Soil improvement measure decision-making method and system based on multi-modal deep learning

Abstract

The invention discloses a soil improvement measure decision-making method and system based on multi-modal deep learning. According to the method, firstly, soil state parameter change data and crop growth data of a target planting site are acquired, and a preliminary soil improvement scheme is established by combining the two data; then, a multi-modal deep learning network model is constructed, climate change characteristics and soil conditioner content change characteristics after improvement measures are implemented are fused and analyzed, and data characteristics of influences of climate changes on the conditioner are extracted; and based on the influence data, further identifying the failure characteristics of the modifier, including the failure propagation path and the dose change characteristics of the modifier. And finally, performing dynamic compensation on the original soil improvement measures according to the failure characteristics to form a refined compensation scheme. According to the method, intelligent adjustment and dynamic optimization of the soil improvement strategy can be realized, and the method has relatively high agricultural application value and popularization prospect.

Description

Technical Field

[0001] The present invention relates to the field of soil improvement technology, and in particular to a soil improvement measure decision-making method and system based on multimodal deep learning. Background Art

[0002] Soil, the foundation of agricultural production, has a direct impact on crop growth and yields due to its physical and chemical properties. However, long-term farming practices, excessive use of fertilizers and pesticides, and inappropriate land management practices have led to increasingly serious problems such as soil acidification, salinization, decreased organic matter content, and heavy metal contamination, hindering the healthy development of agriculture.

[0003] Traditional decision-making methods for soil improvement measures rely primarily on the empirical judgment of agronomists and limited soil testing data, making them incapable of addressing complex soil environmental changes influenced by multivariate interactions. In practical applications, these approaches ignore the temporal and spatial evolution of soil properties, as well as the dynamic relationships among heterogeneous factors such as climate change and crop root distribution. This leads to significant deviations in the application method, dosage, and timing of soil amendments, compromising improvement effectiveness and even causing waste of resources or secondary soil contamination.

[0004] In recent years, data-driven soil analysis and prediction methods have gradually emerged. Multimodal deep learning, an intelligent algorithmic framework capable of integrating information from diverse data sources, has demonstrated powerful modeling and decision-making capabilities in fields such as medical imaging, intelligent transportation, and environmental monitoring. However, systematic research and application of multimodal deep learning methods integrated with agricultural data for the development of soil improvement measures is still lacking.

[0005] Existing technologies lack comprehensive modeling capabilities for soil state parameters, crop growth data, climate change characteristics, and the dynamic behavior of amendments. This makes it impossible to dynamically predict and optimize soil amendment measures throughout their entire lifecycle, hindering the urgent need for high-precision, real-time, and intelligent decision-making in modern precision agriculture. Therefore, there is an urgent need for an intelligent decision-making method and system for soil amendments that integrates multiple data types and possesses spatiotemporal analysis capabilities to improve the scientific nature and efficiency of soil management. Summary of the Invention

[0006] In order to solve at least one of the above technical problems, the present invention proposes a soil improvement measure decision-making method and system based on multimodal deep learning.

[0007] A first aspect of the present invention provides a soil improvement measure decision-making method based on multimodal deep learning, comprising:

[0008] Acquiring soil state parameter change data and crop growth data of the target planting site, and determining soil improvement measures for the target planting site based on the soil state parameters and crop growth data;

[0009] A multimodal deep learning network was constructed to conduct a correlation analysis between the climate change characteristics and the changes in soil amendment content at the target planting site after the implementation of soil amendment measures within a preset time period, thereby determining the impact of climate change on soil amendments.

[0010] Determining, based on the impact data, the failure characteristics of the improver at the target planting site, wherein the failure characteristics include the failure propagation path characteristics and the failure dose variation characteristics of the improver;

[0011] Compensation operations are performed on soil improvement measures based on the failure characteristics of the amendments, and a compensation plan is constructed.

[0012] In this solution, the soil state parameter change data and crop growth data of the target planting site are obtained, and soil improvement measures for the target planting site are determined based on the soil state parameters and crop growth data, specifically:

[0013] Divide the soil layer at a preset depth in the target planting area into N three-dimensional sub-areas, and obtain soil state parameter change data for each sub-area within a preset historical time period, wherein the soil state parameter change data includes changes in pH value, organic matter content, and heavy metal pollutant type and content within the preset historical time period;

[0014] Acquire data on the type of crops to be planted in the target planting area, and acquire crop growth data of the type of crops to be planted, wherein the crop growth data includes soil condition requirement data for crop growth, crop root growth depth distribution data, crop planting range data, and growth cycle;

[0015] Determine the distribution of each soil state parameter feature in the target planting area based on the soil state parameter change data of each sub-area, and construct soil state parameter distribution maps at multiple historical preset time points based on the distribution;

[0016] Performing a time series analysis on the soil state parameter distribution map to determine the migration path and migration rate of each soil state parameter characteristic in the target planting area within a historical preset time period, thereby obtaining the migration characteristics of the soil state in the target planting area;

[0017] Building a migration prediction model based on the ARIMA algorithm, importing the migration features into the prediction model for training, and determining model parameters, wherein the model parameters include autoregressive order, difference order, and moving average order;

[0018] Importing the migration characteristics of the preset time period before the planting of the crops to be planted into the fully trained migration prediction model, predicting the migration characteristics of the soil state during the growth cycle of the crops to be planted, and obtaining the predicted migration characteristics;

[0019] Soil improvement measures for the target planting site are determined based on the predicted migration characteristics and crop growth data.

[0020] In this solution, the soil improvement measures for the target planting site are determined based on the predicted migration characteristics and crop growth data, specifically:

[0021] According to the crop root growth depth distribution data, the root growth area of ​​the crops to be planted in the target planting area during the growth cycle is used as the soil improvement area, and the soil conditioner application position is determined according to the soil improvement area;

[0022] predicting soil parameter changes in each sub-region of the soil improvement region during the growth cycle based on the predicted migration characteristics to obtain predicted soil parameter changes;

[0023] Determining the type of soil parameters to be improved in the soil improvement area during the growth cycle based on the predicted soil parameter changes, determining the type of amendment in the soil improvement area based on the type of soil parameters to be improved, and determining the amount of amendment to be applied based on the predicted soil parameter changes;

[0024] Soil improvement measures for the target planting site are determined based on the application location, type and amount of the soil conditioner.

[0025] In this solution, a multimodal deep learning network is constructed to perform a correlation analysis on the climate change characteristics and soil conditioner content change characteristics of the target planting site after the implementation of soil improvement measures within a preset time period, and to determine the impact data of climate change on soil conditioners, specifically:

[0026] Construct a multimodal deep learning network based on the XGBoost gradient boosting tree model, recurrent neural network, and temporal graph neural network;

[0027] Implement the soil improvement measures and obtain time series data on climate change characteristics and soil conditioner content change characteristics of the target planting site within a preset time period after the soil improvement measures are implemented, wherein the climate change characteristics include changes in temperature, precipitation, wind speed, and air humidity; and the soil conditioner content change characteristics include concentration change values ​​of each soil conditioner type in each sub-region of the target planting area;

[0028] Performing a time series alignment operation on the climate change characteristic time series data and the soil conditioner content change characteristic time series data, and importing the climate change characteristic as the input feature vector and the soil conditioner content change characteristic as the model output label into the XGBoost gradient boosting tree model for training;

[0029] All climate variables are ranked in importance based on the feature gain scoring mechanism built into the XGBoost gradient boosting tree model, and the top climate variables with the most significant impact on the change in soil amendment concentration are identified as key climate factors based on their cumulative gain contribution during the training process.

[0030] Inputting the time series of the key climate factors into a recurrent neural network model including a gating mechanism to perform time series modeling training, and extracting the time series change characteristics of the key climate factors;

[0031] A graph structure model is established based on the diffusion behavior of soil amendments between each sub-region, with each sub-region as a graph node and the propagation probability, diffusion time lag, and concentration gradient of the amendment between adjacent regions as graph edge attributes. A graph adjacency matrix and a time-step transfer matrix are constructed based on a temporal graph neural network model, and the temporal graph neural network is trained to extract the spatial diffusion characteristics of the soil amendments.

[0032] A correlation analysis is performed on the time series variation characteristics of the key climate factors and the spatial diffusion characteristics of the soil conditioner to determine the impact of climate change on the soil conditioner and obtain impact data.

[0033] In this solution, the failure characteristics of the modifier in the target planting site are determined based on the impact data. The failure characteristics include the failure propagation path characteristics and the failure dose variation characteristics of the modifier, specifically:

[0034] Obtaining historical climate change characteristic data for a target planting area, importing the historical climate change characteristic data into a climate prediction model based on a long short-term memory network for training based on the historical climate change characteristic data, and predicting climate change for a preset implementation period after soil improvement measures are implemented in the target planting area to obtain climate change prediction data;

[0035] Determine, based on the impact data and climate change prediction data, a time series of concentration changes of the soil amendment in each sub-region after the implementation of the soil amendment measures, and determine a failure time node at which the concentration falls below a preset threshold;

[0036] Connecting each sub-region in order of failure time nodes to determine the failure propagation direction of the soil conditioner, and determining the failure propagation path characteristics of the soil conditioner based on the failure propagation direction;

[0037] The modifier concentration change rate of each sub-region is calculated according to the concentration change time series, and the modifier failure dose change characteristics are determined based on the concentration change rate.

[0038] In this solution, the soil improvement measures are compensated according to the failure characteristics of the improver, and a compensation solution is constructed, specifically:

[0039] Determine a sub-region set of the soil conditioner on the failure propagation path according to the characteristics of the conditioner failure path, mark the sub-region to be compensated, and determine the conditioner compensation priority of each sub-region to be compensated according to the conditioner failure time node in the sub-region to be compensated;

[0040] Determining the initial compensation application amount of the soil conditioner for each sub-area to be compensated according to the failure dose variation characteristics;

[0041] Comparing the initial compensation application amount with a preset safety application amount threshold, if the initial compensation application amount is greater than the safety application amount threshold, dividing the initial compensation application amount into multiple stage application amounts based on the failure propagation path characteristics, and determining the application time of each stage application amount according to the climate change mutation time node in the climate change prediction data to obtain an application time set;

[0042] If the initial compensation application amount is not greater than the safe application amount threshold, a soil conditioner compensation application operation is performed on the sub-area to be compensated according to the initial compensation application amount, and the time point at which the soil conditioner concentration change rate is minimum after the expiration time node is determined based on the climate change prediction data and impact data, and the time point at which the concentration change rate is minimum is used as the soil conditioner compensation application time point;

[0043] Compensation operations are performed on soil improvement measures according to the amendment compensation priority, stage application amount or initial compensation application amount, application time set or compensation application time point to construct a compensation plan.

[0044] A second aspect of the present invention further provides a soil improvement measure decision-making system based on multimodal deep learning, the system comprising: a memory and a processor, wherein the memory includes a soil improvement measure decision-making method program based on multimodal deep learning, and when the soil improvement measure decision-making method program based on multimodal deep learning is executed by the processor, the following steps are implemented:

[0045] Acquiring soil state parameter change data and crop growth data of the target planting site, and determining soil improvement measures for the target planting site based on the soil state parameters and crop growth data;

[0046] A multimodal deep learning network was constructed to conduct a correlation analysis between the climate change characteristics and the changes in soil amendment content at the target planting site after the implementation of soil amendment measures within a preset time period, thereby determining the impact of climate change on soil amendments.

[0047] Determining, based on the impact data, the failure characteristics of the improver at the target planting site, wherein the failure characteristics include the failure propagation path characteristics and the failure dose variation characteristics of the improver;

[0048] Compensation operations are performed on soil improvement measures based on the failure characteristics of the amendments, and a compensation plan is constructed.

[0049] The present invention discloses a soil improvement measure decision-making method and system based on multimodal deep learning. The method first obtains the soil state parameter change data and crop growth data of the target planting site, and combines the two to establish a preliminary soil improvement plan. Subsequently, a multimodal deep learning network model is constructed to integrate and analyze the climate change characteristics and soil conditioner content change characteristics after the implementation of the improvement measures, and extract the data characteristics of the impact of climate change on the conditioner. Based on the impact data, the failure characteristics of the conditioner are further identified, including its failure propagation path and dosage change characteristics. Finally, the original soil improvement measures are dynamically compensated according to the failure characteristics to form a refined compensation plan. This method can realize the intelligent adjustment and dynamic optimization of soil improvement strategies, and has high agricultural application value and promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of a soil improvement decision-making method based on multimodal deep learning according to the present invention is shown;

[0051] Figure 2 A flow chart showing soil improvement measures for determining a target planting site according to the present invention is shown;

[0052] Figure 3 A flow chart illustrating the present invention for determining failure characteristics of an amendment for a target planting site;

[0053] Figure 4 A block diagram of a soil improvement measure decision-making system based on multimodal deep learning of the present invention is shown. DETAILED DESCRIPTION

[0054] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present application and the features therein can be combined with each other.

[0055] 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. Therefore, the scope of protection of the present invention is not limited to the specific embodiments disclosed below.

[0056] Figure 1 A flowchart of a soil improvement measure decision-making method based on multimodal deep learning of the present invention is shown.

[0057] like Figure 1 As shown, the first aspect of the present invention provides a soil improvement measure decision-making method based on multimodal deep learning, comprising:

[0058] S102, obtaining soil state parameter change data and crop growth data of the target planting site, and determining soil improvement measures for the target planting site based on the soil state parameters and crop growth data;

[0059] S104, constructing a multimodal deep learning network, and performing a correlation analysis on the climate change characteristics and soil amendment content change characteristics of the target planting site after the implementation of the soil amendment measures within a preset time period based on the multimodal deep learning network, to determine the impact data of climate change on the soil amendment;

[0060] S106, determining the failure characteristics of the improver in the target planting site based on the impact data, wherein the failure characteristics include the failure propagation path characteristics and the failure dose variation characteristics of the improver;

[0061] S108, performing compensation operations on the soil improvement measures according to the failure characteristics of the improver and constructing a compensation plan.

[0062] It should be noted that by obtaining the soil state parameter change data and crop growth data of the target planting site, a comprehensive perception of soil quality and crop growth needs can be achieved, and the scientificity and adaptability of soil improvement measures can be improved; by constructing a multimodal deep learning network that integrates XGBoost, recurrent neural networks and time series graph neural networks, the complex relationship between climate change characteristics and soil conditioner content change characteristics after the implementation of soil improvement measures can be deeply mined, thereby accurately identifying the impact of key climate factors on the behavior of conditioners and enhancing the system's sensitivity to environmental disturbances; further by analyzing the above-mentioned impact data, the failure propagation path and dosage change characteristics of conditioners in different sub-regions can be extracted, which can realize the prediction and positioning of the failure trend of conditioners in time and space dimensions, and improve the timeliness and accuracy of dynamic monitoring of soil improvement; finally, differentiated compensation operations can be carried out according to failure characteristics, which can realize intelligent optimization and regulation of the application amount and timing of conditioners, improve the utilization rate of conditioners and reduce resource waste.

[0063] According to an embodiment of the present invention, the step of obtaining soil state parameter change data and crop growth data of a target planting site and determining soil improvement measures for the target planting site based on the soil state parameters and crop growth data is as follows:

[0064] Divide the soil layer at a preset depth in the target planting area into N three-dimensional sub-areas, and obtain soil state parameter change data for each sub-area within a preset historical time period, wherein the soil state parameter change data includes changes in pH value, organic matter content, and heavy metal pollutant type and content within the preset historical time period;

[0065] Acquire data on the type of crops to be planted in the target planting area, and acquire crop growth data of the type of crops to be planted, wherein the crop growth data includes soil condition requirement data for crop growth, crop root growth depth distribution data, crop planting range data, and growth cycle;

[0066] Determine the distribution of each soil state parameter feature in the target planting area based on the soil state parameter change data of each sub-area, and construct soil state parameter distribution maps at multiple historical preset time points based on the distribution;

[0067] Performing a time series analysis on the soil state parameter distribution map to determine the migration path and migration rate of each soil state parameter characteristic in the target planting area within a historical preset time period, thereby obtaining the migration characteristics of the soil state in the target planting area;

[0068] Building a migration prediction model based on the ARIMA algorithm, importing the migration features into the prediction model for training, and determining model parameters, wherein the model parameters include autoregressive order, difference order, and moving average order;

[0069] Importing the migration characteristics of the preset time period before the planting of the crops to be planted into the fully trained migration prediction model, predicting the migration characteristics of the soil state during the growth cycle of the crops to be planted, and obtaining the predicted migration characteristics;

[0070] Soil improvement measures for the target planting site are determined based on the predicted migration characteristics and crop growth data.

[0071] It should be noted that since soil state parameters (such as pH value, organic matter content, heavy metal (such as chromium, lead, arsenic, chromium, etc.) pollution, etc.) often show temporal variation patterns under natural conditions and human intervention, it will be difficult to effectively reflect their future change trends if we only rely on static historical data, which may lead to lags or mismatches in improvement measures. The ARIMA algorithm is used to construct a migration prediction model, which can perform differential stabilization on the change series of various soil parameters, and combine the autoregressive and sliding average structures to model and fit their change trends, periodic fluctuations and disturbance patterns, thereby fully exploring the intrinsic time series characteristics of soil parameters and achieving high-precision predictions of parameter migration paths and change amplitudes in future time periods. The historical preset time period is the time period between the current time point and the preset time length in the past. The soil state parameter characteristics are each state parameter item, namely the pH value parameter item, the organic matter content item, and the heavy metal pollutant type and content parameter item.

[0072] Figure 2 A flow chart of the present invention for determining soil improvement measures for a target planting site is shown.

[0073] According to an embodiment of the present invention, determining soil improvement measures for a target planting site based on the predicted migration characteristics and crop growth data is specifically as follows:

[0074] S202, based on the crop root growth depth distribution data, the root growth area of ​​the crop to be planted in the target planting area during the growth cycle is used as the soil improvement area, and the soil conditioner application position is determined based on the soil improvement area;

[0075] S204, predicting soil parameter changes in each sub-region of the soil improvement region during the growth cycle based on the predicted migration characteristics to obtain predicted soil parameter changes;

[0076] S206, determining the type of soil parameters to be improved in the soil improvement area within the growth cycle based on the predicted soil parameter changes, determining the type of amendment in the soil improvement area based on the type of soil parameters to be improved, and determining the amount of amendment to be applied based on the predicted soil parameter changes;

[0077] S208: Determine soil improvement measures for the target planting site based on the application location, type, and application amount of the soil conditioner.

[0078] It should be noted that determining soil improvement measures for target planting sites based on predicted migration characteristics and crop growth data can achieve dual optimization of spatial accuracy and temporal adaptability of soil improvement operations, significantly improving the scientific nature and resource utilization efficiency of improvement measures. By introducing crop root growth depth distribution data, the soil layer where the roots may be distributed is accurately delineated as the soil improvement area, which can effectively focus the improvement operation on the soil space where the crops actually need to be improved, avoiding waste of resources and interference with non-target areas; based on the predicted migration characteristics, the soil parameter change trend of each sub-area in the improvement area during the entire growth cycle is dynamically predicted, which can not only identify in advance the risk points that the soil parameters will exceed the crop adaptation range in the future, but also accurately locate the change rate and severity of each parameter, and then select targeted amendment types according to different types of parameters to be improved (such as abnormal pH, poor organic matter, excessive heavy metals, etc.); in addition, by scientifically determining the amount of amendment applied in combination with the predicted parameter change trend, it can dynamically match actual needs and avoid adverse consequences caused by insufficient or excessive application. The types of amendments include lime, organic fertilizer, silicon fertilizer and selenium fertilizer.

[0079] According to an embodiment of the present invention, the multimodal deep learning network is constructed, and a correlation analysis is performed on the climate change characteristics and soil conditioner content change characteristics of the target planting site after the implementation of soil improvement measures within a preset time period based on the multimodal deep learning network to determine the impact data of climate change on soil conditioners, specifically:

[0080] Construct a multimodal deep learning network based on the XGBoost gradient boosting tree model, recurrent neural network, and temporal graph neural network;

[0081] Implement the soil improvement measures and obtain time series data on climate change characteristics and soil conditioner content change characteristics of the target planting site within a preset time period after the soil improvement measures are implemented, wherein the climate change characteristics include changes in temperature, precipitation, wind speed, and air humidity; and the soil conditioner content change characteristics include concentration change values ​​of each soil conditioner type in each sub-region of the target planting area;

[0082] Performing a time series alignment operation on the climate change characteristic time series data and the soil conditioner content change characteristic time series data, and importing the climate change characteristic as the input feature vector and the soil conditioner content change characteristic as the model output label into the XGBoost gradient boosting tree model for training;

[0083] All climate variables are ranked in importance based on the feature gain scoring mechanism built into the XGBoost gradient boosting tree model, and the top climate variables with the most significant impact on the change in soil amendment concentration are identified as key climate factors based on their cumulative gain contribution during the training process.

[0084] Inputting the time series of the key climate factors into a recurrent neural network model including a gating mechanism to perform time series modeling training, and extracting the time series change characteristics of the key climate factors;

[0085] A graph structure model is established based on the diffusion behavior of soil amendments between each sub-region, with each sub-region as a graph node and the propagation probability, diffusion time lag, and concentration gradient of the amendment between adjacent regions as graph edge attributes. A graph adjacency matrix and a time-step transfer matrix are constructed based on a temporal graph neural network model, and the temporal graph neural network is trained to extract the spatial diffusion characteristics of the soil amendments.

[0086] A correlation analysis is performed on the time series variation characteristics of the key climate factors and the spatial diffusion characteristics of the soil conditioner to determine the impact of climate change on the soil conditioner and obtain impact data.

[0087] It should be noted that by combining the XGBoost gradient boosting tree model, recurrent neural network (RNN) and time series graph neural network (TGNN) to construct a multimodal deep learning network, it is possible to fully integrate the nonlinear characteristic contribution, temporal dynamic evolution law and spatial diffusion behavior of climate factors, so as to accurately identify the key data of the impact of climate change on soil conditioners. The specific logical steps are as follows: First, the XGBoost gradient boosting tree model is used to perform supervised learning training on the input climate change characteristic time series and the output soil conditioner content change characteristic time series. The model has powerful nonlinear fitting capabilities and feature gain analysis mechanisms. It can scientifically evaluate the degree of influence of each type of climate variable (such as temperature, precipitation, wind speed, humidity) on the change of conditioner concentration through the accumulated feature gain contribution during the training process, thereby screening out the key climate factors that really have a significant impact; secondly, for these key climate factors, a recurrent neural network with a gating mechanism is used for modeling. This structure can capture the dynamics of key climate variables over a long span of time. Change pattern, excavate its internal time dependence and trend characteristics; then, combined with the diffusion characteristics of soil amendments in the planting area, model the sub-regions as nodes in the graph structure, construct edge attributes through the diffusion probability, time lag and concentration gradient between adjacent sub-regions, introduce the time series graph neural network to model the graph structure, thereby extracting the diffusion law characteristics of the amendment at multiple times and across space; finally, the time series characteristics of the climate factors extracted from the RNN and the diffusion characteristics of the amendments extracted from the TGNN are correlated to achieve a quantitative analysis of how climate change drives the evolution of soil amendment concentrations and diffusion behavior, thereby forming accurate, dynamic and regionally differentiated impact data. The first several climate variables are user-set values. The impact data include quantitative impact information of changes in key climate factors on the concentration change rate, diffusion path offset and attenuation period of various types of soil amendments in different sub-regions.

[0088] Figure 3 A flow chart for determining the failure characteristics of an amendment for a target planting site according to the present invention is shown.

[0089] According to an embodiment of the present invention, the failure characteristics of the modifier in the target planting site are determined based on the impact data. The failure characteristics include the failure propagation path characteristics of the modifier and the failure dose variation characteristics. Specifically,

[0090] S302, obtaining historical climate change characteristic data for the target planting area, importing the historical climate change characteristic data into a climate prediction model based on a long short-term memory network for training, and predicting climate change for a preset implementation period after soil improvement measures are implemented in the target planting area, thereby obtaining climate change prediction data;

[0091] S304, determining a time series of concentration changes of the soil conditioner in each sub-region after the implementation of the soil improvement measures based on the impact data and the climate change prediction data, and determining an expiration time node at which the concentration falls below a preset threshold;

[0092] S306: Connect each sub-region in order of the failure time nodes to determine the failure propagation direction of the soil conditioner, and determine the failure propagation path characteristics of the soil conditioner based on the failure propagation direction;

[0093] S308 , calculating the modifying agent concentration change rate of each sub-region according to the concentration change time series, and determining the modifying agent failure dose change characteristics based on the concentration change rate.

[0094] It should be noted that the impact data itself has been modeled based on the deep correlation between climate change characteristics and changes in soil amendment content, enabling prediction of changes in amendment behavior driven by climate conditions. By integrating historical climate change characteristic data into a climate prediction model based on a long short-term memory (LSTM) network, we can accurately capture climate change trends within a specific future time period. Furthermore, by combining the "climate change-amendment concentration change" response mechanism reflected in the impact data, we can deduce the future evolution of soil amendment concentrations within different subregions. Based on the predicted concentration change time series, we can clearly identify the time point when the concentration in each subregion falls below the safety threshold. By sequencing these failure times, we can infer the direction and trajectory of the amendment failure propagation within the planting site, thereby determining the characteristics of the failure propagation path. Furthermore, by calculating the derivative of the concentration gradient in each subregion's concentration change time series, we can extract information on the rate of amendment consumption or degradation, namely, the amendment concentration change rate. This allows us to further derive the evolutionary trend of the failure dose in different regions, namely, the failure dose variation characteristics. Overall, this method organically integrates climate impact mechanisms, time series prediction, and spatial diffusion dynamics, enabling the precise characterization and quantification of amendment failure behavior.

[0095] According to an embodiment of the present invention, the compensation operation for the soil improvement measures is performed according to the failure characteristics of the improver, and a compensation scheme is constructed, specifically:

[0096] Determine a sub-region set of the soil conditioner on the failure propagation path according to the characteristics of the conditioner failure path, mark the sub-region to be compensated, and determine the conditioner compensation priority of each sub-region to be compensated according to the conditioner failure time node in the sub-region to be compensated;

[0097] Determining the initial compensation application amount of the soil conditioner for each sub-area to be compensated according to the failure dose variation characteristics;

[0098] Comparing the initial compensation application amount with a preset safety application amount threshold, if the initial compensation application amount is greater than the safety application amount threshold, dividing the initial compensation application amount into multiple stage application amounts based on the failure propagation path characteristics, and determining the application time of each stage application amount according to the climate change mutation time node in the climate change prediction data to obtain an application time set;

[0099] It should be noted that during the application of soil conditioners, excessive application dosages may lead to increased soil environmental burden, increased cumulative toxicity risks of conditioners, and deterioration of the crop growth environment. Therefore, when the initial compensation application amount exceeds the safe application amount threshold, the system will split the compensation application amount into multiple stages based on the failure propagation path characteristics. Climate change is usually accompanied by significant changes in key environmental factors such as temperature, precipitation, and humidity. These changes directly affect the diffusion rate, degradation rate, and effectiveness of soil conditioners. By accurately identifying the time nodes of climate change, the application time of soil conditioners can be reasonably arranged so that the application amount of each stage coincides with the time period when soil conditions are most conducive to the absorption and action of the conditioner, avoiding the loss or failure of the conditioner caused by application under adverse climatic conditions, and maximizing the application efficiency and sustained effect.

[0100] If the initial compensation application amount is not greater than the safe application amount threshold, a soil conditioner compensation application operation is performed on the sub-area to be compensated according to the initial compensation application amount, and the time point at which the soil conditioner concentration change rate is minimum after the expiration time node is determined based on the climate change prediction data and impact data, and the time point at which the concentration change rate is minimum is used as the soil conditioner compensation application time point;

[0101] It should be noted that when the initial compensatory application amount is not greater than the preset safe application amount threshold, the time point with the minimum concentration change rate of the soil conditioner after the expiration time node is determined based on climate change prediction data and impact data, and this time point is used as the compensation application time point. This is because the time point with the minimum concentration change rate represents the stage where the concentration of the soil conditioner tends to stabilize, and the loss and degradation rate slow down. At this time, the application of the compensatory agent can minimize the application waste caused by environmental disturbances and ensure the continued effect of the effective ingredients of the conditioner.

[0102] Compensation operations are performed on soil improvement measures according to the amendment compensation priority, stage application amount or initial compensation application amount, application time set or compensation application time point to construct a compensation plan.

[0103] It should be noted that compensating soil amendment measures based on the priority of amendment compensation, stage application amount or initial compensation application amount, and application time set or compensation application time point can achieve precise and phased management of soil amendments, effectively avoiding environmental risks and resource waste caused by one-time excessive application, while ensuring the timeliness and continuity of compensation application and improving the utilization efficiency of amendments. Prioritization ensures that key failure areas receive priority compensation, combined with scientific and reasonable application amount and application time, adapting to the dynamic impact of climate change on the soil environment, and dynamically adjusting compensation plans to achieve stable release and sustained effect of soil amendments.

[0104] Figure 4 A block diagram of a soil improvement measure decision-making system based on multimodal deep learning of the present invention is shown.

[0105] A second aspect of the present invention further provides a soil improvement measure decision-making system 4 based on multimodal deep learning, the system comprising: a memory 41 and a processor 42, wherein the memory includes a soil improvement measure decision-making method program based on multimodal deep learning, and when the soil improvement measure decision-making method program based on multimodal deep learning is executed by the processor, the following steps are implemented:

[0106] Acquiring soil state parameter change data and crop growth data of the target planting site, and determining soil improvement measures for the target planting site based on the soil state parameters and crop growth data;

[0107] A multimodal deep learning network was constructed to conduct a correlation analysis between the climate change characteristics and the changes in soil amendment content at the target planting site after the implementation of soil amendment measures within a preset time period, thereby determining the impact of climate change on soil amendments.

[0108] Determining, based on the impact data, the failure characteristics of the improver at the target planting site, wherein the failure characteristics include the failure propagation path characteristics and the failure dose variation characteristics of the improver;

[0109] Compensation operations are performed on soil improvement measures based on the failure characteristics of the amendments, and a compensation plan is constructed.

[0110] The present invention discloses a soil improvement measure decision-making method and system based on multimodal deep learning. The method first obtains the soil state parameter change data and crop growth data of the target planting site, and combines the two to establish a preliminary soil improvement plan. Subsequently, a multimodal deep learning network model is constructed to integrate and analyze the climate change characteristics and soil conditioner content change characteristics after the implementation of the improvement measures, and extract the data characteristics of the impact of climate change on the conditioner. Based on the impact data, the failure characteristics of the conditioner are further identified, including its failure propagation path and dosage change characteristics. Finally, the original soil improvement measures are dynamically compensated according to the failure characteristics to form a refined compensation plan. This method can realize the intelligent adjustment and dynamic optimization of soil improvement strategies, and has high agricultural application value and promotion prospects.

[0111] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0112] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the scheme of this embodiment.

[0113] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0114] Those skilled in the art will appreciate that all or part of the steps of the above-mentioned method embodiments may be implemented by hardware associated with program instructions, and the aforementioned program may be stored in a computer-readable storage medium. When the program is executed, the program executes the steps of the above-mentioned method embodiments. The aforementioned storage medium includes various media that can store program codes, such as mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0115] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a number of instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0116] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A soil improvement decision-making method based on multimodal deep learning, characterized in that: The following steps are involved: Acquiring soil state parameter change data and crop growth data of the target planting site, and determining soil improvement measures for the target planting site based on the soil state parameters and crop growth data; A multimodal deep learning network was constructed to conduct a correlation analysis between the climate change characteristics and the changes in soil amendment content at the target planting site after the implementation of soil amendment measures within a preset time period, thereby determining the impact of climate change on soil amendments. Determining, based on the impact data, the failure characteristics of the improver at the target planting site, wherein the failure characteristics include the failure propagation path characteristics and the failure dose variation characteristics of the improver; Compensation operations are performed on soil improvement measures based on the failure characteristics of the amendments, and a compensation plan is constructed.

2. A soil improvement measure decision-making method based on multimodal deep learning according to claim 1, characterized in that: The acquiring of soil state parameter change data and crop growth data of the target planting site, and determining soil improvement measures for the target planting site based on the soil state parameters and crop growth data, specifically includes: Divide the soil layer at a preset depth in the target planting area into N three-dimensional sub-areas, and obtain soil state parameter change data for each sub-area within a preset historical time period, wherein the soil state parameter change data includes changes in pH value, organic matter content, and heavy metal pollutant type and content within the preset historical time period; Acquire data on the type of crops to be planted in the target planting area, and acquire crop growth data of the type of crops to be planted, wherein the crop growth data includes soil condition requirement data for crop growth, crop root growth depth distribution data, crop planting range data, and growth cycle; Determine the distribution of each soil state parameter feature in the target planting area based on the soil state parameter change data of each sub-area, and construct soil state parameter distribution maps at multiple historical preset time points based on the distribution; Performing a time series analysis on the soil state parameter distribution map to determine the migration path and migration rate of each soil state parameter characteristic in the target planting area within a historical preset time period, thereby obtaining the migration characteristics of the soil state in the target planting area; Building a migration prediction model based on the ARIMA algorithm, importing the migration features into the prediction model for training, and determining model parameters, wherein the model parameters include autoregressive order, difference order, and moving average order; Importing the migration characteristics of the preset time period before the planting of the crops to be planted into the fully trained migration prediction model, predicting the migration characteristics of the soil state during the growth cycle of the crops to be planted, and obtaining the predicted migration characteristics; Soil improvement measures for the target planting site are determined based on the predicted migration characteristics and crop growth data.

3. A soil improvement measure decision-making method based on multimodal deep learning according to claim 2, characterized in that: The soil improvement measures for the target planting site are determined based on the predicted migration characteristics and crop growth data, specifically: According to the crop root growth depth distribution data, the root growth area of ​​the crops to be planted in the target planting area during the growth cycle is used as the soil improvement area, and the soil conditioner application position is determined according to the soil improvement area; predicting soil parameter changes in each sub-region of the soil improvement region during the growth cycle based on the predicted migration characteristics to obtain predicted soil parameter changes; Determining the type of soil parameters to be improved in the soil improvement area during the growth cycle based on the predicted soil parameter changes, determining the type of amendment in the soil improvement area based on the type of soil parameters to be improved, and determining the amount of amendment to be applied based on the predicted soil parameter changes; Soil improvement measures for the target planting site are determined based on the application location, type and amount of the soil conditioner.

4. The soil improvement measure decision-making method based on multimodal deep learning according to claim 1, characterized in that: The multimodal deep learning network is constructed, and a correlation analysis is performed on the climate change characteristics and soil conditioner content change characteristics of the target planting site after the implementation of soil improvement measures within a preset time period based on the multimodal deep learning network to determine the impact data of climate change on soil conditioners, specifically: Construct a multimodal deep learning network based on the XGBoost gradient boosting tree model, recurrent neural network, and temporal graph neural network; Implement the soil improvement measures and obtain time series data on climate change characteristics and soil conditioner content change characteristics of the target planting site within a preset time period after the soil improvement measures are implemented, wherein the climate change characteristics include changes in temperature, precipitation, wind speed, and air humidity; and the soil conditioner content change characteristics include concentration change values ​​of each soil conditioner type in each sub-region of the target planting area; Performing a time series alignment operation on the climate change characteristic time series data and the soil conditioner content change characteristic time series data, and importing the climate change characteristic as the input feature vector and the soil conditioner content change characteristic as the model output label into the XGBoost gradient boosting tree model for training; All climate variables are ranked in importance based on the feature gain scoring mechanism built into the XGBoost gradient boosting tree model, and the top climate variables with the most significant impact on the change in soil amendment concentration are identified as key climate factors based on their cumulative gain contribution during the training process. Inputting the time series of the key climate factors into a recurrent neural network model including a gating mechanism to perform time series modeling training, and extracting the time series change characteristics of the key climate factors; A graph structure model is established based on the diffusion behavior of soil amendments between each sub-region, with each sub-region as a graph node and the propagation probability, diffusion time lag, and concentration gradient of the amendment between adjacent regions as graph edge attributes. A graph adjacency matrix and a time-step transfer matrix are constructed based on a temporal graph neural network model, and the temporal graph neural network is trained to extract the spatial diffusion characteristics of the soil amendments. A correlation analysis is performed on the time series variation characteristics of the key climate factors and the spatial diffusion characteristics of the soil conditioner to determine the impact of climate change on the soil conditioner and obtain impact data.

5. The soil improvement measure decision-making method based on multimodal deep learning according to claim 1, characterized in that: The failure characteristics of the modifier in the target planting site are determined based on the impact data, and the failure characteristics include the failure propagation path characteristics and the failure dose variation characteristics of the modifier, specifically: Obtaining historical climate change characteristic data for a target planting area, importing the historical climate change characteristic data into a climate prediction model based on a long short-term memory network for training based on the historical climate change characteristic data, and predicting climate change for a preset implementation period after soil improvement measures are implemented in the target planting area to obtain climate change prediction data; Determine, based on the impact data and climate change prediction data, a time series of concentration changes of the soil amendment in each sub-region after the implementation of the soil amendment measures, and determine a failure time node at which the concentration falls below a preset threshold; Connecting each sub-region in order of failure time nodes to determine the failure propagation direction of the soil conditioner, and determining the failure propagation path characteristics of the soil conditioner based on the failure propagation direction; The modifier concentration change rate of each sub-region is calculated according to the concentration change time series, and the modifier failure dose change characteristics are determined based on the concentration change rate.

6. The soil improvement measure decision-making method based on multimodal deep learning according to claim 1, characterized in that: The compensation operation for the soil improvement measures is performed according to the failure characteristics of the improver, and a compensation plan is constructed, specifically: Determine a sub-region set of the soil conditioner on the failure propagation path according to the characteristics of the conditioner failure path, mark the sub-region to be compensated, and determine the conditioner compensation priority of each sub-region to be compensated according to the conditioner failure time node in the sub-region to be compensated; Determining the initial compensation application amount of the soil conditioner for each sub-area to be compensated according to the failure dose variation characteristics; Comparing the initial compensation application amount with a preset safety application amount threshold, if the initial compensation application amount is greater than the safety application amount threshold, dividing the initial compensation application amount into multiple stage application amounts based on the failure propagation path characteristics, and determining the application time of each stage application amount according to the climate change mutation time node in the climate change prediction data to obtain an application time set; If the initial compensation application amount is not greater than the safe application amount threshold, a soil conditioner compensation application operation is performed on the sub-area to be compensated according to the initial compensation application amount, and the time point at which the soil conditioner concentration change rate is minimum after the expiration time node is determined based on the climate change prediction data and impact data, and the time point at which the concentration change rate is minimum is used as the soil conditioner compensation application time point; Compensation operations are performed on soil improvement measures according to the amendment compensation priority, stage application amount or initial compensation application amount, application time set or compensation application time point to construct a compensation plan.

7. A soil improvement measure decision-making system based on multimodal deep learning, characterized in that: The soil improvement measure decision-making system based on multimodal deep learning includes a storage device and a processor. The storage device includes a soil improvement measure decision-making method program based on multimodal deep learning. When the soil improvement measure decision-making method program based on multimodal deep learning is executed by the processor, the following steps are implemented: Acquiring soil state parameter change data and crop growth data of the target planting site, and determining soil improvement measures for the target planting site based on the soil state parameters and crop growth data; A multimodal deep learning network was constructed to conduct a correlation analysis between the climate change characteristics and the changes in soil amendment content at the target planting site after the implementation of soil amendment measures within a preset time period, thereby determining the impact of climate change on soil amendments. Determining, based on the impact data, the failure characteristics of the improver at the target planting site, wherein the failure characteristics include the failure propagation path characteristics and the failure dose variation characteristics of the improver; Compensation operations are performed on soil improvement measures based on the failure characteristics of the amendments, and a compensation plan is constructed.

Images