A method for analyzing the application effect of liquid fertilizer based on self-supervised learning

Abstract

The present invention discloses a method for analyzing the application effect of liquid fertilizer based on self-supervised learning, which includes the following steps: S1, generating initial feature data; S2, generating environmental feature representations based on a heterogeneous environment adaptive module; S3, extracting short-term and long-term spatio-temporal features of the liquid fertilizer application effect based on a dual spatio-temporal feature decoupling and dynamic fusion model, and generating comprehensive spatio-temporal feature representations; S4, using self-supervised learning to generate fertilization features under different scenarios, and eliminating environmental interference factors through a causal inference module; S5, generating fertilization intention features and dynamically adjusting fertilization strategies; S6, when the difference between the prediction result and the actual effect exceeds a threshold, triggering a re-learning module to update model parameters; S7, performing integrated optimization, realizing feature sharing and task coordination through multi-task collaborative learning, and using Bayesian optimization for automatic tuning. The present invention realizes intelligent analysis of the liquid fertilizer application effect through self-supervised learning and multi-task collaborative optimization methods.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent agricultural fertilization, and particularly to a method for analyzing the application effect of liquid fertilizer based on self-supervised learning. Background Art

[0002] In modern agricultural production, the application of liquid fertilizer, as an efficient fertilization method, is widely used because it can be absorbed by crops more quickly and improve fertilizer utilization efficiency. The accuracy and efficiency of liquid fertilizer application directly affect the growth quality and yield of crops. However, the application effect of liquid fertilizer is complexly affected by environmental conditions (such as climate, soil humidity, soil pH value, etc.) and the growth state of crops. There may be significant differences in its application effect under different fertilization conditions. Therefore, in order to ensure the stability of fertilization effect, the existing technologies generally rely on manual experience or single-sensor monitoring methods to make fertilization decisions, and it is difficult to fully consider the diversity of the environment and the needs of crops at different growth stages.

[0003] Traditional liquid fertilizer application methods mainly determine the fertilization amount, frequency, and concentration through simple summary and analysis of environmental sensor data. The limitation of this method is that it is difficult to adapt to diverse agricultural environments and the dynamic growth needs of crops. In addition, in the existing technologies, most fertilization systems only use fixed fertilization parameters and lack an intelligent adjustment mechanism. Even after monitoring environmental changes or crop feedback data, the adjustment of fertilization parameters still depends on fixed preset rules or manual intervention. This lack of intelligent adaptability in fertilization methods results in low fertilizer utilization efficiency and unstable fertilization effects. Especially when facing complex agricultural environments (such as areas with large climate fluctuations and diverse soil types), it is very difficult for the existing fertilization methods to be optimized according to real-time feedback.

[0004] To solve these problems, in recent years, some machine learning-based agricultural fertilization methods have emerged, which provide data support for fertilization through modeling and prediction of historical data. Such technologies usually rely on supervised learning algorithms and train models through a large amount of labeled historical data. However, due to the complexity and variability of the agricultural environment, it is very difficult to obtain a large amount of high-quality labeled data. The dependence on supervised learning makes these methods perform poorly in new environments or unlabeled data, and at the same time, they lack effective utilization of real-time feedback data during the fertilization process, making it difficult to achieve dynamic analysis and adaptive optimization of the liquid fertilizer application effect.

[0005] When dealing with various environmental factors, existing intelligent fertilization technologies often lack comprehensive consideration of characteristics at different time scales and are unable to accurately capture the different effects of liquid fertilizer application in the short and long terms. The impact of liquid fertilizer on crops has both immediate short-term effects and relatively slow long-term effects. However, existing technologies usually adopt a single time-series model to analyze fertilization effects, ignoring the differences between short-term and long-term effects. In addition, traditional models lack the ability to adaptively adjust to environmental data characteristics during the feature fusion process, resulting in poor generalization ability under diverse environmental conditions, which imposes certain limitations on the application scope and accuracy of the models.

[0006] In addition, when analyzing liquid fertilizer application, existing technologies lack the support of causal reasoning. The agricultural environment has complex causal relationships, and only through correlation analysis, it is impossible to accurately eliminate the interference of environmental variables on fertilization effects, which greatly affects the prediction accuracy of fertilization effects. Traditional models often only conduct correlation analysis and ignore the in-depth modeling of causal relationships, resulting in difficulties in optimizing fertilization decisions according to the real needs of crop growth. The lack of causal analysis will lead to the misjudgment of the true relationship between environmental factors and fertilization effects by the model, thus affecting the accuracy and adaptability of fertilization strategies.

[0007] Regarding the adjustment of fertilization strategies, existing technologies mainly rely on fixed rules and thresholds and lack dynamic adjustment and intelligent optimization means. With the continuous update of environmental changes and crop growth states, fixed fertilization strategies are difficult to meet the dynamic changing requirements. Although some technologies attempt to optimize fertilization strategies through feedback mechanisms, they are often based on simple comparative analysis without more complex feedback control mechanisms. Once the actual fertilization effect fails to meet expectations, it is difficult for existing systems to automatically update fertilization strategies, usually requiring manual intervention or full model retraining, resulting in a waste of time cost and computing resources. This inefficient adjustment mechanism is particularly insufficient in the actual agricultural environment, especially in situations where the crop growth cycle is long or environmental changes are frequent. The lag of existing fertilization models is obvious, and it is difficult to achieve real-time optimization and update.

[0008] In the optimization of fertilization strategy parameters, existing intelligent fertilization systems lack automated and precise means for tuning key parameters and often use manual settings or rough parameter searches, which are inefficient and have poor effects. For parameter adjustment in multi-task fertilization scenarios, traditional methods mostly lack multi-task collaboration mechanisms and fail to effectively achieve feature sharing and dynamic collaborative optimization between tasks. In addition, commonly used parameter optimization algorithms, such as grid search and random search, are difficult to find the optimal parameter combination in complex agricultural environments, and existing machine learning-based parameter tuning algorithms are limited by single-task learning models and do not comprehensively consider the relationships between different tasks and potential synergistic effects, resulting in limitations in the optimization of the overall fertilization effect.

[0009] Therefore, how to provide a method for analyzing the application effect of liquid fertilizer based on self-supervised learning is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0010] An object of the present invention is to propose a method for analyzing the application effect of liquid fertilizer based on self-supervised learning. The present invention adopts self-supervised learning, multi-task collaborative optimization and causal reasoning methods to intelligently analyze and dynamically optimize the application effect of liquid fertilizer. Through the generation of multi-scenario fertilization features, the dual spatio-temporal decoupling of short-term and long-term fertilization effects, and a real-time feedback mechanism, it is ensured that the fertilization strategy can be adaptively adjusted according to the crop growth status and environmental changes. Bayesian optimization further improves the automatic tuning efficiency of key parameters, making the present invention have strong environmental adaptability and real-time response ability, and significantly improving the application accuracy of liquid fertilizer and resource utilization efficiency.

[0011] A method for analyzing the application effect of liquid fertilizer based on self-supervised learning according to an embodiment of the present invention includes the following steps:

[0012] S1. Collect environmental data and crop images through multiple sensors, and perform cleaning, standardization processing and feature extraction to generate initial feature data;

[0013] S2. Input the initial feature data into the heterogeneous environment adaptive module, and dynamically adjust the convolution kernel size and feature fusion strategy according to different environmental features to generate environmental feature representations;

[0014] S3. Based on the dual spatio-temporal feature decoupling and dynamic fusion model, input the environmental feature representations into the multi-scale time series feature extraction module to extract the short-term spatio-temporal features and long-term spatio-temporal features of the liquid fertilizer application effect. Use the spatio-temporal decoupling mechanism to separate the feature channels in different time periods, and fuse the short-term spatio-temporal features and long-term spatio-temporal features through a dynamic weighting mechanism. Automatically adjust the fusion ratio according to the crop growth cycle and fertilization effect changes to generate comprehensive spatio-temporal feature representations;

[0015] S4. Based on the generated comprehensive spatio-temporal feature representations, use self-supervised learning to generate fertilization features in different scenarios, and eliminate environmental interference factors through the causal reasoning module to extract the true influence features of liquid fertilizer at different growth stages;

[0016] S5. Generate fertilization intention features based on the true influence features and crop growth status, dynamically adjust the fertilization strategy, and regenerate the fertilization intention when the feedback effect is not ideal;

[0017] S6. Collect the crop growth change data after fertilization through the real-time feedback loop, calculate the difference between the prediction result and the actual effect, and when the difference exceeds the threshold, trigger the re-learning module to update the fertilization parameters;

[0018] S7. Perform integration optimization to achieve feature sharing and task coordination through multi-task collaborative learning, and use Bayesian optimization to automatically tune key parameters.

[0019] Furthermore, the specific steps of S2 are as follows:

[0020] S21. Input the initial feature data generated in S1 into the heterogeneous environment adaptation module, where the initial feature data includes soil humidity, nutrient content, pH value, and climate data.

[0021] S22. In the heterogeneous environment adaptation module, dynamically select and initialize the convolutional kernel parameters according to the soil humidity, nutrient content, pH value, and climate data, including the initial size and shape of the convolutional kernel.

[0022] S23. Based on the multiple statistical characteristics of the initial feature data, adaptively adjust the size and shape of the convolutional kernel:

[0023]

[0024] Among them, K adjust represents the adjusted convolutional kernel size, K init represents the initial convolutional kernel size, E var represents the variance of the initial feature data, E avg represents the mean of the initial feature data, E std represents the standard deviation of the initial feature data, E range represents the value range of the initial feature data, E max represents the maximum value of the initial feature data, E min represents the minimum value of the initial feature data, α a α d β b β f γ c γ g and δ e represent control coefficients;

[0025] S24. Use the adjusted convolutional kernel parameters to perform convolution operations to extract multi-scale features from the initial feature data and generate environmental feature matrices at different scales.

[0026] S25. In the heterogeneous environment adaptation module, apply a multi-level feature fusion strategy to fuse the environmental feature matrices at different scales and different modalities in space, channel, and scale.

[0027] S26. Finally, output the generated environmental feature representation.

[0028] Even further, the specific steps of S3 are as follows:

[0029] S31. Input the generated environmental feature representation into the dual spatio-temporal feature decoupling and dynamic fusion model, and perform temporal decomposition on the environmental feature representation through a multi-scale temporal feature extraction module in the dual spatio-temporal feature decoupling and dynamic fusion model;

[0030] S32. Through the short-term spatio-temporal feature extraction channel, send the input environmental feature data into a long short-term memory network structure to extract the short-term effect features after fertilization, record the rapid response features of the crops within several days or weeks, and generate the short-term spatio-temporal feature F short ;

[0031] S33. Utilize the long-term spatio-temporal feature extraction channel to input the environmental feature data into a temporal convolutional network to extract the long-term impact features of the fertilization effect, and generate the long-term spatio-temporal feature F containing the impact of fertilization within several months or the entire growing season long ;

[0032] S34. Apply the spatio-temporal decoupling mechanism to separate the short-term spatio-temporal feature F short and the long-term spatio-temporal feature F long into independent feature channels at different time periods to form a short-term channel and a long-term channel;

[0033] S35. Based on the dynamic weighting mechanism, set adaptive weight parameters and apply them to the spatio-temporal features of the short-term channel and the long-term channel respectively to form the weighted short-term feature and long-term feature;

[0034] S36. Fuse the weighted short-term feature and long-term feature to generate a comprehensive spatio-temporal feature representation:

[0035]

[0036] where, F fusion represents the comprehensive spatio-temporal feature representation, exp represents the exponential function, α represents the weighting parameter of the short-term feature, β represents the weighting parameter of the long-term feature, γ represents the power control coefficient, δ represents the exponential control coefficient, η represents the transformation parameter of the long-term feature, sinh represents the hyperbolic sine function, θ represents the coefficient of the hyperbolic sine term, and ζ represents the normalization parameter.

[0037] Furthermore, the specific steps of S4 include:

[0038] S41. Input the generated comprehensive spatio-temporal feature representation into the self-supervised learning module to generate fertilization feature representations under multiple different scenarios. The self-supervised learning module generates multi-scenario fertilization features by constructing a contrast task of unlabeled data and constructs a multi-scenario fertilization feature library;

[0039] S42. During the self-supervised learning process, create the target scenario feature F target and the contrast scenario feature F through the contrast learning methodcontrast and use the contrast loss function L contrast to represent the difference degree between situations:

[0040]

[0041] where exp represents the exponential function, sim represents the similarity function, and F j contrast represents the j-th contrast situation feature, and F positive represents the positive sample feature representation, ∥·∥ represents the vector norm operation, κ1 and κ2 represent the weight coefficients of the norm terms, and τ represents the parameter that regulates the similarity distribution in contrast learning;

[0042] S43. Input the generated multi-situation fertilization feature library into the causal inference module, establish the causal relationship between features through the structural equation model, and eliminate environmental interference factors. The causal inference module identifies the direct causal relationship between fertilization features and crop growth;

[0043] S44. Use the latent confounder variable model to correct the fertilization features and crop growth features, identify the influence of latent environmental variables and eliminate noise interference:

[0044]

[0045] where Y represents the crop growth influence, X represents the fertilization feature, Z represents the latent environmental variable, α x , β z , γ p , δ s and θ x represent the weight coefficients, and η y represents the normalization coefficient;

[0046] S45. Apply the counterfactual reasoning method to create the counterfactual situation feature F counter , simulate the crop growth effect under different fertilization conditions, adjust the fertilization features through intervention operations, and generate the crop response under the hypothetical situation;

[0047] S46. By comparing the actual situation feature F actual and the counterfactual situation feature F counter , extract the true influence features of liquid fertilizer at different growth stages, eliminate the interference of latent environmental variables, and finally obtain the true fertilization effect during the crop growth process.

[0048] Furthermore, the specific steps of S5 include:

[0049] S51. Input the generated true influence feature F true and the current growth state S of the crop into the fertilization intention generation module to generate the initial fertilization intention feature Iinitial , the initial fertilization intention feature I initial is used to represent the best strategy for liquid fertilization in the current environment;

[0050] S52. During the generation of the fertilization intention, based on the dynamic mapping of the real influence feature F true and the current growth state S of the crop, a fertilization intention function is constructed. The fertilization intention function synthesizes the weights of different features to generate the initial fertilization intention feature I that adapts to different environments and crop requirements initial ;

[0051] S53. According to the generated initial fertilization intention feature I initial , a set of fertilization strategy parameters P is generated, including the fertilization amount P amount , the fertilization frequency P freq and the fertilization concentration P conc to achieve the optimal fertilization effect;

[0052] S54. During the fertilization process, the fertilization effect is monitored according to the real-time feedback data R of the crop growth actual . The feedback result is compared with the expected effect R expected . If the deviation exceeds the set threshold, the fertilization intention update mechanism is triggered:

[0053]

[0054] where D represents the deviation between the actual effect and the expected effect, α growth represents the growth-related adjustment coefficient, β env represents the weight coefficient of the environmental feature, γ fert represents the adjustment coefficient of the fertilization feature, δ crop represents the adjustment coefficient of the crop feature, and ∈ represents the smoothing term;

[0055] S55. The fertilization intention update mechanism adjusts the initial fertilization intention feature I according to the deviation D between the actual effect and the expected effect initial :

[0056]

[0057] where I update represents the updated fertilization intention feature, exp represents the exponential function, λ growth represents the growth adjustment coefficient, λ env represents the environmental adjustment coefficient, θ fert represents the fertilization feature adjustment coefficient, ζ crop represents the crop feature adjustment coefficient, and η growth represents the gain control coefficient;

[0058] S56. According to the updated fertilization intention feature I update , regenerate the set of fertilization strategy parameters P and apply it to the liquid fertilizer application process.

[0059] Furthermore, the S7 specifically includes:

[0060] S71. Based on the multi-task collaborative learning mechanism, construct a unified set of feature vectors to achieve feature sharing and form a stable multi-task fertilization model;

[0061] S72. On the basis of feature sharing, generate independent task representations for each fertilization task. While each fertilization task shares features, it retains its own characteristics; based on the designed task-specific coding module, automatically identify and utilize the feature subsets suitable for specific tasks when dealing with different tasks;

[0062] S73. In multi-task collaborative learning, introduce an interaction module between tasks to capture the potential correlation relationships between different tasks, generate joint feature representations, and through the weighting mechanism of the interaction module, enable the learning process of one task to feedback to the feature update of other tasks;

[0063] S74. On the basis of the multi-task learning framework, use the Bayesian optimization method to automatically tune the key parameters, construct a parameter space, and gradually update the parameter distribution according to the feedback data of the fertilization task;

[0064] S75. Apply a surrogate model in Bayesian optimization to quickly evaluate different parameter combinations, simulate the feedback situation in the fertilization process through the surrogate model, and the surrogate model is updated based on historical data to learn the optimal parameter configuration from less experimental data;

[0065] S76. After the parameter tuning is completed, apply the obtained set of optimal parameter configurations to the multi-task collaborative model to achieve the overall optimization of the fertilization strategy. Through the integrated and optimized multi-task collaborative learning model, each fertilization task can obtain the best fertilization effect in different environments.

[0066] The beneficial effects of the present invention are:

[0067] First of all, by introducing a self-supervised learning module, the present invention can generate multi-situation fertilization features based on unlabeled data, significantly reduce the dependence on labeled data, reduce the training cost of the model, and improve the adaptability of the model in new environments. At the same time, the introduction of the dual spatio-temporal feature decoupling and dynamic fusion mechanism enables the model to accurately capture the short-term response and long-term effects of liquid fertilizer application effects, meet the nutrient requirements of crops at different growth stages, avoid the limitations of single-time series feature models in complex agricultural environments, and thus improve the accuracy of fertilization strategies.

[0068] Secondly, through the use of the causal reasoning module, the present invention can effectively eliminate the interference of environmental variables, thereby enabling a more authentic and reliable analysis of the fertilization effect. Compared with the limitations of only correlation analysis in traditional methods, this method can identify the true impact of liquid fertilizers at different growth stages, avoid the misleading of environmental variables on the prediction of fertilization effects, and fundamentally improve the scientificity and reliability of fertilization decisions. In addition, the design of dynamically adjusting fertilization intention features and fertilization strategies enables the present invention to respond in real time to the growth status of crops and changes in the external environment, ensuring the flexibility of fertilization strategies. When the fertilization effect fails to meet expectations, this method can automatically adjust the fertilization intention and optimize fertilization parameters based on feedback data, thereby avoiding manual intervention, improving the autonomy and intelligence of the fertilization process, and further enhancing the utilization efficiency of liquid fertilizers.

[0069] In addition, the combination of multi-task collaborative learning and Bayesian optimization adopted by the present invention ensures the integrated optimization of fertilization strategies and the automatic tuning of parameters. The multi-task collaborative learning mechanism enables feature sharing and task coordination among different fertilization tasks, enhancing the stability of the model in complex and variable agricultural environments. At the same time, Bayesian optimization can efficiently search for the optimal combination in the multi-dimensional parameter space, providing precise parameter configurations for each fertilization scenario to ensure the best effect of fertilization strategies under different environmental and crop conditions. Through integrated optimization, the fertilization model of the present invention has excellent generalization ability and adaptability, and can not only effectively cope with changes in various environmental variables, but also continuously provide optimal fertilization decisions under different crops and growth stages, thus realizing the intelligent, refined and efficient application of liquid fertilizers in agricultural production. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] The drawings are used to provide a further understanding of the present invention, and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention, and do not constitute a limitation to the present invention. In the drawings:

[0071] Figure 1 is a flowchart of a method for analyzing the application effect of liquid fertilizers based on self-supervised learning proposed by the present invention;

[0072] Figure 2 is a schematic structural diagram of a dual spatio-temporal feature decoupling and dynamic fusion model of a method for analyzing the application effect of liquid fertilizers based on self-supervised learning proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0073] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0074] Refer to Figure 1 andFigure 2 , a method for analyzing the application effect of liquid fertilizer based on self-supervised learning, comprising the following steps:

[0075] S1. Collect environmental data and crop images through multiple sensors, and perform cleaning, standardization processing, and feature extraction to generate initial feature data;

[0076] S2. Input the initial feature data into the heterogeneous environment adaptive module, and dynamically adjust the convolution kernel size and feature fusion strategy according to different environmental features to generate environmental feature representations;

[0077] S3. Based on the dual spatio-temporal feature decoupling and dynamic fusion model, input the environmental feature representations into the multi-scale temporal feature extraction module to extract the short-term spatio-temporal features and long-term spatio-temporal features of the liquid fertilizer application effect. Use the spatio-temporal decoupling mechanism to separate the feature channels at different time periods, and fuse the short-term spatio-temporal features and long-term spatio-temporal features through the dynamic weighting mechanism. Automatically adjust the fusion ratio according to the crop growth cycle and the change of fertilization effect to generate comprehensive spatio-temporal feature representations;

[0078] S4. Based on the generated comprehensive spatio-temporal feature representations, use self-supervised learning to generate fertilization features in different scenarios, and eliminate environmental interference factors through the causal reasoning module to extract the true influence features of the liquid fertilizer at different growth stages;

[0079] S5. Generate fertilization intention features based on the true influence features and crop growth status, dynamically adjust the fertilization strategy, and regenerate the fertilization intention when the feedback effect is not ideal;

[0080] S6. Collect data on the growth changes of crops after fertilization through a real-time feedback loop, calculate the difference between the prediction result and the actual effect, and trigger the re-learning module to update the fertilization parameters when the difference exceeds the threshold;

[0081] S7. Perform integrated optimization, achieve feature sharing and task coordination through multi-task collaborative learning, and automatically optimize the key parameters using Bayesian optimization.

[0082] In this embodiment, the S2 specifically includes:

[0083] S21. Input the initial feature data generated in S1 into the heterogeneous environment adaptive module, and the initial feature data includes soil humidity, nutrient components, pH value, and climate data;

[0084] S22. In the heterogeneous environment adaptive module, dynamically select and initialize the convolution kernel parameters according to the soil humidity, nutrient components, pH value, and climate data, including the initial size and initial shape of the convolution kernel;

[0085] S23. Adaptively adjust the size and shape of the convolutional kernel based on the multiple statistical characteristics of the initial feature data:

[0086]

[0087] Among them, K adjust represents the size of the adjusted convolutional kernel, K init represents the size of the initial convolutional kernel, E var represents the variance of the initial feature data, E avg represents the mean of the initial feature data, E std represents the standard deviation of the initial feature data, E range represents the value range of the initial feature data, E max represents the maximum value of the initial feature data, E min represents the minimum value of the initial feature data, α a α d β b β f γ c γ g and δ e represent control coefficients;

[0088] S24. Perform a convolution operation using the adjusted convolutional kernel parameters to extract multi-scale features from the initial feature data and generate environmental feature matrices at different scales;

[0089] S25. In the heterogeneous environment adaptive module, apply a multi-level feature fusion strategy to fuse the environmental feature matrices of different scales and different modalities in space, channel, and scale;

[0090] S26. Finally, output the generated environmental feature representation.

[0091] In this embodiment, the specific steps of S3 are as follows:

[0092] S31. Input the generated environmental feature representation into the dual spatio-temporal feature decoupling and dynamic fusion model, and perform temporal decomposition on the environmental feature representation through the multi-scale temporal feature extraction module in the dual spatio-temporal feature decoupling and dynamic fusion model;

[0093] S32. Through the short-term spatio-temporal feature extraction channel, input the environmental feature data into the long short-term memory network structure to extract the short-term effect features after fertilization, record the rapid response features of the crops in several days or weeks, and generate the short-term spatio-temporal feature F short ;

[0094] S33. Use the long-term spatio-temporal feature extraction channel to input the environmental feature data into the temporal convolutional network, extract features of the long-term impact of fertilization effect, and generate the long-term spatio-temporal feature F that includes the impact of fertilization over several months or the entire growing season. long ;

[0095] S34. Apply the spatio-temporal decoupling mechanism to separate the short-term spatio-temporal feature F short and the long-term spatio-temporal feature F long into independent feature channels at different time periods to form a short-term channel and a long-term channel.

[0096] S35. Based on the dynamic weighting mechanism, set the adaptive weight parameters, which act on the spatio-temporal features of the short-term channel and the long-term channel respectively, to form the weighted short-term feature and long-term feature.

[0097] S36. Fuse the weighted short-term feature and long-term feature to generate the comprehensive spatio-temporal feature representation:

[0098]

[0099] where F fusion represents the comprehensive spatio-temporal feature representation, exp represents the exponential function, α represents the weighting parameter of the short-term feature, β represents the weighting parameter of the long-term feature, γ represents the power control coefficient, δ represents the exponential control coefficient, η represents the transformation parameter of the long-term feature, sinh represents the hyperbolic sine function, θ represents the coefficient of the hyperbolic sine term, and ζ represents the normalization parameter.

[0100] In this embodiment, the S4 specifically includes:

[0101] S41. Input the generated comprehensive spatio-temporal feature representation into the self-supervised learning module to generate the fertilization feature representations in multiple different scenarios. The self-supervised learning module generates the multi-scenario fertilization features by constructing a contrast task of unlabeled data and constructs a multi-scenario fertilization feature library.

[0102] S42. In the self-supervised learning process, create the target scenario feature F target and the contrast scenario feature F contrast through the contrast learning method, and use the contrast loss function L contrast to represent the difference degree between scenarios:

[0103]

[0104] where exp represents the exponential function, sim represents the similarity function, represents the j-th contrast scenario feature, and F positiveIt represents the positive sample feature representation, ∥·∥ represents the norm operation of the vector, κ1 and κ2 represent the weight coefficients of the norm terms, and τ represents the parameter for regulating the similarity distribution in contrastive learning;

[0105] S43. Input the generated multi - context fertilization feature library into the causal inference module, establish the causal relationship between features through the structural equation model, and eliminate environmental interference factors. The causal inference module identifies the direct causal relationship between fertilization features and crop growth;

[0106] S44. Use the latent confounder variable model to correct the fertilization features and crop growth features, identify the influence of latent environmental variables and eliminate noise interference:

[0107]

[0108] Among them, Y represents the crop growth impact, X represents the fertilization feature, Z represents the latent environmental variable, α x 、β z 、γ p 、δ s and θ x represent the weight coefficients, and η y represents the normalization coefficient;

[0109] S45. Apply the counterfactual reasoning method to create the counterfactual context feature F counter , simulate the crop growth effect under different fertilization conditions, adjust the fertilization features through intervention operations, and generate the crop response under the hypothetical situation;

[0110] S46. By comparing the actual context feature F actual and the counterfactual context feature F counter , extract the true influence features of liquid fertilizer at different growth stages, eliminate the interference of latent environmental variables, and finally obtain the true fertilization effect during the crop growth process.

[0111] In this embodiment, the specific steps of S5 are as follows:

[0112] S51. Input the generated true influence feature F true and the current growth state S of the crop into the fertilization intention generation module to generate the initial fertilization intention feature I initial , and the initial fertilization intention feature I initial is used to represent the best strategy for liquid fertilization in the current environment;

[0113] S52. During the generation of the fertilization intention, based on the dynamic mapping of the true influence feature F true and the current growth state S of the crop, construct a fertilization intention function. The fertilization intention function synthesizes the weights of different features and generates the initial fertilization intention feature I that adapts to different environments and crop requirementsinitial ;

[0114] S53. Generate a set of fertilization strategy parameters P according to the generated initial fertilization intention feature I initial , including the fertilization amount P amount , the fertilization frequency P freq and the fertilization concentration P conc , to achieve the optimal fertilization effect;

[0115] S54. During the fertilization process, monitor the fertilization effect according to the real-time feedback data R actual of the crop growth, compare the feedback result with the expected effect R expected . If the deviation exceeds the set threshold, trigger the fertilization intention update mechanism:

[0116]

[0117] where D represents the deviation between the actual effect and the expected effect, and α growth represents the regulation coefficient related to growth, β env represents the weight coefficient of the environmental characteristics, γ fert represents the regulation coefficient of the fertilization characteristics, δ crop represents the regulation coefficient of the crop characteristics, and ∈ represents the smoothing term;

[0118] S55. The fertilization intention update mechanism adjusts the initial fertilization intention feature I initial according to the deviation D between the actual effect and the expected effect:

[0119]

[0120] where I update represents the updated fertilization intention feature, exp represents the exponential function, and λ growth represents the growth regulation coefficient, λ env represents the environmental regulation coefficient, θ fert represents the fertilization characteristic regulation coefficient, ζ crop represents the crop characteristic regulation coefficient, and η growth represents the gain control coefficient;

[0121] S56. According to the updated fertilization intention feature I update , regenerate the set of fertilization strategy parameters P and apply it to the liquid fertilizer application process.

[0122] In this embodiment, the S7 specifically includes:

[0123] S71. Based on the multi-task collaborative learning mechanism, construct a unified set of feature vectors to achieve feature sharing and form a stable multi-task fertilization model;

[0124] S72. On the basis of feature sharing, generate independent task representations for each fertilization task. While sharing features, each fertilization task retains its own characteristics; based on a designed task-specific encoding module, automatically identify and utilize a feature subset suitable for a specific task when dealing with different tasks;

[0125] S73. In multi-task collaborative learning, introduce an interaction module between tasks to capture potential correlation relationships between different tasks, generate joint feature representations, and through the weighted mechanism of the interaction module, enable the learning process of one task to feedback to the feature update of other tasks;

[0126] S74. On the basis of the multi-task learning framework, use the Bayesian optimization method to automatically tune key parameters, construct a parameter space, and gradually update the parameter distribution according to the feedback data of the fertilization task;

[0127] S75. Apply a surrogate model in Bayesian optimization to quickly evaluate different parameter combinations, simulate the feedback situation in the fertilization process through the surrogate model, and the surrogate model is updated based on historical data to learn the optimal parameter configuration from less experimental data;

[0128] S76. After the parameter tuning is completed, apply the obtained set of optimal parameter configurations to the multi-task collaborative model to achieve the overall optimization of the fertilization strategy. Through the integrated and optimized multi-task collaborative learning model, each fertilization task achieves the best fertilization effect in different environments.

[0129] Example 1:

[0130] To verify the feasibility of the present invention in implementation, apply the present invention to a certain experimental field project. The area of the experimental field is 20 hectares, and the planted crop is wheat. The agricultural production in this area faces diverse climate changes and complex soil types. Traditional fertilization techniques are difficult to effectively meet the needs of different regions and different growth stages, resulting in uneven crop growth and unstable fertilization effects. Traditional liquid fertilizer application relies on fixed fertilization parameters and cannot be dynamically adjusted when environmental factors change, leading to low fertilizer utilization rate and serious soil nutrient loss. Through the method of the present invention, attempt to solve the problem of intelligent application of liquid fertilizer in multiple environments in order to achieve efficient and precise fertilization effects.

[0131] In this embodiment, the system of the present invention is installed with a variety of environmental sensors, including soil moisture sensors, temperature sensors, pH value sensors, and nitrogen, phosphorus, and potassium content sensors. These sensors collect data every 30 minutes and transmit the data to the central processing system through a wireless network. After the data is cleaned, standardized, and feature extracted, it is input into the heterogeneous environment adaptive module. This module adaptively adjusts the size of the convolutional kernel and the feature fusion strategy according to the soil and climate characteristics of different regions to generate a highly adaptable environmental feature representation. In the experiment, the initial data distribution of soil moisture was between 10% - 30%, the pH value was between 6.0 - 7.5, and the temperature fluctuated between 15 - 30°C due to seasonal changes. Through the processing of the heterogeneous environment adaptive module, these environmental features are transformed into standard features that the system can handle.

[0132] Next, the system uses the dual spatio-temporal feature decoupling and dynamic fusion model to perform multi-scale feature extraction on the environmental feature representation. The short-term spatio-temporal feature extraction module captures the immediate effects after fertilization, such as short-term growth performances like the color change of wheat leaves and the growth height of stems within a few days; the long-term spatio-temporal features are used to analyze the continuous impact of liquid fertilizer on the entire growing season, such as long-term performances like increasing plant height and leaf area index. In the first month of the initial stage of the experiment, the changes in short-term features showed the rapid response of wheat growth to liquid fertilizer, while as the growing season progressed, the long-term features captured the effects of liquid fertilizer in enhancing crop disease resistance and increasing stem strength. These short-term and long-term features are dynamically weighted and fused in the dual spatio-temporal feature decoupling and dynamic fusion model to generate comprehensive spatio-temporal features, ensuring that the fertilization strategy can adapt to the actual needs of different growth stages of the crop.

[0133] To further improve the fertilization accuracy, the system generates a fertilization feature library under multiple scenarios through the self-supervised learning module. The causal inference module analyzes the fertilization data, eliminates environmental interference factors, and identifies the true impact of liquid fertilizer at different growth stages. In the experimental field, the system eliminates the random impacts of climate and soil structure through causal inference and obtains the actual impact values of fertilization amount on leaf thickness, plant height, and grain weight. For example, in the case of relatively low soil moisture (10% - 15%), the positive impact of fertilization on the leaf thickness and grain weight of wheat is more significant, while when the moisture is relatively high (25% - 30%), this impact is weakened. These causal features provide a scientific basis for the system's fertilization strategy.

[0134] During the experiment, the system adjusts the fertilization strategy in real time according to the growth feedback data of the crops. If there is a large difference between the actual fertilization effect and the expected effect, the system will automatically regenerate the fertilization intention and optimize the fertilization parameters. In practical applications, the monitoring data in the experimental fields show that as the wheat enters the heading stage, the system automatically increases the proportion of nitrogen fertilizer to promote grain development. At the same time, if the expected plant height growth rate is not reached after fertilization (such as a 10% increase), the system will trigger the re-learning module to re-adjust the fertilization parameters and apply the optimized parameters in subsequent fertilization processes.

[0135] The experimental data show that the method of the present invention can significantly improve the application efficiency of liquid fertilizers and the growth effect of wheat. After the fertilization process throughout the growing season, the experimental fields using the method of the present invention have a 15% increase in average plant height, a 12% increase in leaf area index, a 10% increase in grain weight, and a 12% reduction in fertilizer usage compared to the control fields using traditional fertilization methods. In addition, the fertilization strategy with dynamic optimization and feedback control also significantly improves the fertilizer utilization rate and reduces the phenomena of soil salinization and nutrient loss caused by excessive fertilization. The following are some experimental data of the present invention in wheat experimental fields.

[0136] Table 1 Monthly average data table of environmental conditions in wheat experimental fields

[0137]

[0138] Table 2 Statistical table of wheat growth feedback data

[0139] Month Average plant height (cm) Leaf area index Stem strength (N) Grain weight (g / ear) March 12 1.5 5.8 1.2 April 24 2.1 6.5 1.5 May 42 2.8 7.2 1.8 June 55 3.2 8.0 2.2

[0140] Table 3 Record table of dynamic adjustment of liquid fertilizer application strategy

[0141]

[0142] In the experimental data, Table 1 records the monthly average data of the environmental conditions in wheat experimental fields, reflecting the soil humidity, pH value, nitrogen, phosphorus, potassium content, and temperature changes in the experimental fields in different months. During the period from March to June, the soil humidity increased month by month, gradually increasing from 15% to 28%, indicating that the experimental fields maintained a relatively moist state in the middle of the growing season. At the same time, the pH value of the soil changed little, basically remaining stable between 6.8 and 7.2, within the suitable pH range for wheat growth. The nitrogen, phosphorus, and potassium contents also gradually increased under the action of seasonal fertilization, effectively supporting the growth needs of wheat. The temperature gradually rose from 16°C to 26°C in June, providing suitable temperature conditions for the rapid growth of wheat.

[0143] Table 2 summarizes the growth feedback data of wheat in different months, including important growth indicators such as plant height, leaf area index, stem strength, and grain weight. From March to June, the average plant height of wheat increased from 12 cm to 55 cm, and the leaf area index also increased accordingly, rising from 1.5 to 3.2, indicating an increase in the leaf coverage area and photosynthesis ability of the plants. The stem strength increased from 5.8 N in March to 8.0 N in June, showing an enhanced lodging resistance of the crop. At the same time, the continuous increase in grain weight indicates that the precise application of liquid fertilizer has played a positive role in effectively improving the yield. These data further support the beneficial effects of the method of the present invention in the process of crop growth.

[0144] Table 3 shows the dynamic adjustment of the liquid fertilizer application strategy during the test period. The test field adjusted the fertilizer application rate, the ratio of nitrogen, phosphorus, and potassium, and the fertilization frequency according to the real-time feedback data to meet the nutritional requirements of wheat at each growth stage. In March, the system set the nitrogen fertilizer ratio at 40% to promote early leaf growth; as the crop entered the vigorous growth stage in April, the nitrogen fertilizer ratio was gradually increased to 45%. In the critical growth stages of May and June, the nitrogen fertilizer ratio was further increased to 50% and 55% to support the growth and grain development of wheat. The fertilization frequency was also increased accordingly, from the initial 2 times per month to 3 times per month to ensure continuous nutrient supply. The ratio of phosphorus and potassium fertilizers was also dynamically adjusted according to soil and crop requirements, effectively supporting root development and improving disease resistance.

[0145] Based on the above data analysis, it can be seen that the method for analyzing the application effect of liquid fertilizer of the present invention effectively improves the precision and adaptability of fertilization through real-time optimization of environmental characteristics, crop feedback, and fertilization strategies. The growth data show that the dynamically adjusted fertilization strategy significantly promotes the growth and yield of wheat, especially in terms of plant height, leaf area index, stem strength, and grain weight. At the same time, the feedback optimization mechanism of the present invention can make fine adjustments according to environmental changes and crop growth feedback, thereby achieving efficient utilization of resources and ensuring that the crop obtains the best nutritional support at different growth stages. This indicates that the present invention has good adaptability under multiple environmental conditions and can effectively improve the fertilizer utilization rate and crop yield in agricultural production.

[0146] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and all should be covered within the protection scope of the present invention.

Claims

1. A method for analyzing the application effect of liquid fertilizer based on self-supervised learning, characterized in that, It includes the following steps: S1. Collect environmental data and crop images through multiple sensors, and perform cleaning, standardization processing and feature extraction to generate initial feature data; S2. Input the initial feature data into the heterogeneous environment adaptive module, and dynamically adjust the convolution kernel size and feature fusion strategy according to different environmental features to generate environmental feature representations; S3. Based on the dual spatio-temporal feature decoupling and dynamic fusion model, input the environmental feature representation into the multi-scale temporal-spatial feature extraction module to extract the short-term spatio-temporal features and long-term spatio-temporal features of the liquid fertilizer application effect. Use the spatio-temporal decoupling mechanism to separate the feature channels in different time periods, and fuse the short-term spatio-temporal features and long-term spatio-temporal features through the dynamic weighting mechanism. Automatically adjust the fusion ratio according to the crop growth cycle and the change of fertilization effect to generate a comprehensive spatio-temporal feature representation; S4. Based on the generated comprehensive spatio-temporal feature representation, use self-supervised learning to generate fertilization features in different scenarios, and eliminate environmental interference factors through the causal reasoning module to extract the true impact features of liquid fertilizer at different growth stages; S5. Generate fertilization intention features based on the true impact features and crop growth status, dynamically adjust the fertilization strategy, and regenerate the fertilization intention when the feedback effect is not ideal; S6. Collect data on the growth changes of crops after fertilization through the real-time feedback loop, calculate the difference between the prediction result and the actual effect, and trigger the re-learning module to update the fertilization parameters when the difference exceeds the threshold; S7. Perform integrated optimization, achieve feature sharing and task coordination through multi-task collaborative learning, and automatically optimize the key parameters using Bayesian optimization.

2. The method for analyzing the application effect of liquid fertilizer based on self-supervised learning according to claim 1, wherein, The specific content of S2 includes: S21. Input the initial feature data generated in S1 into the heterogeneous environment adaptive module, and the initial feature data includes soil humidity, nutrient components, pH value and climate data; S22. In the heterogeneous environment adaptive module, dynamically select and initialize the convolution kernel parameters according to the soil humidity, nutrient components, pH value and climate data, including the initial size and initial shape of the convolution kernel; S23. Based on the multiple statistical characteristics of the initial feature data, adaptively adjust the size and shape of the convolution kernel: Among them, K adjust represents the adjusted convolution kernel size, and K init represents the initial convolution kernel size. E var represents the variance of the initial feature data, and E avg represents the mean of the initial feature data. E std represents the standard deviation of the initial feature data, and E range represents the value range of the initial feature data. E max represents the maximum value of the initial feature data, and E min represents the minimum value of the initial feature data. α a , α d , β b , β f , γ c , γ g and δ e represent control coefficients; S24. Use the adjusted convolution kernel parameters to perform convolution operations, perform multi-scale feature extraction on the initial feature data, and generate environmental feature matrices at different scales; S25. In the heterogeneous environment adaptive module, apply a multi-level feature fusion strategy to fuse the environmental feature matrices at different scales and different modalities in space, channels and scales; S26. Finally, output the generated environmental feature representation.

3. The method for analyzing the application effect of liquid fertilizer based on self-supervised learning according to claim 1, wherein, The specific content of S3 includes: S31. Input the generated environmental feature representation into the dual spatio-temporal feature decoupling and dynamic fusion model, and perform temporal decomposition on the environmental feature representation through the multi-scale temporal-spatial feature extraction module in the dual spatio-temporal feature decoupling and dynamic fusion model; S32. Through the short-term spatio-temporal feature extraction channel, the input environmental feature data is sent into the long short-term memory network structure to extract the short-term effect features after fertilization, record the rapid response features of the crop within several days or weeks, and generate the short-term spatio-temporal feature F short ; S33. Use the long-term spatio-temporal feature extraction channel to input the environmental feature data into the temporal convolutional network, extract features of the long-term impact of fertilization effects, and generate long-term spatio-temporal features F that include the impact of fertilization over several months or the entire growing season long ; S34. Apply the spatio-temporal decoupling mechanism to separate the short-term spatio-temporal feature F short and the long-term spatio-temporal feature F long into independent feature channels in different time periods, forming a short-term channel and a long-term channel; S35. Based on the dynamic weighting mechanism, set adaptive weight parameters, which act on the spatio-temporal features of the short-term channel and the long-term channel respectively to form weighted short-term features and long-term features; S36. Fuse the weighted short-term features and long-term features to generate a comprehensive spatio-temporal feature representation: Among them, F fusion represents the comprehensive spatio-temporal feature representation, exp represents the exponential function, α represents the weighting parameter of the short-term features, β represents the weighting parameter of the long-term features, γ represents the power control coefficient, δ represents the exponential control coefficient, η represents the transformation parameter of the long-term features, sinh represents the hyperbolic sine function, θ represents the coefficient of the hyperbolic sine term, and ζ represents the normalization parameter.

4. The method for analyzing the application effect of liquid fertilizer based on self-supervised learning according to claim 1, wherein, The specific content of S4 includes: S41. Input the generated comprehensive spatio-temporal feature representation into the self-supervised learning module to generate fertilization feature representations in multiple different scenarios. The self-supervised learning module generates multi-scenario fertilization features by constructing a contrast task for unlabeled data and constructs a multi-scenario fertilization feature library. S42. During the self-supervised learning process, create the target situation feature F target and the contrast situation feature F contrast using the contrastive learning method, and use the contrastive loss function L contrast to represent the difference degree between situations: where exp represents the exponential function, sim represents the similarity function, and F j contrast represents the j-th contrast scenario feature, and F positive represents the positive sample feature representation, ∥·∥ represents the norm operation of the vector, κ1 and κ2 represent the weight coefficients of the norm term, and τ represents the parameter that regulates the similarity distribution in contrastive learning; S43. Input the generated multi-scenario fertilization feature library into the causal inference module, establish the causal relationship between features through the structural equation model, and eliminate environmental interference factors. The causal inference module identifies the direct causal relationship of fertilization features on crop growth. S44. Use the latent confounder variable model to correct the fertilization features and crop growth features, identify the influence of latent environmental variables and eliminate noise interference. Among them, Y represents the crop growth impact, X represents the fertilization characteristics, Z represents the potential environmental variables, α x , β z , γ p , δ s and θ x represent weight coefficients, and η y represents the normalization coefficient; S45. Apply the counterfactual reasoning method to create counterfactual scenario features F counter , simulate the crop growth effects under different fertilization conditions, adjust the fertilization features through intervention operations, and generate crop responses under hypothetical scenarios; S46. By comparing the actual situation feature F actual with the counterfactual situation feature F counter , extract the true impact features of the liquid fertilizer at different growth stages, eliminate the interference of potential environmental variables, and finally obtain the true fertilization effect during the crop growth process.

5. A method for analyzing the application effect of liquid fertilizer based on self-supervised learning according to claim 1, characterized in that, The specific content of S5 includes: S51. Input the generated true impact feature F true and the current growth state S of the crop into the fertilization intention generation module to generate the initial fertilization intention feature I initial , where the initial fertilization intention feature I initial is used to represent the best strategy for liquid fertilization in the current environment; S52. During the generation process of fertilization intention, based on the dynamic mapping of the true influence feature F true and the current growth state S of the crop, a fertilization intention function is constructed. The fertilization intention function synthesizes the weights of different features to generate an initial fertilization intention feature I that adapts to different environments and crop requirements initial ; S53. According to the generated initial fertilization intention feature I initial , generate a set of fertilization strategy parameters P, including the fertilization amount P amount , the fertilization frequency P freq and the fertilization concentration P conc to achieve the optimal fertilization effect; S54. During the fertilization process, based on the real-time feedback data R of crop growth actual monitor the fertilization effect, and compare the feedback result with the expected effect R expected If the deviation exceeds the set threshold, trigger the fertilization intention update mechanism: Among them, D represents the deviation between the actual effect and the expected effect, α growth represents the growth-related adjustment coefficient, β env represents the weight coefficient of environmental characteristics, γ fert represents the adjustment coefficient of fertilization characteristics, δ crop represents the adjustment coefficient of crop characteristics, ∈ represents the smoothing term; The fertilization intention update mechanism S55 adjusts the initial fertilization intention feature I according to the deviation D between the actual effect and the expected effect initial : Among them, I update represents the updated fertilization intention feature, exp represents the exponential function, λ growth represents the growth regulation coefficient, λ env represents the environmental regulation coefficient, θ fert represents the fertilization feature regulation coefficient, ζ crop represents the crop feature regulation coefficient, η growth represents the gain control coefficient; S56. According to the updated fertilization intention feature I update , regenerate the set of fertilization strategy parameters P and apply it to the process of liquid fertilizer application.

6. The method for analyzing the application effect of liquid fertilizer based on self-supervised learning according to claim 1, characterized in that, The specific content of S7 includes: S71. Based on the multi-task collaborative learning mechanism, construct a unified feature vector set to achieve feature sharing and form a stable multi-task fertilization model. S72. On the basis of feature sharing, generate independent task representations for each fertilization task. While sharing features, each fertilization task retains its own characteristics. Based on the designed task-specific encoding module, automatically identify and utilize the feature subset suitable for a specific task when dealing with different tasks. S73. In multi-task collaborative learning, introduce an interaction module between tasks to capture the potential correlation relationships between different tasks, generate joint feature representations, and through the weighted mechanism of the interaction module, enable the learning process of one task to feedback to the feature update of other tasks. S74. On the basis of the multi-task learning framework, use the Bayesian optimization method to automatically tune the key parameters, construct a parameter space, and gradually update the parameter distribution according to the feedback data of the fertilization tasks. S75. Apply a surrogate model in Bayesian optimization to quickly evaluate different parameter combinations, simulate the feedback situation in the fertilization process through the surrogate model, and update the surrogate model based on historical data to learn the optimal parameter configuration from less experimental data. S76. After the parameter tuning is completed, apply the obtained set of optimal parameter configurations to the multi-task collaborative model to achieve the overall optimization of the fertilization strategy. Through the integrated and optimized multi-task collaborative learning model, each fertilization task obtains the best fertilization effect in different environments.

Images