Artificial intelligence rice water and fertilizer real-time monitoring method and system

Abstract

The invention discloses an artificial intelligence rice water and fertilizer real-time monitoring method and system, and relates to the technical field of agricultural intelligent decision making, and the method comprises the steps: inputting a farmland feature data set into a soil thermodynamic diagram generation model, carrying out the high-resolution reconstruction of the farmland feature data set through a GAN adversarial network, and generating a whole-field high-precision soil thermodynamic diagram; based on the whole-field high-precision soil thermodynamic diagram, a collision relation between the fertilization amount and historical farming data is detected according to an FCL causal algorithm, a preliminary causal diagram is generated, and a causal diagram structure of the fertilization amount and the historical farming data is constructed by adopting a multiple linear regression method; based on a causal diagram structure, the multi-order causal effect of the fertilization amount, the soil parameters and the historical yield is analyzed through a dynamic allocation algorithm, and a water and fertilizer regulation and control strategy is formulated in combination with a multi-objective optimization algorithm. According to the method, the soil thermodynamic diagram generation model is constructed, so that the fuzzy problem of the edge of the field and the salinization area is solved, a high-fidelity soil space state substrate is provided for water and fertilizer regulation and control, and invalid irrigation is reduced.

Description

Technical Field

[0001] The present invention relates to the field of agricultural intelligent decision-making technology, and in particular to an artificial intelligence-based real-time monitoring method and system for rice water and fertilizer. Background Art

[0002] In recent years, data-driven precision control technologies in rice cultivation have rapidly developed. These technologies rely on the integrated analysis and intelligent decision-making of multi-source sensor data. With the widespread adoption of IoT sensors, low-altitude remote sensing platforms, and farmland robots, the ability to collect key soil parameters (such as conductivity, moisture content, and organic matter content) in real time has significantly improved, enabling local monitoring with centimeter-level spatial resolution.

[0003] Existing methods for spatially reconstructing soil parameters often rely on low-resolution remote sensing data, resulting in blurring or distortion in heat map generation models at the edges of fields, making it difficult to accurately characterize localized soil anomalies such as salinization and compaction. Furthermore, existing causal inference methods often employ static correlation analysis, failing to capture the bidirectional feedback mechanism between fertilization operations and the dynamic changes in soil parameters. This results in insufficient timeliness and robustness in regulatory strategies. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an artificial intelligence-based real-time monitoring method for rice water and fertilizer to solve the problems of insufficient accuracy in soil parameter reconstruction and lack of a causal feedback mechanism.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an artificial intelligence-based real-time monitoring method for rice water and fertilizer, comprising:

[0008] The farmland characteristic dataset is input into the soil thermodynamic map generation model, and the farmland characteristic dataset is reconstructed at high resolution through the GAN adversarial network to generate a high-precision soil thermodynamic map of the entire field. Based on the high-precision soil thermodynamic map of the entire field, the collision relationship between fertilizer application amount and historical agricultural data is detected according to the FCL causal algorithm to generate a preliminary causal graph, and the multiple linear regression method is used to construct the causal graph structure of fertilizer application amount and historical agricultural data. Based on the causal graph structure, the multi-order causal effects of fertilizer application amount, soil parameters and historical yield are analyzed through a dynamic allocation algorithm, and a water and fertilizer control strategy is formulated in combination with a multi-objective optimization algorithm. According to the water and fertilizer control strategy, the upper limit of fertilizer application amount and the baseline value of irrigation amount are defined for each farmland characteristic block, and the optimal combination parameter table of fertilization and irrigation is generated. According to the irrigation rules, the regional fertilization and irrigation instructions are output and executed, and an execution effect report is generated.

[0009] As a preferred embodiment of the artificial intelligence-based real-time monitoring method for rice water and fertilizer of the present invention, the following steps are described:

[0010] The farmland characteristic dataset includes soil parameters, image data and historical farming data;

[0011] The historical agricultural data includes historical yield, fertilizer application amount and irrigation amount.

[0012] As a preferred embodiment of the artificial intelligence-based real-time monitoring method for rice water and fertilizer of the present invention, the following steps are described:

[0013] The specific steps of constructing the soil thermal map generation model are as follows:

[0014] Use the encoder and decoder to build the generator structure, use the convolution kernel cascade architecture to build the discriminator structure, couple the generator structure and the discriminator structure through the adversarial training framework to form a GAN adversarial network, and use the gradient fusion algorithm to build the splicing layer;

[0015] A soil heat map generation model is constructed based on the GAN adversarial network and splicing layer.

[0016] As a preferred embodiment of the artificial intelligence-based real-time monitoring method for rice water and fertilizer of the present invention, the following steps are described:

[0017] The GAN adversarial network is used to reconstruct the farmland feature dataset at high resolution and generate a high-precision soil thermal map of the entire field. The specific steps are as follows:

[0018] The GAN adversarial network is used to divide the farmland feature dataset into pixel blocks, generate farmland feature blocks, and then use the generator structure to perform high-resolution reconstruction to generate soil thermal map blocks;

[0019] The stitching layer stitches the soil thermal map blocks, uses the central difference method to identify high mutation areas, and uses the soil moisture diffusion equation to optimize the high mutation areas to generate a high-precision soil thermal map of the entire field.

[0020] As a preferred embodiment of the artificial intelligence-based real-time monitoring method for rice water and fertilizer of the present invention, the following steps are described:

[0021] The collision relationship between the amount of fertilizer applied and the historical farming data is detected according to the FCL causal algorithm, a preliminary causal graph is generated, and a causal graph structure of the amount of fertilizer applied and the historical farming data is constructed using the multivariate linear regression method. The specific steps are as follows:

[0022] A spatiotemporal gridded data fusion method was used to correlate high-precision soil thermal maps of the entire field with fertilizer application rates, soil parameters, and historical yields to form a causal analysis dataset.

[0023] The Pearson correlation coefficient is used to screen the correlation items between fertilizer application amount, soil parameters and historical yield in the causal analysis data set, and the significant variable set is output;

[0024] Using the FCL causal algorithm, we detect collision relationships among significant variable sets and generate a preliminary causal graph of fertilizer application, soil parameters, and historical yields.

[0025] The direct causal effects of the preliminary causal diagram were verified by the multiple linear regression method to generate the causal diagram structure.

[0026] As a preferred embodiment of the artificial intelligence-based real-time monitoring method for rice water and fertilizer of the present invention, the following steps are described:

[0027] Based on the causal graph structure, the dynamic allocation algorithm is used to analyze the multi-order causal effects of fertilizer application amount, soil parameters and historical yield, and a water and fertilizer control strategy is formulated in combination with a multi-objective optimization algorithm. The specific steps are as follows:

[0028] A dynamic allocation algorithm was used to decompose the direct effect of fertilizer application on historical yield and the indirect effect of soil parameters, and the total effect was calculated using a multi-path accumulation method.

[0029] Define the gain coefficient of fertilizer application on historical yield based on the total effect, and define the fertilizer application baseline value and soil parameter safety threshold based on the gain coefficient and soil parameters;

[0030] High conductivity areas and low humidity areas are identified based on the safety thresholds of soil parameters, the fertilizer application amount is adjusted for high conductivity areas and low humidity areas based on the fertilizer application benchmark value, and water and fertilizer control strategies are formulated through irrigation rules.

[0031] As a preferred embodiment of the artificial intelligence-based real-time monitoring method for rice water and fertilizer of the present invention, the following steps are described:

[0032] According to the water and fertilizer control strategy, the upper limit of fertilizer application and the reference value of irrigation amount for each farmland block are defined, the optimal combination parameter table of fertilization and irrigation is generated, and the regional fertilization and irrigation instructions are output and executed according to the irrigation rules, and the execution effect report is generated. The specific steps are as follows:

[0033] According to the water and fertilizer control strategy, combined with the soil safety threshold, the upper limit of fertilizer application and the baseline value of irrigation amount for each farmland characteristic block are determined;

[0034] Using farmland characteristic blocks as index units, a combined table containing fertilizer and irrigation amounts is constructed to generate a fertilization and irrigation instruction table, and the optimal fertilization and irrigation instructions are identified based on the water and fertilizer control strategy.

[0035] Execute optimal fertilization and irrigation instructions, record actual fertilization amount, actual irrigation amount and changes in soil parameters, and generate execution effect reports.

[0036] In a second aspect, the present invention provides an artificial intelligence real-time monitoring system for rice water and fertilizer, comprising a heat map generation module, a cause-effect diagram generation module, a strategy generation module, and an instruction execution module; the heat map generation module is used to input a farmland feature data set into a soil heat map generation model, and perform high-resolution reconstruction of the farmland feature data set through a GAN adversarial network to generate a high-precision soil heat map for the entire field; the cause-effect diagram generation module is used to detect the collision relationship between the fertilizer amount and historical farming data based on the FCL cause-effect algorithm based on the high-precision soil heat map of the entire field, generate a preliminary cause-effect diagram, and use a multiple linear regression method to construct a cause-effect diagram structure of the fertilizer amount and historical farming data; the strategy generation module is used to analyze the multi-order causal effects of fertilizer amount, soil parameters and historical yield through a dynamic allocation algorithm based on the cause-effect diagram structure, and formulate a water and fertilizer control strategy in combination with a multi-objective optimization algorithm; the instruction execution module is used to define the fertilizer amount upper limit and irrigation amount baseline value for each farmland feature block according to the water and fertilizer control strategy, generate an optimal combination parameter table of fertilization and irrigation, output and execute regional fertilization and irrigation instructions according to irrigation rules, and generate an execution effect report.

[0037] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the artificial intelligence real-time monitoring method for rice water and fertilizer as described in the first aspect of the present invention is implemented.

[0038] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the artificial intelligence real-time monitoring method for rice water and fertilizer as described in the first aspect of the present invention is implemented.

[0039] The beneficial effects of the present invention are as follows: by constructing a soil thermodynamic map generation model, the ambiguity problem of field edges and salinized areas is solved, the conductivity gradient mutation area is accurately identified, a high-fidelity soil spatial state basis is provided for water and fertilizer regulation, and ineffective irrigation is reduced; through the causal graph structure constructed based on the FCL causal algorithm, the direct causal effect of fertilizer application amount, soil parameters and yield is verified, and a water and fertilizer regulation strategy is formulated in combination with soil parameter thresholds to ensure a balance between yield maximization and salt control. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0041] Figure 1Flowchart of the artificial intelligence-based real-time monitoring method for rice water and fertilizer.

[0042] Figure 2 Schematic diagram of the artificial intelligence-based real-time monitoring system for rice water and fertilizer.

[0043] Figure 3 Flowchart for generating high-precision soil thermodynamic maps for the entire field.

[0044] Figure 4 Flowchart for cause-effect diagram structure construction and water and fertilizer regulation strategy generation. DETAILED DESCRIPTION

[0045] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0046] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0047] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0048] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides an artificial intelligence real-time monitoring method for rice water and fertilizer, comprising the following steps:

[0049] S1. Input the farmland feature dataset into the soil thermodynamic map generation model, and reconstruct the farmland feature dataset with high resolution through the GAN adversarial network to generate a high-precision soil thermodynamic map of the entire field.

[0050] S1.1. Collect soil parameters, image data, and historical agricultural data, acquiring multi-source data at a preset density and frequency;

[0051] It should be noted that soil conductivity sensors, pH sensors, and temperature and humidity sensors are deployed to collect soil parameters at a frequency of once per minute at a density of one node per 10 mu, generating original soil data tables with timestamps and latitude and longitude coordinates; multi-band images of farmland with a resolution of 10 meters are obtained weekly through multispectral satellite images, and aerial photography of crops during the tillering and heading periods is performed simultaneously using drone visible light images to generate original image data sets; historical agricultural data are extracted from the local agricultural management database, and the fertilizer application amount, irrigation amount, and meteorological records corresponding to the field numbers are linked to generate structured historical agricultural data tables.

[0052] S1.2. Perform missing value filling, outlier removal, timestamp alignment, and spatial grid mapping on multi-source data to generate a spatiotemporally aligned farmland feature dataset.

[0053] It should be noted that the original soil data table, original image dataset and historical agricultural data table were merged according to timestamps, duplicate records and null value fields were removed, and a preliminary multi-source dataset was generated. The preliminary multi-source dataset was interpolated using cubic spline to fill missing values, and soil parameter outliers were removed to generate a cleaned multi-source dataset. The cleaned multi-source dataset was uniformly converted to UTC timestamps and UTM coordinate systems, and the spatiotemporal dimensions were resampled to 0.1-meter grid accuracy to generate a spatiotemporally aligned farmland characteristic dataset.

[0054] S1.3. Use the encoder and decoder to build a generator structure, use the convolutional kernel cascade architecture to build a discriminator structure, couple the generator structure and the discriminator structure through the adversarial training framework to form a GAN adversarial network, use the gradient fusion algorithm to build the splicing layer, and build a soil heat map generation model based on the GAN adversarial network and the splicing layer;

[0055] It should be noted that when constructing the encoder, a 7×7 convolution kernel is used as the input layer, and then batch normalization is performed, and the ReLU activation function is defined; the input feature map size is compressed through a 3×3 maximum pooling layer, and then 4 ResNet blocks are stacked. Each ResNet block consists of two 3×3 convolution layers, and the number of channels increases to 256 in sequence. The residual connection is used to avoid gradient disappearance, forming a hierarchical feature extraction structure; when constructing the decoder, a 4×4 deconvolution layer is used as the initial upsampling layer, and a 3×3 convolution layer is used to refine the input features, and batch Normalize and define the ReLU activation function; repeatedly perform deconvolution and convolution operations to gradually improve the resolution of the input feature map, and reduce the number of channels to 64; finally adjust the number of output channels through a 1×1 convolution layer to complete the decoder construction, and use the encoder and decoder to build the generator structure; adopt a convolution kernel cascade architecture, take the feature map output by the generator structure as input, and extract different feature maps through three parallel convolution kernels of 3×3, 5×5, and 7×7. The three sets of feature maps are spliced ​​along the channel dimension and connected to the cascade convolution unit. The cascade convolution unit consists of The model is composed of 4 groups of 3×3 depth-separable convolutional layers, and the output of each layer is superimposed with the features of the previous layer through jump connections to construct a discriminator structure. The feature map output by the generator structure is input into the discriminator structure, and the classification error of the discriminator structure for the generated image and soil parameters is calculated. At the same time, the adversarial loss function of the generator structure and the classification loss function of the discriminator structure are constructed to form an adversarial optimization target. The gradient penalty mechanism is used to constrain the update of the discriminator structure, and the adversarial training framework coupling is achieved through an alternating iterative training strategy. The trainable computing unit is defined by the gradient fusion algorithm, which receives the feature map output by the encoder and the feature map of the current layer of the decoder. The bilinear interpolation operation is performed on the encoder output feature map to match the spatial size of the feature map of the current layer of the decoder. The encoder feature map and the decoder feature map are then spliced ​​along the channel dimension to generate a fused feature map, and the pixel weight coefficients of the encoder feature map and the decoder feature map are calculated based on the cosine similarity. Finally, the pixel weight coefficients are element-wise multiplied with the fused feature map to generate a dynamically adjusted weighted feature map, completing the construction of the splicing layer. The soil thermal map generation model is constructed based on the generator structure and the splicing layer.

[0056] It should also be noted that when training the soil thermodynamic map generation model, the L1 reconstruction loss of the whole-field high-precision soil thermodynamic map and the measured values ​​is calculated based on the measured soil parameter values ​​in the farmland feature dataset; the classification error of the spatial consistency between the whole-field high-precision soil thermodynamic map and the original soil parameter distribution in the farmland feature dataset is calculated through the discriminator structure of the GAN adversarial network to generate the adversarial loss; the discriminator structure parameters are fixed, and the generator structure is optimized by back propagation to minimize the L1 reconstruction loss and adversarial loss; the generator structure and the discriminator structure are alternately optimized until the soil thermodynamic map generation model converges. The expression of the soil thermodynamic map generation model is,

[0057]

[0058] Among them, H represents the high-precision soil thermal map of the whole field; D l (X) is the decoder function, which represents the feature map of the decoder layer l when the farmland feature dataset X is input; G(X) is the concatenation layer, which represents the fusion feature map when the farmland feature dataset X is input; E l (X) is the encoder function, which represents the feature map of the encoder layer l when the farmland feature dataset X is input; α l Represents the pixel weight coefficient of the feature map of the lth layer of the encoder and decoder; L represents the total number of levels in the encoder and decoder; l represents the level index of the current operation, and the value range is 1-L; d represents the differential element, which is used to mark the level index.

[0059] S1.4. Use the GAN adversarial network to divide the farmland feature dataset into pixel blocks, generate farmland feature blocks, and use the generator structure to perform high-resolution reconstruction to generate soil heat map blocks;

[0060] It should be noted that based on the GAN framework, the farmland feature dataset is divided into 256×256 pixel farmland feature blocks through a sliding window and input into the generator structure. The generator structure extracts multi-scale features based on the encoder, and the splicing layer dynamically weights and fuses the encoder feature map and the decoder feature map. It is reconstructed into a high-resolution soil thermal map block through deconvolution upsampling. The discriminator compares the spatial consistency between the generated thermal map block and the real soil parameter distribution to calculate the adversarial loss. The generator combines the adversarial loss and the L1 reconstruction loss to optimize the generator structure and then outputs the soil thermal map block.

[0061] S1.5. Splice the soil thermodynamic map blocks and locate the splicing boundaries of the soil thermodynamic map blocks using the phase correlation positioning method. Use the central difference method to analyze the gradient values ​​of the soil parameters at the splicing boundaries, identify the high mutation areas, and optimize the high mutation areas using the soil water diffusion equation to generate a high-precision soil thermodynamic map for the entire field.

[0062] It should be noted that the soil thermogram blocks are spliced ​​into a complete thermogram based on the sliding window, and the normalized cross-correlation values ​​of the boundaries of adjacent soil thermogram blocks are calculated by the phase correlation positioning method to determine the optimal matching offset; the normalized cross-correlation calculation is performed on the splicing boundary areas of adjacent soil thermogram blocks to generate a cross-correlation matrix, and the coordinates corresponding to the maximum value in the cross-correlation matrix are the optimal matching offset, indicating that the two soil thermogram blocks are most aligned under the optimal matching offset; after locating the splicing boundary, the gradient values ​​of the soil parameters in the historical agricultural data are statistically analyzed by the statistical distribution method, and the gradient value mean + 3 times the gradient value standard deviation is taken as the gradient mutation threshold, and the gradient mutation threshold = gradient value mean + 3 gradient value standard deviation; the gradient values ​​of the soil parameters on both sides of the splicing boundary are calculated based on the central difference method, and the high mutation area whose gradient value exceeds the gradient mutation threshold is identified; the high mutation area is extracted Soil parameters are used as initial conditions, and the diffusion coefficient and time step are defined by the measured soil texture and soil moisture content. The soil moisture diffusion equation is discretized into a gridded iterative formula using the finite difference method. The natural gradient fluctuation range of soil parameters in historical agricultural data is analyzed using the statistical quantile method, and the 95% quantile is taken as the convergence threshold. In each iterative calculation, the gradient components in the x and y directions of the soil parameter values ​​of each soil thermodynamic map block in the high mutation area are calculated, and the gradient amplitude is calculated based on the gradient components. The average absolute change of the gradient amplitude of all grid points is taken as the soil parameter gradient change. When the soil parameter gradient change is less than the convergence threshold, the optimization is terminated, and the optimized soil parameters are interpolated back to the original high mutation area using the Kriging interpolation method to ensure a smooth transition with the non-mutation area, thereby generating a spatially continuous high-precision soil thermodynamic map of the entire field that conforms to the actual distribution of farmland.

[0063] S2. Based on the high-precision soil thermodynamic map of the entire field, the collision relationship between fertilizer application amount and historical agricultural data was detected according to the FCL causal algorithm, a preliminary causal diagram was generated, and the causal diagram structure of fertilizer application amount and historical agricultural data was constructed using the multivariate linear regression method;

[0064] S2.1. Use spatiotemporal gridded data fusion methods to correlate farmland feature blocks with fertilizer application rates, soil parameters, and historical yields to form a causal analysis dataset.

[0065] It should be noted that based on the soil parameters of each grid in the whole-field high-precision soil thermodynamic map, combined with the fertilizer amount, historical yield and farmland characteristic blocks, the fertilizer amount is aligned with the timestamp of the whole-field high-precision soil thermodynamic map, and the fertilizer amount is allocated to the corresponding grid of the whole-field high-precision soil thermodynamic map through spatial interpolation. The historical yield is mapped to the same spatial resolution according to the boundaries of the farmland characteristic blocks, and a causal analysis dataset is constructed according to grid coordinates. Each row represents a grid unit, and the columns contain soil parameters, fertilizer amount, historical yield and time labels.

[0066] S2.2. Use the Pearson correlation coefficient to screen the correlations between fertilizer application, soil parameters, and historical yields in the causal analysis dataset and output a set of significant variables.

[0067] It should be noted that the linear correlation between fertilizer application and soil parameters and historical yield was calculated using the Pearson correlation coefficient r, with a value range of -1≤r≤1, where r=1 indicates a perfect positive correlation, r=-1 indicates a perfect negative correlation, and r=0 indicates no linear correlation. The significance level threshold α was set based on the Fisher criterion, with a value of α=0.05. The p-value was generated using the chi-square test, and the p-value indicated the probability of observing the current or stronger correlation under the condition that the null hypothesis (fertilizer application is unrelated to soil parameters) holds. The p-value screening rule is that when p<α, the null hypothesis is rejected, and it is considered that there is a significant correlation, and the fertilizer application and soil parameters are not independent. When p≥α, the null hypothesis is accepted, and it is considered that there is no significant correlation, and the fertilizer application and soil parameters are independent. The correlation items with r>0.3 and p<α were screened, and a set of significant variables significantly correlated with fertilizer application, soil parameters and historical yield was output.

[0068] S2.3. Detect collision relationships among significant variable sets using the FCL causal algorithm to generate a preliminary causal diagram for fertilizer application, soil parameters, and historical yields.

[0069] It should be noted that all fertilizer amounts, soil parameters, and historical yields in the significant variable set were connected in pairs as undirected edges to generate an undirected causal skeleton; all triplets containing fertilizer amount, soil parameters, and historical yield in the undirected causal skeleton were traversed, and the partial correlation coefficient test was performed on the fertilizer amount and historical yield in each triplet under the condition of controlling soil parameters. Linear regression analysis was performed on fertilizer amount and historical yield using soil parameters, and the residuals of fertilizer amount and historical yield after regression were extracted. The Pearson correlation coefficient between the residuals of fertilizer amount and historical yield was calculated to obtain the partial correlation coefficient. The p value was calculated based on the partial correlation coefficient. If the fertilizer amount and historical yield after control were still significantly correlated (p < α), the soil parameter was judged as a collision node and connected as a directed edge fertilizer amount → soil parameter ← historical yield; based on all directed edges, a preliminary causal graph with a clear causal direction between fertilizer amount, soil parameters, and historical yield was generated.

[0070] S2.4. Verify the direct causal effect of fertilizer application on historical yield in the preliminary causal diagram using the multivariate linear regression method to generate a causal diagram structure.

[0071] It should be noted that the causal path from fertilizer application to historical yield in the preliminary causal diagram was extracted as the verification target, and the direct causal effect of fertilizer application on historical yield was verified by the multiple linear regression method. Fertilizer application was used as the independent variable and historical yield was used as the dependent variable. The soil parameters directly connected with fertilizer application and historical yield in the preliminary causal diagram were included as covariates in the multiple linear regression method. The regression coefficient of fertilizer application on historical yield was calculated using the least squares method. At the same time, the interference of soil parameters was controlled to eliminate confounding effects. The p-value of the regression coefficient was calculated using the t-test method. If p < α, it means that the regression coefficient of fertilizer application is significant, and the direct causal edge is retained. Otherwise, the causal path is deleted. Finally, the connection status of fertilizer application and historical yield in the preliminary causal diagram is updated according to the test results, and the causal diagram structure is output.

[0072] S3. Based on the causal graph structure, a dynamic allocation algorithm is used to analyze the multi-order causal effects of fertilizer application amount, soil parameters and historical yield, and a water and fertilizer control strategy is formulated in combination with a multi-objective optimization algorithm.

[0073] S3.1. Use a dynamic allocation algorithm to decompose the direct effect of fertilizer application on historical yield and the indirect effect of soil parameters, and calculate the total effect using a multi-path accumulation method;

[0074] It should be noted that a depth-first search is performed starting from the fertilizer application node to traverse all paths pointing to the historical yield node. The local causal effect coefficient is extracted for each node pair on the path. The path coefficient is calculated by multivariate linear regression for multiple intermediary paths. All path coefficients are stored in a sparse matrix in layers according to the path length, where the direct path coefficient is stored in the main diagonal position. The path order is defined by accumulating the number of intermediate nodes from the fertilizer application node to the historical yield node. The indirect path coefficient is filled in the corresponding super-diagonal position according to the path order. Finally, the causal path coefficient matrix containing all fertilizer applications to historical yield is output. The direct effect of fertilizer application on historical yield is represented by the direct path coefficient. All indirect paths containing soil parameters are traversed and the path effect value is calculated. The dynamic weight allocation formula is used to determine the contribution weight of each indirect path. Finally, the multi-path effects are aggregated through the total effect calculation formula to generate the total effect.

[0075] S3.2. Define the gain coefficient of fertilizer application rate on historical yield based on the total effect, and define the fertilizer application rate baseline and soil parameter safety threshold based on the gain coefficient and soil parameters;

[0076] It should be noted that the contribution of each unit increase in fertilizer application to the historical yield is defined as the gain coefficient; the historical yield in each farmland characteristic block is set as the target yield, and the APSIM simulation method is used to input the soil parameters in each farmland characteristic block. The no-fertilizer scenario simulation is run to output the basic yield, and the fertilizer application baseline value is calculated based on the gain coefficient, target yield and basic yield. The fertilizer application baseline value = (target yield-basic yield) / gain coefficient; the conductivity threshold (EC≤3.0dS / m) and moisture threshold (volumetric water content <15%) are set according to FAO recommendations and the soil texture (sandy texture, clay texture) in the farmland characteristic block, and the conductivity threshold and moisture threshold are used as the soil parameter safety thresholds.

[0077] S3.3. Identify high conductivity and low humidity areas based on soil parameter safety thresholds, adjust fertilizer application rates for these areas using baseline fertilizer application rates, and develop water and fertilizer control strategies based on irrigation rules.

[0078] It should be noted that GIS software was used to align the high-precision soil thermal map of the entire field with the real-time sensor data to ensure consistency in spatial position. The discrete sensor data was converted into a continuous spatial layer using the Kriging interpolation method to identify and mark high conductivity areas (such as conductivity = 4.2 dS / m in the southeast of the field) and low humidity areas (such as humidity = 10% in the northwest). Irrigation rules were combined (drip irrigation was used for low humidity areas and leaching irrigation was used for high conductivity areas). A water and fertilizer control strategy was formulated. The leaching water volume in high conductivity areas was calculated using the FAO leaching water requirement, where leaching volume = conventional irrigation volume × (1 + (measured conductivity - conductivity threshold) ÷ conductivity threshold). Irrigation volume was increased in high conductivity areas and reduced in high humidity areas. The nitrogen fertilizer rate is 30%, and potassium fertilizer is added to alleviate salt damage. Drip irrigation is started in low-humidity areas, and water-soluble fertilizers are added according to the fertilizer application benchmark value. Priority rules are formulated based on the water and fertilizer regulation strategy. High priority refers to characteristic farmland blocks with electrical conductivity > 3.0 dS / m, which have salinization risks and inhibit rice root water absorption, or humidity < 15%, resulting in obstructed rice photosynthesis. If any of the conditions is met, high priority is triggered; medium priority refers to characteristic farmland blocks with pH < 6.0, resulting in excessive soil acidity or pH > 7.5, resulting in excessive soil alkalinity, but electrical conductivity ≤ 3.0 dS / m and humidity ≥ 15%; low priority refers to characteristic farmland blocks with electrical conductivity ≤ 3.0 dS / m, humidity ≥ 15%, and pH 6.0 ≤ ≤ 7.5.

[0079] S4. Based on the water and fertilizer control strategy, define the upper limit of fertilizer application and the baseline value of irrigation for each characteristic block of farmland, generate the optimal combination parameter table of fertilization and irrigation, output and execute regional fertilization and irrigation instructions according to the irrigation rules, and generate an execution effect report.

[0080] S4.1. Determine the upper limit of fertilizer application and the baseline value of irrigation for each characteristic area of ​​farmland based on the water and fertilizer control strategy and soil safety threshold;

[0081] It should be noted that based on the priority rule, the real-time monitoring data of each farmland characteristic block is screened through the conductivity threshold and the humidity threshold; combined with the fertilizer application record and the target yield demand, the dynamic threshold adjustment algorithm is used to calculate the maximum allowable fertilizer application amount of each farmland characteristic block in the current planting cycle; the soil moisture deviation value is defined according to the difference between the real-time monitored soil moisture value and the humidity threshold; when the real-time soil moisture is lower than the humidity threshold, the basic irrigation demand is dynamically adjusted proportionally based on the current soil moisture deviation value, and the irrigation amount benchmark value is generated under the constraint condition of controlling the maximum adjustment range, and finally a control parameter table containing the fertilizer application upper limit and the irrigation amount benchmark value is generated.

[0082] S4.2. Using farmland feature blocks as index units, construct a combination table containing fertilizer and irrigation amounts, generate a fertilization and irrigation instruction table, and identify the optimal fertilization and irrigation instructions based on priority rules;

[0083] It should be noted that the amount of fertilizer and irrigation was mapped to the farmland feature blocks at a resolution of 0.1 m, generating a data set containing the amount of fertilizer (kg / mu) and the amount of irrigation (m 3 / mu); according to the soil parameter safety threshold and the priority rules in the water and fertilizer control strategy, the farmland characteristic blocks in the initial instruction table are sorted in the order of excessive conductivity, insufficient humidity, and normal conductivity and humidity, and three-level priority labels of high, medium, and low are generated; the priority labels are bound to the fertilizer and irrigation amounts, and a fertilization and irrigation instruction table of fertilizer amount, irrigation amount and priority is generated through spatial data mapping conversion; according to the soil parameter safety threshold and gain coefficient, the fertilizer and irrigation amounts in each farmland characteristic block are calculated, and the optimal fertilization and irrigation instructions are identified according to the priority rules.

[0084] S4.3. Execute the optimal fertilization and irrigation instructions, record the actual fertilization amount, actual irrigation amount and soil parameter changes, and generate an execution effect report.

[0085] It should be noted that the amount of fertilizer and irrigation in the optimal fertilization and irrigation instructions is converted into a control signal through the Internet of Things controller and sent to the actuator, the actual amount of fertilizer and the actual amount of irrigation are recorded, and the soil conductivity sensor and the humidity sensor are used to collect the soil parameters after the execution of the optimal fertilization and irrigation instructions and calculate the change. The actual amount of fertilizer and the actual amount of irrigation are compared with the amount of fertilizer and irrigation in the fertilization and irrigation instruction table, the deviation rate is calculated, and an execution effect report is generated based on the deviation rate.

[0086] This embodiment also provides an artificial intelligence real-time monitoring system for rice water and fertilizer, including: a heat map generation module, a cause-effect diagram generation module, a strategy generation module, and an instruction execution module;

[0087] The heat map generation module is used to input the farmland feature dataset into the soil heat map generation model, reconstruct the farmland feature dataset at high resolution through the GAN adversarial network, and generate a high-precision soil heat map for the entire field;

[0088] The causal graph generation module is used to detect the collision relationship between fertilizer application amount and historical farming data based on the FCL causal algorithm based on the high-precision soil thermal map of the whole field, generate a preliminary causal graph, and use the multivariate linear regression method to construct the causal graph structure of fertilizer application amount and historical farming data;

[0089] The strategy generation module is used to analyze the multi-order causal effects of fertilizer application, soil parameters, and historical yields through a dynamic allocation algorithm based on a causal graph structure, and to formulate water and fertilizer control strategies in combination with a multi-objective optimization algorithm;

[0090] The instruction execution module is used to define the upper limit of fertilizer application and the baseline value of irrigation amount for each characteristic block of farmland according to the water and fertilizer control strategy, generate the optimal combination parameter table of fertilization and irrigation, output and execute regional fertilization and irrigation instructions according to irrigation rules, and generate an execution effect report.

[0091] This embodiment also provides a computer device suitable for the case of an artificial intelligence real-time monitoring method for rice water and fertilizer, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the artificial intelligence real-time monitoring method for rice water and fertilizer as proposed in the above embodiment.

[0092] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0093] This embodiment also provides a storage medium having a computer program stored thereon. When the program is executed by a processor, the method for real-time monitoring of rice water and fertilizer using artificial intelligence as proposed in the above embodiment is implemented. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0094] In summary, the present invention solves the ambiguity problem of field edges and salinized areas by: constructing a soil thermodynamic map generation model, accurately identifying areas with sudden changes in conductivity gradients, providing a high-fidelity soil spatial state basis for water and fertilizer regulation, and reducing ineffective irrigation; through the causal graph structure constructed based on the FCL causal algorithm, verifying the direct causal effect of fertilizer application amount, soil parameters and yield, and formulating water and fertilizer regulation strategies based on soil parameter thresholds to ensure a balance between yield maximization and salt control.

[0095] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An artificial intelligence-based real-time monitoring method for rice water and fertilizer, characterized by: include, The farmland feature dataset is input into the soil thermal map generation model, and the farmland feature dataset is reconstructed with high resolution through the GAN adversarial network to generate a high-precision soil thermal map of the entire field. Based on a high-precision soil thermodynamic map of the entire field, the collision relationship between fertilizer application and historical farming data was detected using the FCL causal algorithm to generate a preliminary causal diagram. The causal diagram structure of fertilizer application and historical farming data was then constructed using a multiple linear regression method. Based on the causal graph structure, a dynamic allocation algorithm is used to analyze the multi-order causal effects of fertilizer application, soil parameters, and historical yields, and a water and fertilizer control strategy is formulated in combination with a multi-objective optimization algorithm. According to the water and fertilizer control strategy, the upper limit of fertilizer application and the baseline value of irrigation amount for each characteristic block of farmland are defined, and the optimal combination parameter table of fertilization and irrigation is generated. According to the irrigation rules, the regional fertilization and irrigation instructions are output and executed, and an execution effect report is generated.

2. The artificial intelligence rice water and fertilizer real-time monitoring method according to claim 1, characterized in that: The farmland characteristic dataset includes soil parameters, image data and historical farming data; The historical agricultural data includes historical yield, fertilizer application amount and irrigation amount.

3. The artificial intelligence real-time monitoring method for rice water and fertilizer according to claim 2, characterized in that: The specific steps of constructing the soil thermal map generation model are as follows: Use the encoder and decoder to build the generator structure, use the convolution kernel cascade architecture to build the discriminator structure, couple the generator structure and the discriminator structure through the adversarial training framework to form a GAN adversarial network, and use the gradient fusion algorithm to build the splicing layer; A soil heat map generation model is constructed based on the GAN adversarial network and splicing layer.

4. The artificial intelligence real-time monitoring method for rice water and fertilizer according to claim 3, characterized in that: The GAN adversarial network is used to reconstruct the farmland feature dataset at high resolution and generate a high-precision soil thermal map of the entire field. The specific steps are as follows: The GAN adversarial network is used to divide the farmland feature dataset into pixel blocks, generate farmland feature blocks, and then use the generator structure to perform high-resolution reconstruction to generate soil thermal map blocks; The stitching layer stitches the soil thermal map blocks, uses the central difference method to identify high mutation areas, and uses the soil moisture diffusion equation to optimize the high mutation areas to generate a high-precision soil thermal map of the entire field.

5. The artificial intelligence real-time monitoring method for rice water and fertilizer according to claim 4, characterized in that: The collision relationship between the amount of fertilizer applied and the historical farming data is detected according to the FCL causal algorithm, a preliminary causal graph is generated, and a causal graph structure of the amount of fertilizer applied and the historical farming data is constructed using the multivariate linear regression method. The specific steps are as follows: A spatiotemporal gridded data fusion method was used to correlate high-precision soil thermal maps of the entire field with fertilizer application rates, soil parameters, and historical yields to form a causal analysis dataset. The Pearson correlation coefficient is used to screen the correlation items between fertilizer application amount, soil parameters and historical yield in the causal analysis data set, and the significant variable set is output; Using the FCL causal algorithm, we detect collision relationships among significant variable sets and generate a preliminary causal graph of fertilizer application, soil parameters, and historical yields. The direct causal effects of the preliminary causal diagram were verified by the multiple linear regression method to generate the causal diagram structure.

6. The artificial intelligence real-time monitoring method for rice water and fertilizer according to claim 5, characterized in that: Based on the causal graph structure, the dynamic allocation algorithm is used to analyze the multi-order causal effects of fertilizer application amount, soil parameters and historical yield, and a water and fertilizer control strategy is formulated in combination with a multi-objective optimization algorithm. The specific steps are as follows: A dynamic allocation algorithm was used to decompose the direct effect of fertilizer application on historical yield and the indirect effect of soil parameters, and the total effect was calculated using a multi-path accumulation method. Define the gain coefficient of fertilizer application on historical yield based on the total effect, and define the fertilizer application baseline value and soil parameter safety threshold based on the gain coefficient and soil parameters; High conductivity areas and low humidity areas are identified based on the safety thresholds of soil parameters, the fertilizer application amount is adjusted for high conductivity areas and low humidity areas based on the fertilizer application benchmark value, and water and fertilizer control strategies are formulated through irrigation rules.

7. The artificial intelligence real-time monitoring method for rice water and fertilizer according to claim 6, characterized in that: According to the water and fertilizer control strategy, the upper limit of fertilizer application and the reference value of irrigation amount for each farmland block are defined, the optimal combination parameter table of fertilization and irrigation is generated, and the regional fertilization and irrigation instructions are output and executed according to the irrigation rules, and the execution effect report is generated. The specific steps are as follows: According to the water and fertilizer control strategy, combined with the soil safety threshold, the upper limit of fertilizer application and the baseline value of irrigation amount for each farmland characteristic block are determined; Using farmland characteristic blocks as index units, a combined table containing fertilizer and irrigation amounts is constructed to generate a fertilization and irrigation instruction table, and the optimal fertilization and irrigation instructions are identified based on the water and fertilizer control strategy. Execute optimal fertilization and irrigation instructions, record actual fertilization amount, actual irrigation amount and changes in soil parameters, and generate execution effect reports.

8. An artificial intelligence real-time monitoring system for rice water and fertilizer, based on the artificial intelligence real-time monitoring method for rice water and fertilizer according to any one of claims 1 to 7, characterized in that: Including, heat map generation module, cause and effect diagram generation module, strategy generation module, instruction execution module; The heat map generation module is used to input the farmland feature dataset into the soil heat map generation model, reconstruct the farmland feature dataset at high resolution through the GAN adversarial network, and generate a high-precision soil heat map for the entire field; The causal graph generation module is used to detect the collision relationship between fertilizer application amount and historical farming data based on the FCL causal algorithm based on the high-precision soil thermal map of the whole field, generate a preliminary causal graph, and use the multivariate linear regression method to construct the causal graph structure of fertilizer application amount and historical farming data; The strategy generation module is used to analyze the multi-order causal effects of fertilizer application, soil parameters, and historical yields through a dynamic allocation algorithm based on a causal graph structure, and to formulate water and fertilizer control strategies in combination with a multi-objective optimization algorithm; The instruction execution module is used to define the upper limit of fertilizer application and the baseline value of irrigation amount for each characteristic block of farmland according to the water and fertilizer control strategy, generate the optimal combination parameter table of fertilization and irrigation, output and execute regional fertilization and irrigation instructions according to irrigation rules, and generate an execution effect report.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the artificial intelligence real-time monitoring method for rice water and fertilizer according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the artificial intelligence real-time monitoring method for rice water and fertilizer according to any one of claims 1 to 7 are implemented.

Images