice to salt stress. This index transcends the limitations of traditional single-dimensional monitoring, capturing both immediate signals of canopy photosynthetic impairment and the cumulative effects of salt accumulation in the root zone. Its core technology lies in establishing a coordinated quantitative model for aboveground and belowground stress responses.

[0111] The canopy salt stress index is a spectral response item, and its physical basis is the destruction of leaf cell structure caused by salt damage, namely the change of near-infrared reflectance and degradation of photosynthetic pigments, namely the red edge shift; the soil salt stress coefficient is a root zone environmental item, which quantifies the continuous effects of sodium ion toxicity and osmotic stress. The two are combined through the root sensitivity index. Dynamic Coupling: >0.5 emphasizes root-dominated ion toxicity (such as the tillering stage). When the value is <0.5, it focuses on canopy-dominated photosynthetic inhibition (such as during the grain filling period). This design reflects the transmission law of salt stress from underground initiation to aboveground manifestation, making the model adaptive to changes in sensitivity during crop growth stages.

[0112] CCI is positively correlated with SSI but is affected by Adjust, when =0.6, each 1 unit increase in SSI leads to a 0.6 unit increase in CCI; it is positively correlated with SRI but with a weight of (1- ), the relationship between the two is specific to the growth period - the tillering period A higher value such as 0.6-0.7 makes CCI more dependent on SSI, This dynamic weight distribution reveals the physiological laws governing the evolution of salt damage: early on, root ion absorption impairment is the primary cause, while later on, a decrease in canopy photosynthetic efficiency dominates.

[0113] In this example, 30 sets of salt stress coefficient, salt stress index and comprehensive stress index sample data were collected, and the root sensitivity index was set to 0.6. The specific sample data are shown in the following table:

[0114] Table 1: Sample data collection table for comprehensive stress index

[0115]

[0116] refer to Figure 2-3 As can be seen from the data in the table above, as the salt stress coefficient and salt stress index increase, the comprehensive stress index also increases. Changes in the comprehensive stress index directly reflect the overall physiological response of rice to salt stress. As a spectral response term, an increase in the salt stress index reflects the visible salt damage response of the rice canopy's physiological state. Essentially, this is the immediate damage to the leaf cell structure and photosynthetic system caused by salt. The increase in the salt stress coefficient reflects the temporal and spatial cumulative effects of salt accumulation in the root zone, and its physical significance is the sustained intensity of ionic stress in the soil solution. Both are positively correlated with the comprehensive stress index.

[0117] In this embodiment, determining the salt feedback item specifically includes:

[0118] The calculation formula for determining the salt feedback term is:

[0119] ;

[0120] in, represents the salt feedback term at time t, is the maximum compensation potential, which indicates the self-regulation ability of crops under mild stress, and its value range is [0.1, 0.3]. is the attenuation coefficient, which characterizes the irreversible degree of salt damage, such as sandy soil ,clay , which can be determined through relevant references; express Soil conductivity at all times, It represents the salt activation threshold, which is determined by measuring the inflection point of rice root elongation speed in hydroponic experiments. represents the positive function, only when When points accumulate, is a time variable, and .

[0121] The salt feedback term, as the dependent variable, quantifies the attenuation of rice's physiological compensation capacity to salt stress. Its numerical value dynamically reflects the degree of exhaustion of the crop's self-regulation potential under sustained salt stress. This indicator converts soil salt excess ( ) is converted into a quantitative description of the crop growth compensation capacity, revealing that short-term salt excess may be compensated by crops, but long-term accumulation will lead to irreversible damage.

[0122] The integral term is the core independent variable, and its physical essence is the accumulation of salt stress dose effects, reflecting the irreversible nature of stress damage; the maximum compensation potential Reflects the genetic characteristics of varieties (e.g. salt-tolerant varieties >0.25), the attenuation coefficient characterizes the buffering capacity of the soil and crop system. The combination of these three independent variables is derived from stress dose (integral term), genetic potential ( ) and environmental buffer ( ) three dimensions to construct a salt damage cumulative response model, the innovation of which is to incorporate instantaneous salt concentration, duration and variety tolerance into a unified framework for evaluation.

[0123] When the integral term decreases, it indicates that the soil conductivity is close to the salt activation threshold, which reflects that the rice is gradually escaping from the critical state of salt stress and the physiological metabolism of the crop is tending to normal. It is negatively correlated with the cumulative amount of excess salt. As the integral term increases, the index value will become smaller, and the corresponding salt feedback term will decrease, reflecting that the continuous accumulation of salt stress has exceeded the compensatory capacity of crops, causing irreversible physiological damage.

[0124] Step 5: The convolutional neural network model was used to analyze the soil electrical conductivity and root density of the planting sample area, and the long short-term memory network model was used to analyze the sodium adsorption ratio, stress comprehensive index, and leaf area index of the planting sample area to construct a rice growth model of the hybrid machine learning model.

[0125] In this embodiment, constructing the rice growth model specifically includes:

[0126] Record the average bioaccumulation of rice samples at the monitoring point during the monitoring period. Bioaccumulation refers to the total dry matter accumulated by rice through photosynthesis and nutrient absorption during the monitoring period, including the total dry matter of the above-ground parts (stems, leaves, and panicles) and underground parts (roots). It reflects the crop's ability to convert light energy, water, and inorganic salts into organic matter. The rice samples were processed. The above-ground parts were cut to the ground level, including all stems, leaves, and panicles, and placed in marked bags. The underground parts were measured by root length. After drying and weighing in the laboratory, the above-ground dry weight and root dry weight were obtained to calculate the bioaccumulation. The formula is referenced. .

[0127] The input data of the rice growth model are soil conductivity, root density, sodium adsorption ratio, stress comprehensive index, and leaf area index of the planting sample area, and the output data are the mean bioaccumulation and salt feedback items of the planting sample area during the corresponding monitoring period. All data within the monitoring period are divided into training, validation, and test sets. The rice growth model is trained using the data in the training set. A convolutional neural network is used to process the soil conductivity and root density data of the planting sample area. The input data of the convolutional neural network model is a three-dimensional matrix containing three dimensions: time, depth, and features. The feature dimension includes two channels: soil conductivity and root density. The depth resolution is 1 cm and the time resolution is 1 hour. The discrete sensor data is converted into a continuous depth-time conductivity distribution matrix through the Kriging interpolation method. At the same time, the root density data is also interpolated and aligned at the same depth resolution. The convolutional neural network model extracts the spatiotemporal characteristics of the vertical migration of soil salt through convolutional layers and pooling layers, and ultimately outputs a 320-dimensional feature vector. A long-short-term memory (LSTM) network (LSTM) was used to process data on the sodium adsorption ratio, stress index, and leaf area index (LAI) of the planted sample area. The sodium adsorption ratio and stress index were sampled hourly, while the LAI was measured three times daily and averaged. The LAI was then expanded from three daily measurements to hourly measurements using linear or cubic spline interpolation. The input data consisted of the hourly sampled sodium adsorption ratio, stress index, and interpolated hourly LAI. These data together formed a multivariate time series. The LSTM model, using its unique gating mechanism, captured the long-term cumulative effects of salt stress and its lagged impact on crop growth. A 30-day sliding window was used to ensure the model could identify the dynamic evolution of salt stress. After processing by the LSTM model, a 128-dimensional feature vector was output, encoding the complex relationship between the temporal changes in salt stress and canopy physiological responses.

[0128] The convolutional neural network branch includes a convolution layer, a maximum pooling layer, and a flattening layer. The convolution layer uses a 3×3 convolution kernel, the activation function is ReLU, and the pooling size of the pooling layer is 2. The long short-term memory network branch includes an LSTM layer with 128 units and dropout=0.2. The fusion layer is an attention mechanism layer with softmax weight calculation method. The output layer is a fully connected layer with 2 neurons and a linear activation function. The output of the convolutional neural network model (320 dimensions) and the output of the long short-term memory network model (128 dimensions) are spliced ​​into a 448-dimensional feature vector. The attention score is calculated through the trainable parameter matrix, and then the weighted fusion result is output based on the attention score of each model to obtain the mean bioaccumulation and salt feedback term at each moment in the monitoring period. The mean square error of the bioaccumulation mean prediction and the Huber loss of the salt feedback term are used as loss functions. The Adam optimizer sets the initial learning rate to 0.001. , , set the early stopping mechanism, i.e. patience = 15epoch; the mean bioaccumulation The mean absolute error of the salt feedback term is used as the evaluation index. ,When the mean absolute error of the salt feedback term is less than 0.1, the model performance is considered to be excellent, otherwise the model is retrained.

[0129] See also Figure 2 The present invention further provides a system for constructing a rice growth model based on salt stress, wherein the system for constructing a rice growth model based on salt stress is used to implement the above-mentioned method for constructing a rice growth model based on salt stress, comprising:

[0130] The soil dynamic monitoring module is used to determine the planting sample area and set a monitoring point in the center of the planting sample area. During the monitoring period, the soil conductivity at different depths at the monitoring point is collected and the irrigation amount of the sample planting area where the monitoring point is located is recorded. The soil conductivity at different depths is determined by the Kriging interpolation method;

[0131] The vegetation remote sensing analysis module is used to collect rice samples at monitoring points and measure the root density of the sample rice; obtain multispectral remote sensing images of the planting sample area, analyze the reflectance of different bands in the multispectral remote sensing images to obtain the salt stress index of the canopy response of the planting sample area, and simultaneously obtain the leaf area index of the planting sample area;

[0132] Soil ion stress quantification module, used to measure the relevant ion concentrations at the monitoring points to obtain the sodium adsorption ratio of the planting sample area, and analyze the salt stress coefficient of the planting sample area in combination with soil conductivity and irrigation amount;

[0133] The salt stress comprehensive assessment module is used to combine the canopy-response salt stress index and the soil-driven salt stress coefficient for joint calibration to determine the comprehensive stress index of the planting sample area; and to determine the salt feedback term based on the relationship between soil conductivity and the salt activation threshold;

[0134] The hybrid growth model construction module is used to analyze the soil conductivity and root density of the planting sample area using a convolutional neural network model, and to analyze the sodium adsorption ratio, stress comprehensive index and leaf area index of the planting sample area using a long short-term memory network model to construct a rice growth model of the hybrid machine learning model.

[0135] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters in the formulas are set by technicians in this field according to actual conditions.

[0136] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software depends on the specific application and design constraints of the technical solution.

[0137] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, and may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment as needed.

[0138] The above is only a specific implementation method of the present application, but the scope of protection of the present application is not limited thereto. Any technician familiar with this technical field can easily think of changes or replacements within the technical scope disclosed in this application, which should be covered by the scope of protection of the present application.

Claims

1. A method for constructing a rice growth model based on salt stress, characterized in that: The specific steps include: Step 1: Determine the planting sample area and set up a monitoring point in the center of the planting sample area. During the monitoring period, collect soil electrical conductivity at different depths at the monitoring point and record the irrigation amount of the sample planting area where the monitoring point is located. Determine the soil electrical conductivity at different depths through Kriging interpolation. Step 2: Collect rice samples at the monitoring points and measure the root density of the rice samples; obtain multispectral remote sensing images of the planting sample area and analyze the reflectance of different bands in the multispectral remote sensing images to obtain the salt stress index of the canopy response of the planting sample area and the leaf area index of the planting sample area; Step 3: Measure the concentration of relevant ions at the monitoring points to obtain the sodium adsorption ratio of the planting sample area, and analyze the salt stress coefficient of the planting sample area in combination with soil conductivity and irrigation amount; Step 4: Combine the canopy-response salt stress index and the soil-driven salt stress coefficient for joint calibration to determine the comprehensive stress index for the planting sample area; determine the salt feedback term based on the relationship between soil electrical conductivity and the salt activation threshold; Step 5: The convolutional neural network model was used to analyze the soil electrical conductivity and root density of the planting sample area, and the long short-term memory network model was used to analyze the sodium adsorption ratio, stress comprehensive index, and leaf area index of the planting sample area to construct a rice growth model of the hybrid machine learning model.

2. The method for constructing a rice growth model based on salt stress according to claim 1, characterized in that: Determining the soil electrical conductivity at different depths specifically includes: The rice planting area is evenly divided, and a certain area is randomly selected as the planting sample area. The center point of the planting sample area is determined as the monitoring point. Sensors are arranged at different depths in the soil at the monitoring point along the vertical direction to collect soil conductivity data. Kriging interpolation is performed on the conductivity data at discrete depths. The formula is expressed as: ； in, represents the soil conductivity at a depth of z meters at time t, Indicates the depth of the i-th sensor The conductivity value measured at time t, represents the depth of the i-th sensor position, i is the sensor index, n is the total number of sensors, and i∈[1,n], is the spatial weight coefficient, which represents the weight of the i-th sensor to the target depth z, where z represents the depth variable and t is the time variable within the monitoring period; The soil conductivity at different depths of the monitoring point was integrated, and the ratio of the integral to the total depth was used as the soil conductivity at the monitoring point, which was also used as the soil conductivity of the sample planting area.

3. The method for constructing a rice growth model based on salt stress according to claim 1, characterized in that: Collecting the root density specifically includes: Randomly select several rice samples at the monitoring point, dig a 10cm×10cm×50cm soil column with the sample rice as the center, and measure the root length of the sample rice in the laboratory to obtain the root density of the sample rice. The formula is as follows: ； Among them, RLD(o) is the root density of the oth sample rice, is the root length of the oth sample rice, V(o) is the volume of the soil column measured in the laboratory for the oth sample rice, and o is the sample rice index; The mean root density of all rice samples at the monitoring point was taken as the root density of the planting sample area where the monitoring point was located.

4. The method for constructing a rice growth model based on salt stress according to claim 1, wherein: Obtaining salt stress index and leaf area index specifically includes: Obtain multispectral remote sensing images of the planting sample area, extract the red edge band reflectance, red band reflectance, and near-infrared band reflectance from the multispectral remote sensing images of the area to generate the salt stress index. The calculation formula is: ； Among them, SRI(t) represents the salt stress index of the canopy response at time t, R735(t) represents the red edge band reflectance at time t, R660(t) represents the red band reflectance at time t, and R790(t) represents the near infrared band reflectance at time t. is the weight coefficient; The quantum sensor was placed horizontally 1m above the rice canopy at the monitoring point in the planting sample area. PAR was incident from above the canopy. A linear photosynthetically active radiation array was placed below the canopy to measure the canopy-projected photosynthetically active radiation to obtain the leaf area index of the planting sample area. The calculation formula is: ； Where LAI(t) represents the leaf area index at time t, represents the canopy-projected photosynthetically active radiation at time t, represents the incident PAR above the canopy at time t, k is the extinction coefficient, is the solar zenith angle.

5. The method for constructing a rice growth model based on salt stress according to claim 1, wherein: Determining the salt stress coefficient specifically includes: The relevant ion concentrations include sodium ion concentration, calcium ion concentration, and magnesium ion concentration. A pore water sampler is used to collect the soil solution of the root active layer at the monitoring point, and the sodium ion concentration, calcium ion concentration, and magnesium ion concentration are measured to calculate the sodium adsorption ratio. The calculation formula is: ； in, is the sodium adsorption ratio at time t, is the sodium ion concentration at time t, is the calcium ion concentration at time t, is the magnesium ion concentration at time t; The salt stress coefficient in the soil driving process was analyzed based on soil electrical conductivity and irrigation amount, and the formula is expressed as: ； in, represents the soil-driven salt stress coefficient at time t, is the initial salt stress coefficient, express Soil conductivity of the sample planting area at each moment, for The amount of irrigation at any given time, is a time variable, and , It represents the time interval from the start of the monitoring period to time t. is the salt accumulation coefficient, is the irrigation leaching coefficient.

6. The method for constructing a rice growth model based on salt stress according to claim 5, characterized in that: Determining the comprehensive stress index specifically includes: The comprehensive stress index is calculated by combining the salt stress index of the canopy response and the salt stress coefficient driven by the soil: ； Among them, CCI represents the comprehensive index of coercion at time t, Root sensitivity index, SRI The salt stress index representing the canopy response at time t.

7. The method for constructing a rice growth model based on salt stress according to claim 6, characterized in that: Determining the salt feedback item specifically includes: The calculation formula for determining the salt feedback term is: ； in, represents the salt feedback term at time t, For maximum compensation potential, is the attenuation coefficient, express Soil conductivity at all times, represents the salt activation threshold, represents the positive function, only when When points accumulate, is a time variable, and .

8. The method for constructing a rice growth model based on salt stress according to claim 1, characterized in that: Constructing the rice growth model specifically includes: The mean bioaccumulation of sample rice at the monitoring point during the monitoring period was recorded. The soil conductivity and root density data of the planting sample area were processed using a convolutional neural network. The sodium adsorption ratio, stress comprehensive index and leaf area index data of the planting sample area were processed using a long short-term memory network. The outputs of the convolutional neural network and the long short-term memory network were weightedly fused through an attention mechanism. The weights of the attention mechanism were calculated using a softmax function to output the mean bioaccumulation and salt feedback item at the corresponding moment in the monitoring period.

9. A rice growth model construction system based on salt stress, characterized in that: The system for constructing a rice growth model based on salt stress is used to implement the method for constructing a rice growth model based on salt stress according to any one of claims 1 to 8, comprising: The soil dynamic monitoring module is used to determine the planting sample area and set a monitoring point in the center of the planting sample area. During the monitoring period, the soil conductivity at different depths at the monitoring point is collected and the irrigation amount of the sample planting area where the monitoring point is located is recorded. The soil conductivity at different depths is determined by the Kriging interpolation method; The vegetation remote sensing analysis module is used to collect rice samples at monitoring points and measure the root density of the sample rice; obtain multispectral remote sensing images of the planting sample area, analyze the reflectance of different bands in the multispectral remote sensing images to obtain the salt stress index of the canopy response of the planting sample area, and simultaneously obtain the leaf area index of the planting sample area; Soil ion stress quantification module, used to measure the relevant ion concentrations at the monitoring points to obtain the sodium adsorption ratio of the planting sample area, and analyze the salt stress coefficient of the planting sample area in combination with soil conductivity and irrigation amount; The salt stress comprehensive assessment module is used to combine the canopy-response salt stress index and the soil-driven salt stress coefficient for joint calibration to determine the comprehensive stress index of the planting sample area; and to determine the salt feedback term based on the relationship between soil conductivity and the salt activation threshold; The hybrid growth model construction module is used to analyze the soil conductivity and root density of the planting sample area using a convolutional neural network model, and to analyze the sodium adsorption ratio, stress comprehensive index and leaf area index of the planting sample area using a long short-term memory network model to construct a rice growth model of the hybrid machine learning model.

Images