Intelligent agricultural condition monitoring system and method for grain and oil crops

Abstract

The invention discloses an intelligent agricultural condition monitoring system and method for grain and oil crops, and relates to the technical field of agricultural condition monitoring. Farmland multi-source data are acquired and preprocessed in real time through a multi-source data acquisition module; multi-factor step-by-step fusion of farmland multi-source data is realized by using a D-S evidence theory through a multi-source data fusion module, and a multi-factor fusion evaluation index is formed; constructing a grain and oil crop environment-growth state comprehensive model based on LSTM through an environment state fusion module, and fusing the grain and oil crop environment-growth state comprehensive model with farmland multi-source data and multi-factor fusion evaluation indexes to obtain a farmland digital twin system; and a growth deviation index is analyzed and generated through the decision feedback optimization module and is fed back to the optimization system. The problems that in the prior art, data acquisition hysteresis is remarkable, multi-source heterogeneous information is isolated and difficult to fuse, and early warning fails due to lack of correlation between environment mutation and crop response are solved, and accurate monitoring and risk active intervention of grain and oil crops in the whole growth period are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of crop monitoring, in particular to a smart crop monitoring system and method for grain and oil crops. BACKGROUND

[0002] In the production process of grain and oil crops, real-time and accurate crop monitoring is the key to optimizing water and fertilizer management, pest control and disaster response. Although the current mainstream crop monitoring technology has introduced Internet of Things sensors, unmanned aerial vehicles and other equipment, there are still significant bottlenecks: traditional data collection methods rely on manual regular patrols, which cannot balance high-frequency monitoring and large-scale full-coverage scenarios, resulting in missing data during the key growth period of grain and oil crops, and delaying decision-making due to the time lag between environmental changes and crop responses; the multi-source heterogeneous data collected in traditional technology is isolated and scattered, lacking effective fusion analysis mechanisms, making it difficult to establish a dynamic correlation model between environmental factors and crop growth, resulting in pest warning, drought risk assessment and other major relying on manual experience, lacking scientificity; the existing technology applies digital twin technology to grain and oil crop monitoring, but it has poor adaptability in field scenarios, most digital twin systems only realize static environment visualization, and it is difficult to build a growth deduction model combined with the growth period of grain and oil crops, lacking real-time simulation and intervention simulation of sudden risks, resulting in lagging management recommendations and limited accuracy. SUMMARY

[0003] The present application provides a smart crop monitoring system and method for grain and oil crops, which solves the problems of data collection lag, multi-source heterogeneous information isolation, and lack of correlation between environmental changes and crop responses, resulting in ineffective early warning, by real-time collection and fusion of Internet of Things high-frequency multi-source data, dynamic environment-growth coupling modeling based on D-S evidence theory and LSTM, and digital twin system, and realizes accurate monitoring and risk active intervention of grain and oil crops throughout the growth period.

[0004] To achieve the above purpose, the present application provides a smart crop monitoring system for grain and oil crops, comprising:

[0005] A multi-source data acquisition module is used to acquire multi-source data of farmland, pre-process the multi-source data of farmland, and obtain pre-processed data;

[0006] A multi-source data fusion module is used to perform multi-factor distributed step-by-step fusion on the pre-processed data according to the D-S evidence theory, and obtain multi-factor fusion evaluation indexes;

[0007] An environmental state fusion module is used to capture time sequence dependency through a long short-term memory network LSTM based on the pre-processed data and the multi-factor fusion evaluation indexes, and establish a grain and oil crop environment-growth state comprehensive model;

[0008] The digital twin establishment module is used for fusing the preprocessed data and a grain and oil crop environment-growth state comprehensive model to construct a farmland digital twin system.

[0009] The decision feedback optimization module is used for performing real-time analysis on the grain and oil crops by using the farmland digital twin system to obtain a grain and oil crop growth deviation index, and performing feedback optimization on the farmland digital twin system according to the grain and oil crop growth deviation index.

[0010] The existing crop condition monitoring technology is difficult to establish a dynamic correlation model of environmental factors and crop growth due to isolated and scattered multi-source heterogeneous data, resulting in that disaster warning relies on artificial experience and lacks scientificity. Meanwhile, there is significant time lag between environmental mutation and crop response, and traditional static models cannot capture the dynamic coupling relationship of multi-source time series data, resulting in that the key growth period monitoring blind area is expanded and decision delay is caused. In addition, the existing digital twin system is divorced from the deduction of crop growth period rules, only realizes static environment visualization, lacks real-time simulation and intervention simulation capability for sudden risks, and causes that the management strategy is seriously out of touch with the actual response of farmland.

[0011] The present application solves the problem of evidence conflict caused by isolated and scattered multi-source heterogeneous data by using a multi-source data fusion module to construct three types of evidence bodies of soil moisture, weather and pests and diseases based on D-S evidence theory, and using a conflict factor dynamic weighting mechanism driven by historical regression coefficients and real-time conflict factors to coordinate multi-source contradictory data. The anti-interference multi-factor fusion evaluation index is output, and the reliability of pest and drought warning is significantly improved. The environment state fusion module captures the long and short term time series dependence relationship between environment and growth by means of the time series dependence modeling capability of long and short term memory network, realizes the incremental learning optimization of LSTM parameters through the momentum stochastic gradient descent mechanism, eliminates the time lag effect between environmental mutation and crop response, realizes the rapid response of sudden disasters such as rainstorm and hail, eliminates the monitoring blind area of key growth period of grain and oil crops, and reduces the occurrence of warning failure caused by missing correlation between environmental mutation and crop response. The decision feedback module triggers the closed loop self-calibration of the digital twin system through the growth deviation index, dynamically corrects the evidence weight and the LSTM network parameter, predicts the effect of irrigation and pesticide strategy, and reduces the resource mismatch rate.

[0012] Further, the farmland multi-source data includes crop seedling condition, pest and disease situation, soil moisture, environmental temperature, air humidity, wind speed, rainfall and light intensity.

[0013] The preprocessing of the farmland multi-source data includes cleaning, denoising, standardization processing, missing value processing, abnormal value processing, conditional model filling, data standardization and feature selection.

[0014] The application can reflect crop seedling conditions, soil moisture conditions, pest conditions and weather conditions by crop seedling conditions, pest conditions, soil moisture conditions, environmental temperature, air humidity, wind speed, rainfall and light intensity, so as to simulate the dynamic changes of farmland in the real world; the application can effectively filter out distorted signals caused by device drift and environmental mutation by performing cleaning and denoising, outlier removal and multi-modal standardization processing on farmland multi-source data, while retaining key crop conditions such as soil moisture and canopy coverage, a standardized data set with unified dimensions is constructed, providing high-consistency input data for subsequent multi-factor fusion and LSTM modeling, ensuring the accuracy of the digital twin system from the source, and solving the problems of noise interference, dimension confusion and key feature submersion in traditional farmland monitoring due to environmental interference, device abnormalities and differences in collection frequency of multi-source sensor data.

[0015] Further, the multi-source data fusion module comprises:

[0016] The data input unit is configured to divide the preprocessed data into a plurality of independent evidence sources, and construct a basic probability assignment function of each independent evidence source according to the plurality of independent evidence sources.

[0017] The evidence synthesis unit is configured to calculate a conflict factor of each independent evidence source according to the basic probability assignment function of each independent evidence source, and perform an evidence dynamic weight compensation mechanism by introducing historical data according to the conflict factor of each independent evidence source to obtain a joint confidence.

[0018] The decision output unit is configured to obtain a multi-factor fusion evaluation index according to the joint confidence.

[0019] In view of the problem that in the traditional method, after multi-source data is acquired, it is difficult to coordinate the contradictions between the multi-source data, and it is difficult to establish a dynamic correlation model between environmental factors and crop growth, the application is established based on D-S evidence theory, a dynamic weight compensation mechanism is designed, the weight of the evidence source is dynamically adjusted according to the historical regression coefficient, the contradictions between the multi-source data are coordinated, and the model inaccuracy problem caused by multi-source data fragmentation and evidence conflict is solved. The long-term reliability of the evidence source is learned through the historical regression coefficient, the limitations of the traditional static weighted average are solved, the multi-source data are deeply associated, the accuracy of farmland crop condition early warning is improved, meanwhile, by suppressing contradictory evidence interference, the strategy making error caused by invalid data is reduced, the agricultural resource allocation is optimized, and reliable input data is provided for the digital twin system.

[0020] Further, the expression of the multi-factor fusion evaluation index is as follows:

[0021]

[0022] wherein, is the multi-factor fusion evaluation index. is a conflict coefficient decay factor, is a global conflict coefficient, is a state value mapping, is a pignistic probability, is a grain and oil crop growth state, is a historical data interpolation value, is a joint confidence, is a target hypothesis subset to be evaluated, is the number of independent evidence sources, and are indexes of independent evidence sources, is a dynamic weight of introducing historical data, is an actual focal element selection of the th independent evidence source, is a weighted evidence chain confidence, is a confidence distribution value of the th independent evidence source to , is a grain and oil crop state set, is an empty set.

[0023] In view of the problem that the traditional D-S synthesis rule is invalid due to the contradiction and conflict between evidences in the multi-source evidence fusion process, the existing method is difficult to output a reasonable evaluation result when the global conflict coefficient tends to 1, and the fixed interpolation strategy cannot adapt to the dynamic change of the farmland environment, resulting in distortion of the crop growth state evaluation. The present application converts the evidence focal element into a decision-making hypothesis by using the pignistic probability, and dynamically balances the contribution weight of the historical data interpolation value and the current joint confidence through the conflict coefficient decay factor, which can still output stable evaluation indicators in a strong conflict scene, provides anti-interference input for the environment-growth state comprehensive model, and improves the early identification reliability of sudden risks such as drought and pest.

[0024] Further, the expression of the dynamic weight of introducing historical data is as follows:

[0025]

[0026] wherein, is a dynamic weight of introducing historical data, is a historical regression correction coefficient, and are regression parameters (obtained by training historical data), is a natural exponential function, is a conflict sensitive coefficient, is a conflict factor of the th independent evidence source, is a conflict factor of the a historical feature vector of the individual evidence source, a number of focal elements of the individual evidence source, a number of focal elements of the individual evidence source, a number of focal elements of the individual evidence source, a number of focal elements of the individual evidence source, a confidence of the individual evidence source to the subset, a confidence of the individual evidence source to the subset, a confidence of the individual evidence source to the subset, a confidence of the individual evidence source to the subset, a confidence of the individual evidence source to the subset, a confidence of the individual evidence source to the subset, a normalization function, a number of sensors of the individual evidence source, a number of sensors of the individual evidence source, an influence coefficient of the individual evidence source to the sensor, an influence coefficient of the individual evidence source to the sensor, a detection value of the individual evidence source to the sensor, a detection value of the individual evidence source to the sensor, a detection value of the individual evidence source to the sensor, a detection value of the individual evidence source to the sensor, a detection value of the individual evidence source to the sensor, a detection value of the individual evidence source to the sensor, a detection value of the individual evidence source to the sensor,

[0027] In view of the fact that in evidence fusion, due to environmental mutation or sensor abnormality, the reliability of a single evidence source suddenly decreases, the traditional fixed weight allocation mechanism is difficult to dynamically capture the confidence change of the evidence body, and is prone to cause the fusion result to deviate from the real state of grain and oil crops; the present application establishes a conflict-sensitive dynamic weight function, reduces the fusion deviation caused by the absence of dynamic evaluation of the reliability of the evidence source; wherein the historical regression coefficient inherits the long-term data law, the conflict factor quantifies the evidence contradiction degree in real time, and the two data cooperatively drive the adaptive adjustment of the weight, so that the contribution degree of the evidence body is automatically corrected with the environmental mutation, abnormal evidence interference is inhibited, the effective information fusion capability is retained, and the robustness of the multi-factor evaluation index in a complex farmland scene is enhanced.

[0028] Further, the environment state fusion module comprises:

[0029] a data integration unit, configured to align and splice the preprocessed environment data and the multi-factor fusion evaluation index according to a time sequence to generate a time sequence input sample;

[0030] an LSTM modeling unit, configured to train a long short-term memory network (LSTM) based on cross-entropy loss and L2 regularization according to the time sequence input sample, to obtain an environment growth state mapping model;

[0031] The update and optimization unit is used to update the network parameters of the environmental growth state mapping model based on the prediction error of the model and by using stochastic gradient descent with momentum, so as to obtain a comprehensive environmental-growth state model for grain and oil crops.

[0032] To address the problem that traditional models struggle to capture the dynamic coupling relationship between multi-source time-series data due to the time lag effect between farmland environmental factors and crop growth responses, and the weak correlation between environmental abrupt changes and crop state changes, this invention establishes a long- and short-term time-series relationship between environmental changes and crop responses through time-series alignment and recurrent short-term memory (LSTM) neural network modeling. This reduces response time lag and solves the prediction inaccuracy problem caused by the lack of dynamic correlation between environment and growth. Environmental data and multi-factor fusion indicators are concatenated in time series as time-series input samples. The time-series dependency modeling capability of the LSTM network is used to capture the lag response pattern. A comprehensive environment-growth state model is trained using cross-entropy loss and regularization constraints. Through dynamic coupling modeling with LSTM, the time lag effect is captured, realizing the data correlation between environmental changes and time changes. This enhances the early prediction capability of environmental factors with progressive risks such as drought stress and pest and disease spread, providing digital twin systems with dynamic prediction capabilities with time extrapolation.

[0033] Furthermore, the expression for the integrated model of environment and growth status of grain and oil crops is as follows:

[0034]

[0035]

[0036] in, For prediction using a comprehensive environmental-growth status model for grain and oil crops. The growth status of grain and oil crops at all times. In order to be in The output of the optimal environmental growth state mapping model is selected in each iteration. For normalized exponential functions, The parameters are those of the optimal environmental growth state mapping model. The parameters of the environmental growth state mapping model are... The latest time series to be predicted at the current moment. For the number of iterations, For time indexing, The length of the time sliding window. and These are all parameters of the environmental growth state mapping model. For the first The parameters of the environment growth state mapping model in the next iteration For the first The parameters of the environment growth state mapping model in the next iteration For learning rate, For the first The update speed of each iteration The momentum coefficient, For the first Loss function at the next iteration right gradient, For gradient operators, For parameter-dependent loss function, The total number of samples for the time series input. The sample number. It is a logarithmic function. For the first The predicted values ​​obtained from the environmental growth state mapping model for each time-series input sample. For the first The actual values ​​of each time series input sample To The class probability distribution vector obtained after performing softmax. To extract the category probability distribution vector The corresponding probability value, The L2 regularization coefficient is... For parameters The square of the L2 norm, For the environmental growth state mapping model in the first The last time step of the sample The output hidden state vector, This is the output layer weight matrix of the environmental growth state mapping model. This is the output layer bias vector of the environmental growth state mapping model. For parameters Environmental growth state mapping model For the first The complete time series matrix of the training samples, for The one-dimensional feature vector of the time-space environment growth state mapping model. for The environmental data vector at any given time. for Multi-factor fusion evaluation index at any given time.

[0037] In view of the fact that the existing growth prediction model has a lag in parameter adjustment when the environment mutates, so that the traditional batch training mechanism cannot real-time integrate the latest farmland data, and the network gradient oscillation leads to slow convergence, which cannot support high-frequency monitoring data, the present application constructs a random gradient descent mechanism with a momentum factor, after receiving the latest farmland data, a micro-batch sample is constructed, and the gradient vibration is smoothed by using the momentum in the random gradient descent, the network complexity is constrained by combining L2 regularization, the LSTM parameters are updated with the real-time monitoring data increment, the high-frequency monitoring of agricultural data is realized, the minute-level response capability of sudden disasters such as hail and rainstorm is enhanced, and meanwhile, the continuity of the growth trend deduction is ensured by the cross-time step transmission of the hidden state vector.

[0038] Further, the digital twin establishment module comprises:

[0039] The virtual-real synchronization unit is used for synchronously mapping the preprocessed data and the output result of the grain and oil crop environment-growth state comprehensive model based on the time stamp, and constructing a virtual twin corresponding to the real farmland.

[0040] The real-time deduction unit is used for taking the virtual twin as an initial state, predicting the growth state of the grain and oil crop in a set future period by using the grain and oil crop environment-growth state comprehensive model, and obtaining a growth state prediction result in a future multi-granularity time.

[0041] The intervention simulation unit is used for simulating in the virtual twin according to the user-inputted control strategy set and the growth state prediction result in the future multi-granularity time, and obtaining the farmland digital twin system.

[0042] In view of the fact that the existing digital twin system lacks the quantification simulation capability of the control strategy, so that the management decision and the actual response of the crop are disconnected, and it is difficult to dynamically deduce the growth trend of the crop, the present application constructs a virtual-real synchronous deduction architecture, synchronously maps the real-time data and the prediction model into a virtual twin based on the time stamp, ensures that the virtual twin corresponds to the real farmland frame by frame, deduces the evolution path of the crop in the future multi-period by using the growth state comprehensive model, makes the influence of the environment on the growth state of the grain and oil crop more accurate, improves the fitting of the management decision making and the actual grain and oil crop, supports the user to input the control strategy set for intervention simulation, realizes the virtual verification and effect prediction of the management decision such as irrigation and pesticide application, completes the deduction of the future growth trend of the grain and oil crop, and improves the active intervention capability of the digital twin system.

[0043] Further, the expression of the grain and oil crop growth deviation index is as follows:

[0044]

[0045] Wherein, is the grain and oil crop growth deviation index, is a growth state index, is an index of the growth state index, is is a measured value of the growth state index at the moment is a measured value of the growth state index at the moment is a predicted value of the growth state index at the moment is a predicted value of the growth state index at the moment is a predicted value of the growth state index at the moment is a historical maximum value of the growth state index, is a historical maximum value of the growth state index, is a historical minimum value of the growth state index, is a historical minimum value of the growth state index.

[0046] In order to solve the problems that the traditional static model cannot dynamically capture the abnormal response of grain and oil crops to environmental mutations, the prediction result of the digital twin system deviates from the actual farmland state, and there is a lack of closed-loop correction mechanism, resulting in mismatch between the model prediction result and the real farmland state and invalidation of the finally formulated decision, the present application introduces a crop growth deviation index to quantify the prediction deviation degree, improves the ability of the environment mutation and the abnormal response of the grain and oil crops, generates a double path based on a threshold triggering strategy: a new farmland management decision scheme is generated by combining historical management decisions and user historical operations; when the grain and oil crop growth deviation index is greater than a preset threshold, a closed-loop correction mechanism is constructed by updating the evidence weight and the LSTM network parameter, the continuous calibration of the digital twin system and the physical farmland is realized, the intervention accuracy of sudden environmental factors such as hail and pest is enhanced, and the accuracy of the farmland digital twin system and the reliability of the finally formulated strategy are improved.

[0047] The present application also provides a smart farmland monitoring method for grain and oil crops, comprising:

[0048] Obtaining farmland multi-source data through multi-source sensors, preprocessing the farmland multi-source data to obtain preprocessed data;

[0049] According to the D-S evidence theory, the preprocessed data is subjected to multi-factor distributed step-by-step fusion to obtain a multi-factor fusion evaluation index;

[0050] According to the preprocessed data and the multi-factor fusion evaluation index, a long short-term memory network LSTM is used to capture the time sequence dependence relationship, and a grain and oil crop environment-growth state comprehensive model is trained;

[0051] The preprocessed data and the grain and oil crop environment-growth state comprehensive model are fused to construct a farmland digital twin system;

[0052] The farmland digital twin system is used to analyze the grain and oil crops in real time to obtain a grain and oil crop growth deviation index, and the farmland digital twin system is feedback optimized according to the grain and oil crop growth deviation index.

[0053] The present application performs distributed step-by-step fusion on multi-source data by D-S evidence theory, coordinates the conflicts between soil moisture, weather and pest evidence bodies, and generates multi-factor evaluation indexes resistant to interference; combines LSTM network to capture the environment-growth time sequence dependence law, and constructs a comprehensive model with dynamic deduction capability; finally, the growth deviation index drives the digital twin system to close-loop self-optimization, realizes virtual verification and minute-level parameter adjustment of irrigation and pesticide strategies; solves the problem that traditional crop monitoring methods cannot establish a dynamic correlation model between environmental factors and crop growth due to isolated and scattered multi-source heterogeneous data, leading to disaster warning relying on artificial experience and response lag; can quantify the time lag effect between environmental mutation and crop physiological response, reduce the monitoring blind area of grain and oil crops in the key growth period, and at the same time, the present application can reduce the misjudgment of the disconnection between growth state and strategy making, and improve the resource allocation rate of grain and oil crops.

[0054] The present application provides a kind of intelligent crop monitoring system and method for grain and oil crops, at least with following technical effects or advantages:

[0055] The present application solves the problem of isolated and scattered multi-source heterogeneous data in traditional crop monitoring through multi-source data dynamic fusion mechanism; based on D-S evidence theory, independent evidence bodies of soil moisture, weather and pests are divided, and conflict factor dynamic weighting strategy is designed to coordinate contradictory data, improve the correlation reliability between environmental factors and crop growth; with the aid of long short-term memory network, the time sequence coupling law of environmental mutation and crop growth is captured, and the response time lag between environmental factors and growth state is reduced; through sliding window modeling and incremental learning optimization, rapid response to sudden risks such as heavy rain and pest is realized, and the monitoring blind area of grain and oil crops in the key growth period is fully covered, and the timeliness of disaster prevention and control is improved; at the same time, based on the feedback optimization mechanism triggered by growth deviation index, the digital twin system realizes dynamic self-calibration, reduces the risk of resource mismatch, and provides scientific and adaptive decision making for the production of grain and oil crops. BRIEF DESCRIPTION OF DRAWINGS

[0056] The drawings described herein are used to provide further understanding of the embodiments of the present application, constitute a part of the present application, and do not constitute a limitation on the embodiments of the present application;

[0057] Figure 1 is a structural schematic diagram of a kind of intelligent crop monitoring system for grain and oil crops in the present application. DETAILED DESCRIPTION

[0058] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application is further described in detail below in combination with the drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0059] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. However, it will be apparent to one skilled in the art that the present application can be practiced without the specific details set forth in this description, that the present application can be practiced with other systems, components, materials, etc. In other instances, well-known structures have not been described in detail in order to avoid obscuring the present application.

[0060] Embodiment one

[0061] As shown in the figure, the present application provides a grain and oil crop intelligent agricultural condition monitoring system, comprising: Figure 1 A multi-source data acquisition module is configured to acquire multi-source data of farmland, pre-process the multi-source data of farmland, and obtain pre-processed data.

[0062] A multi-source data fusion module is configured to perform multi-factor distributed step-by-step fusion on the pre-processed data according to D-S evidence theory, and obtain a multi-factor fusion evaluation index.

[0063] An environmental state fusion module is configured to capture time sequence dependency through a long short-term memory network (LSTM) based on the pre-processed data and the multi-factor fusion evaluation index, and establish a grain and oil crop environment-growth state comprehensive model.

[0064] A digital twin establishment module is configured to fuse the pre-processed data and the grain and oil crop environment-growth state comprehensive model, and construct a farmland digital twin system.

[0065] A decision feedback optimization module is configured to perform real-time analysis on the grain and oil crops using the farmland digital twin system, obtain a grain and oil crop growth deviation index, and perform feedback optimization on the farmland digital twin system based on the grain and oil crop growth deviation index.

[0066]

[0067] ​In specific embodiments of the application, the multi-source data acquisition module acquires real-time and historical farmland multi-source data, including soil moisture of grain and oil crop planting land acquired through a sensor network in the soil, humidity, temperature, wind speed, light intensity and precipitation of the grain and oil crop planting land acquired through a weather station, and pest density and crop seedling condition acquired through unmanned aerial vehicle images based on existing visual recognition neural network models such as YOLO model; and finally obtains farmland multi-source data including crop seedling condition, pest and disease condition, soil moisture, environmental temperature, air humidity, wind speed, rainfall and light intensity; the acquired farmland multi-source data is preprocessed, including cleaning, denoising, standardization processing, missing value processing, outlier processing, conditional model filling, data standardization and feature selection; the cleaning includes removing invalid data caused by sensor offline period and unmanned aerial vehicle positioning drift; the denoising includes smoothing soil temperature and humidity time series fluctuations by using a Savitzky-Golay filter to eliminate peak noise caused by rainfall impact; the missing value processing includes using a forward filling method for short-time missing meteorological data and using multivariate linear regression compensation for long-time missing meteorological data; the outlier processing includes detecting and identifying soil salinity outliers by using Z-score, and correcting in combination with a spatial interpolation method; the conditional model filling includes generating a compensation value based on a historical and contemporaneous temperature and humidity-pest association model when pest image data is missing; the data standardization includes unifying heterogeneous physical quantities to the interval [0, 1]; and the feature selection includes reducing the dimension of original features by using principal component analysis (PCA), retaining principal components with a cumulative contribution rate > 85%, and generating a low-dimensional feature vector input to the multi-source data fusion module.

[0068] In specific embodiments of the application, the multi-source data fusion module includes:

[0069] a data input unit configured to divide the preprocessed data into a plurality of independent evidence sources, and construct a basic probability assignment function of each independent evidence source according to the plurality of independent evidence sources;

[0070] an evidence synthesis unit configured to calculate a conflict factor of each independent evidence source according to the basic probability assignment function of each independent evidence source, execute an evidence dynamic weight compensation mechanism introducing historical data according to the conflict factor of each independent evidence source, and obtain a joint confidence degree;

[0071] a decision output unit configured to obtain a multi-factor fusion evaluation index according to the joint confidence degree.

[0072] The independent evidence sources can be divided into three types of independent evidence sources, or more independent evidence sources can be divided according to the demand of research accuracy, including:

[0073] a soil sensor-based soil moisture evidence body containing soil humidity, soil salinity and pH value;

[0074] The meteorological evidence body based on the meteorological station includes environmental temperature, air humidity, wind speed, light intensity, and precipitation;

[0075] The disease and pest evidence body based on the unmanned aerial vehicle image includes disease spot area proportion, pest population density, and crop seedling condition;

[0076] The expression of the multi-factor fusion evaluation index is as follows:

[0077]

[0078] wherein, is a multi-factor fusion evaluation index, is a conflict coefficient attenuation factor, is a global conflict coefficient, is a state value mapping, is a pignistic probability, is a grain and oil crop growth state, is a historical data interpolation value, is a joint confidence, is a target hypothesis subset to be evaluated, is the number of independent evidence sources, and are indexes of independent evidence sources, is a dynamic weight of introducing historical data, is an actual focal element selection of the i-th independent evidence source, is a weighted evidence chain confidence, is a confidence distribution value of the i-th independent evidence source to the j-th target hypothesis, is a grain and oil crop state set, is an empty set. The expression of the dynamic weight of introducing historical data is as follows:

[0079]

[0080]

[0081] wherein, is a dynamic weight of introducing historical data, is a historical regression correction coefficient, and are regression parameters (trained by historical data), is a natural exponential function, is a conflict sensitive coefficient, is a conflict factor of the i-th independent evidence source, is a conflict factor of the i-th independent evidence source, is a conflict factor of the i-th independent evidence source, ​​​a historical feature vector of the individual evidence source, the number of focal elements of the first individual evidence source, the number of focal elements of the first individual evidence source, the confidence of the first individual evidence source to the subset , the confidence of the first individual evidence source to the subset , is a normalization function, the number of sensors of the first individual evidence source, the influence coefficient of the first individual evidence source to the first sensor, the detection value of the first individual evidence source to the first sensor, the accuracy weight of the first individual evidence source to the first sensor.

[0082] In an embodiment of the present application, suppose the soil condition evidence body: soil volume water content = 25.3%, soil salinity = 1.8 dS / m, pH = 6.7; the meteorological evidence body: air temperature = 28.5℃, light intensity = 85 klux, precipitation = 0 mm; the pest and disease evidence body: stripe rust spot ratio = 4.7%, aphid density = 12 per leaf; set the grain and oil crop state set ; in the expression , represents a single sensor data; , represents the current detected pest population density; , represents the sensor accuracy weight, which can be obtained by historical data calibration; , represents the artificially set influence coefficient; after softmax normalization, the basic probability assignment function of the pest and disease evidence body is as follows:

[0083]

[0084]

[0085] In the evidence synthesis unit, the meteorological evidence body is embodied as no precipitation + high temperature → support "severe stress"; the soil condition evidence body is embodied as sufficient water → support "healthy", and the meteorological evidence body and the soil condition evidence body are in conflict, so according to:

[0086] ​

[0087] Get:

[0088]

[0089] The introduction of historical regression coefficients , conflict sensitive coefficient , the dynamic weight of the introduction of historical data can be obtained by meteorological station historical data:

[0090]

[0091] Get the global conflict degree of soil and weather evidence:

[0092]

[0093] Pignistic probability conversion is carried out:

[0094]

[0095] In the decision output unit, according to the joint confidence, the multi-factor fusion evaluation index is obtained:

[0096]

[0097] Among them, the conflict coefficient decay factor , the historical data interpolation value , can be obtained from the historical data, State value mapping, which can be set by agricultural experts to health = 0.2, mild stress = 0.8, and severe stress = 0.5.

[0098] The present application is based on D-S evidence theory, D-S evidence theory can fuse multi-source uncertain information through basic probability assignment and synthesis rule, and quantify the confidence in conflict environment, on this basis, a dynamic weight compensation mechanism is designed, which solves the problem that in the traditional method, after obtaining multi-source data, it is difficult to coordinate the contradiction between multi-source data, and it is difficult to establish a dynamic correlation model between environmental factors and crop growth; Avoid the limitations of traditional static weighted average, make the multi-source data deeply related, improve the accuracy of farmland agricultural condition early warning, at the same time, through inhibiting the interference of contradictory evidence, reducing the strategy making error caused by invalid data, optimizing the allocation of agricultural resources, providing reliable input data for digital twin system.

[0099] In the specific embodiments of the present application, the environmental state fusion module comprises:

[0100] The data integration unit is used for aligning and splicing the preprocessed environmental data and the multi-factor fusion evaluation index according to time sequence to generate time sequence input sample;

[0101] The LSTM modeling unit is used to train the Long Short-Term Memory (LSTM) network based on cross-entropy loss and L2 regularization according to the temporal input samples to obtain the environment growth state mapping model.

[0102] The update and optimization unit is used to update the network parameters of the environmental growth state mapping model based on the prediction error of the model and by using stochastic gradient descent with momentum, so as to obtain a comprehensive environmental-growth state model for grain and oil crops.

[0103] The expression for the integrated model of environment and growth status of grain and oil crops is as follows:

[0104]

[0105] in, For prediction using a comprehensive environmental-growth status model for grain and oil crops. The growth status of grain and oil crops at all times. In order to be in The output of the optimal environmental growth state mapping model is selected in each iteration. For normalized exponential functions, The parameters are those of the optimal environmental growth state mapping model. The parameters of the environmental growth state mapping model are... The latest time series to be predicted at the current moment. For the number of iterations, For time indexing, The length of the time sliding window. and These are all parameters of the environmental growth state mapping model. For the first The parameters of the environment growth state mapping model in the next iteration For the first The parameters of the environment growth state mapping model in the next iteration For learning rate, For the first The update speed of each iteration The momentum coefficient, ranging from 0 to 1, represents the retention of historical training directions. A larger momentum coefficient means more historical training data is retained; a value of 0.9 is generally suitable. For the first Loss function at the next iteration right gradient, For gradient operators, For parameter-dependent loss function, The total number of samples for the time series input. The sample number. It is a logarithmic function. For the first The predicted values ​​obtained from the environmental growth state mapping model for each time-series input sample. For the first The actual values ​​of each time series input sample To The class probability distribution vector obtained after performing softmax. To extract the category probability distribution vector The corresponding probability value, The L2 regularization coefficient is... For parameters The square of the L2 norm, For the environmental growth state mapping model in the first The last time step of the sample The output hidden state vector, This is the output layer weight matrix of the environmental growth state mapping model. This is the output layer bias vector of the environmental growth state mapping model. For parameters Environmental growth state mapping model For the first The complete time series matrix of the training samples, for The one-dimensional feature vector of the time-space environment growth state mapping model. for The environmental data vector at any given time. for A multi-factor fusion evaluation index at time points; this invention constructs a stochastic gradient descent mechanism for the driving factors, namely... By combining L2 regularization to constrain network complexity, the LSTM parameters are updated incrementally with real-time monitoring data, enhancing the minute-level response capability to sudden disasters such as hail and rainstorms. Simultaneously, by passing feature vectors across time steps through hidden state vectors, the continuity of growth trend projection is ensured. This addresses the problem that existing growth prediction models suffer from lag in parameter adjustment during sudden environmental changes, making it difficult for traditional batch training mechanisms to integrate the latest farmland data in real time. Furthermore, network gradient oscillations lead to slow convergence, making it unsuitable for high-frequency monitoring data.

[0106] In one embodiment of the present invention, environmental data detected over seven consecutive days are fused with corresponding multi-factor fusion evaluation indicators for each day. For example, assuming detection is performed, the environmental data vector for the first day is... = =[Temperature 28.5℃, Humidity 65%, Illumination 85klux, Wind speed 3.2m / s], the multi-factor fusion evaluation index for the first day is 0.71, and the environmental data vector for the second day... = = [temperature 28.2℃, humidity 67%, illumination 82klux, wind speed 3.3m / s], the multi-factor fusion evaluation index of the second day is 0.70, the environmental data vector of the third day = = [temperature 27.8℃, humidity 69%, illumination 78klux, wind speed 3.5m / s], the multi-factor fusion evaluation index of the third day is 0.69, the environmental data vector of the fourth day = = [temperature 27.3℃, humidity 71%, illumination 72klux, wind speed 3.6m / s], the multi-factor fusion evaluation index of the fourth day is 0.685, the environmental data vector of the fifth day = = [temperature 26.9℃, humidity 73%, illumination 68klux, wind speed 3.65m / s], the multi-factor fusion evaluation index of the fifth day is 0.68, the environmental data vector of the sixth day = = [temperature 26.7℃, humidity 74%, illumination 66klux, wind speed 3.7m / s], the multi-factor fusion evaluation index of the sixth day is 0.68, the environmental data vector of the seventh day = = [temperature 26.5℃, humidity 75%, illumination 65klux, wind speed 3.7m / s], the multi-factor fusion evaluation index of the seventh day is 0.68, and

[0107]

[0108] The data integration unit of the application can solve the time lag problem of environmental changes and growth response of grain and oil crops, provide high information density input for LSTM, facilitate the judgment of grain and oil crop growth trend, and improve the early warning efficiency.

[0109] The LSTM modeling unit of the application trains the long short-term memory network LSTM based on cross-entropy loss and L2 regularization according to the time sequence input sample, and obtains an environmental growth state mapping model; wherein, the input layer of the long short-term memory network LSTM is a 5-dimensional feature, including 4-dimensional environment and 1-dimensional fusion index, the LSTM layer is 64 hidden units, using tanh activation, the output layer is Softmax classification, including healthy, mild stress and severe stress, and the expression of the environmental growth state mapping model is as follows:

[0110]

[0111] At this time, The parameters of the current LSTM network model, i.e., the parameters of the environment growth state mapping model at this time, are optimized by a loss function According to the prediction error of the environment growth state mapping model, the network parameters of the environment growth state mapping model are updated by using a stochastic gradient descent with momentum, and a grain and oil crop environment-growth state comprehensive model is obtained:

[0112]

[0113] Among them, , are specific parameters in the LSTM network model, are input gate, forget gate, output gate and candidate state input weight matrix, which can encode the influence weight of the current environmental factor on the state of grain and oil crops, are the cycle weight matrix of the input gate, the forget gate, the output gate and the candidate state, which can remember the evolution law of the historical growth state , are the bias vectors of the input gate, the forget gate, the output gate and the candidate state, which can adjust the basic physiological response threshold of grain and oil crops, is the weight matrix of the hidden state to the output layer, is the bias vector of the output layer; the present application punishes the deviation degree of the prediction result from the measured result by cross-entropy loss, and constrains the parameters by L2 regularization constraint, reduces the network complexity, prevents small sample overfitting, and enhances the early prediction ability of the grain and oil crop environment-growth state comprehensive model to the environmental factors of progressive risks such as drought stress and pest spread, providing a dynamic prediction capability with time extrapolation function for the digital twin system.

[0114] In specific embodiments of the present application, the digital twin establishment module is used to fuse the preprocessed data and the grain and oil crop environment-growth state comprehensive model to construct a farmland digital twin system; the three-dimensional morphological visualization method and the computer graphics realistic drawing technology can be used to draw the rod organ, individual and group model of crops, and based on the topological structure and spatial configuration law, the Unreal Engine 5 platform of Unreal Engine is used to realize the digital visualization of the crop growth environment and growth trend, describe the virtual scene of the entire crop growth in a digital way, and obtain the farmland digital twin system by combining the user input driving variable and the collision monitoring between crops, which can be used for growth display, process simulation and trend prediction in virtual digital space.

[0115] The digital twin establishment module comprises:

[0116] The virtual-real synchronization unit is used to synchronize and map the preprocessed data with the output results of the integrated model of environment-growth status of grain and oil crops based on timestamps, and to construct a virtual twin corresponding to the real farmland.

[0117] The real-time simulation unit is used to use the virtual twin as the initial state and the integrated model of grain and oil crops environment-growth status to predict the growth status of grain and oil crops in a set future time period, and obtain the growth status prediction results of future multi-granular time.

[0118] The intervention simulation unit is used to simulate in a virtual twin based on the set of control strategies input by the user and the prediction results of the growth status at multiple granular times in the future, so as to obtain a digital twin system for farmland.

[0119] This invention constructs a three-layered farmland digital twin system through a digital twin building module. Specifically, it includes: a timestamp-based virtual-real synchronization mechanism that maps soil moisture, meteorological data, and growth model outputs to a virtual twin at the millisecond level; a comprehensive environment-growth state model that predicts crop evolution paths over multiple time periods; and the ability to receive user-inputted control strategy sets and predict the effects of these strategies. This invention forms a closed-loop process from strategy input to virtual simulation results and then to feedback optimization, solving the problem that existing digital twin systems lack the ability to quantitatively simulate control strategies, leading to a disconnect between management decisions and actual crop responses, and making it difficult to dynamically predict crop growth trends.

[0120] The decision feedback optimization module of this invention can use a farmland digital twin system to perform real-time analysis of grain and oil crops, obtain a grain and oil crop growth deviation index, and perform feedback optimization on the farmland digital twin system based on the grain and oil crop growth deviation index.

[0121] The expression for the growth deviation index of grain and oil crops is as follows:

[0122]

[0123] in, This is the growth deviation index for grain and oil crops. This represents the total number of growth status indicators. An index for growth status indicators. for Time of the first Measured values ​​of the growth status index, for Time of the first Predicted values ​​of growth status indicators, For the first The historical maximum value of the growth status index For the first The historical minimum values ​​of the growth status indicators were used to calculate the growth deviation index of grain and oil crops. Determine the growth deviation index of grain and oil crops Whether it exceeds the preset threshold for the growth deviation index of grain and oil crops (which can be 0.15), if so, the farmland digital twin system is optimized, including updating the network parameters of the environmental growth state mapping model and correcting the weight parameters of independent evidence sources. The grain and oil crop growth deviation index of this invention enables the digital twin system to have the function of continuous calibration, improves the response speed to sudden disasters and the accuracy of resource allocation, realizes real-time and accurate monitoring of grain and oil crop production, and improves the scientificity and reliability of grain and oil crop treatment decisions.

[0124] Example 2

[0125] Based on Example 1, the present invention also provides a smart agricultural condition monitoring method for grain and oil crops, comprising:

[0126] Multi-source data of farmland is acquired through multi-source sensors, and the multi-source data of farmland is preprocessed to obtain preprocessed data;

[0127] Based on the DS evidence theory, the preprocessed data is fused in a multi-factor distributed hierarchical manner to obtain a multi-factor fusion evaluation index.

[0128] Based on the preprocessed data and multi-factor fusion evaluation indicators, a comprehensive model of the environment-growth status of grain and oil crops was trained by capturing temporal dependencies through a long short-term memory network (LSTM).

[0129] By integrating the preprocessed data with the integrated model of the environment and growth status of grain and oil crops, a digital twin system for farmland is constructed.

[0130] A farmland digital twin system is used to analyze grain and oil crops in real time to obtain a growth deviation index for grain and oil crops. The farmland digital twin system is then optimized based on the growth deviation index of grain and oil crops.

[0131] This method addresses the challenges of traditional monitoring methods, such as data lag, isolated multi-source information, and disconnect between environmental abrupt changes and crop responses. It constructs a closed-loop process encompassing data acquisition, fusion, modeling, digital twinning, and optimization. First, it uses IoT multi-source sensors to acquire and preprocess multi-dimensional farmland information in real time. Then, based on the DS evidence theory, it progressively fuses conflicting evidence to form robust multi-factor evaluation indicators. Subsequently, it uses LSTM to capture the coupling between environment and growth time series, establishing a dynamic comprehensive model, and finally generating a farmland digital twin system capable of real-time simulation. Through continuous feedback from the growth deviation index, the model can be quickly calibrated and the effects of regulation can be predicted, significantly shortening the disaster response cycle and improving the reliability of early warnings. Simultaneously, it enables virtual verification and closed-loop optimization of management strategies such as irrigation, fertilization, and pesticide application, achieving precise decision-making, resource efficiency, and risk control in monitoring the entire growth cycle of grain and oil crops.

[0132] While the preferred embodiments of the application have been described, additional variations and modifications can be made to these embodiments by those skilled in the art once they have the benefit of the present disclosure without departing from the spirit and scope of the application. Accordingly, it is intended that the appended claims include all such modifications and variations as fall within the scope of the present application.

[0133] It is apparent that those skilled in the art can make various changes and modifications to the application without departing from the spirit and scope of the application. It is therefore intended that the present application cover all such changes and modifications that are within its scope.

Claims

1. A grain and oil crop intelligent farm condition monitoring system, characterized in that, The method comprises the following steps: A multi-source data acquisition module is used to acquire multi-source data of farmland, pre-process the multi-source data of farmland, and obtain pre-processed data; A multi-source data fusion module is used to perform multi-factor distributed step-by-step fusion on the pre-processed data according to the D-S evidence theory, and obtain a multi-factor fusion evaluation index; An environmental state fusion module is used to capture time sequence dependency through a long short-term memory network (LSTM) based on the pre-processed data and the multi-factor fusion evaluation index, and establish a grain and oil crop environment-growth state comprehensive model; A digital twin establishment module is used to fuse the pre-processed data and the grain and oil crop environment-growth state comprehensive model, and construct a farmland digital twin system; A decision feedback optimization module is used to perform real-time analysis on the grain and oil crops by using the farmland digital twin system, obtain a grain and oil crop growth deviation index, and perform feedback optimization on the farmland digital twin system according to the grain and oil crop growth deviation index. 2.The grain and oil crop intelligent farm condition monitoring system according to claim 1, characterized in that, The multi-source data of farmland includes crop seedling conditions, pest and disease conditions, soil moisture conditions, environmental temperatures, air humidities, wind speeds, rainfall amounts, and light intensities. The pre-processing of the multi-source data of farmland includes cleaning, denoising, standardization processing, missing value processing, abnormal value processing, conditional model filling, data standardization, and feature selection. 3.The grain and oil crop intelligent farm condition monitoring system according to claim 1, characterized in that, The multi-source data fusion module comprises: A data input unit is used to divide the pre-processed data into a plurality of independent evidence sources, and construct a basic probability assignment function of each independent evidence source based on the plurality of independent evidence sources; An evidence synthesis unit is used to calculate a conflict factor of each independent evidence source based on the basic probability assignment function of each independent evidence source, perform an evidence dynamic weight compensation mechanism by introducing historical data based on the conflict factor of each independent evidence source, and obtain a joint confidence degree; A decision output unit is used to obtain a multi-factor fusion evaluation index based on the joint confidence degree.

4. The intelligent crop condition monitoring system for grain and oil crops of claim 3, characterized in that, The expression of the multi-factor fusion evaluation index is as follows: ； wherein, is a multi-factor fusion evaluation index, is a conflict coefficient decay factor, is a global conflict coefficient, is a state value mapping, is a pignistic probability, is a grain and oil crop growth state, is a historical data interpolation value, is a joint confidence, is a target hypothesis subset to be evaluated, is the number of independent evidence sources, and are indexes of independent evidence sources, is a dynamic weight of introducing historical data, is an actual focal element selection of the th independent evidence source, is a weighted evidence chain confidence, is a confidence distribution value of the th independent evidence source to , and is a grain and oil crop state set, is an empty set.

5. The intelligent crop condition monitoring system for grain and oil crops of claim 4, characterized in that, The expression of the dynamic weight of the introduced historical data is as follows: ； wherein, is a dynamic weight for introducing historical data, is a historical regression correction coefficient, and are regression parameters (trained from historical data), is a natural exponential function, is a conflict sensitivity coefficient, is a conflict factor of the th independent evidence source, is a historical feature vector of the th independent evidence source, is a number of focal elements of the th independent evidence source, is a number of focal elements of the th independent evidence source, is a confidence of the th independent evidence source to the subset , is a confidence of the th independent evidence source to the subset , is a normalization function, is a number of sensors of the th independent evidence source, is an impact coefficient of the th independent evidence source to the th sensor, is a detection value of the th independent evidence source to the th sensor, is a precision weight of the th independent evidence source to the th sensor. 6.The grain and oil crop intelligent farm condition monitoring system according to claim 1, characterized in that, The environmental state fusion module comprises: A data integration unit is used to align and splice pre-processed environmental data and the multi-factor fusion evaluation index according to a time sequence to generate a time sequence input sample; An LSTM modeling unit is used to train a long short-term memory network (LSTM) based on a cross-entropy loss and an L2 regularization based on the time sequence input sample, and obtain an environment-growth state mapping model; An update optimization unit is used to update network parameters of the environment-growth state mapping model by using a momentum-based stochastic gradient descent based on a prediction error of the environment-growth state mapping model, and obtain a grain and oil crop environment-growth state comprehensive model.

7. The intelligent crop condition monitoring system for grain and oil crops of claim 6, wherein, The expression of the grain and oil crop environment-growth state comprehensive model is as follows: ； wherein, is the prediction of the comprehensive model of the environment-growth state of grain and oil crops, is the growth state of grain and oil crops at the current time, is the normalized exponential function, is the output result of the optimal environment-growth state mapping model selected in the th iteration, is the parameter of the optimal environment-growth state mapping model, is the parameter of the environment-growth state mapping model, is the latest time sequence input to be predicted at the current time, is the number of iterations, is the time index, is the time sliding window length, and are the parameters of the environment-growth state mapping model, is the parameter of the environment-growth state mapping model in the th iteration, is the parameter of the environment-growth state mapping model in the th iteration, is the learning rate, is the update speed in the th iteration, is the momentum coefficient, is the loss function in the th iteration, is the gradient of is the gradient operator, is the loss function depending on the parameter , is the total number of sample time sequence inputs, is the sample number, is the logarithmic function, is the predicted value of the th time sequence input sample obtained by the environment-growth state mapping model, is the actual value of the th time sequence input sample, is the category probability distribution vector obtained after the softmax is performed on , is the probability value corresponding to in the category probability distribution vector, is the L2 regularization coefficient, is the square of the L2 norm of the parameter , is the hidden state vector output by the environment-growth state mapping model at the last time step of the th sample, This is the output layer weight matrix of the environmental growth state mapping model. This is the output layer bias vector of the environmental growth state mapping model. For parameters Environmental growth state mapping model For the first The complete time series matrix of the training samples, for The one-dimensional feature vector of the time-space environment growth state mapping model. for The environmental data vector at any given time. for Multi-factor fusion evaluation index at any given time. 8.The grain and oil crop intelligent farm condition monitoring system according to claim 1, characterized in that, The digital twin establishment module comprises: A virtual-real synchronization unit is used to synchronize map output results of the pre-processed data and the grain and oil crop environment-growth state comprehensive model based on a timestamp, and construct a virtual twin corresponding to a real farmland. A real-time deduction unit is configured to use a grain-oil crop environment-growth state comprehensive model to predict a growth state of the grain-oil crop in a future time period as an initial state of the virtual twin, and obtain a growth state prediction result of the future multi-granularity time; An intervention simulation unit is configured to simulate in the virtual twin according to a set of user-input control strategies and the growth state prediction result of the future multi-granularity time, and obtain the farmland digital twin system. 9.The grain and oil crop intelligent farm condition monitoring system according to claim 1, characterized in that, The expression of the grain-oil crop growth deviation index is as follows: ； wherein, is a grain and oil crop growth deviation index, is a total number of growth state indexes, is an index of a growth state index, is is a measured value of the th growth state index at the time point, is is a predicted value of the th growth state index at the time point, is a historical maximum value of the th growth state index, is a historical minimum value of the th growth state index.

10. The method according to claim 1, characterized in that, The method comprises the following steps: Obtaining farmland multi-source data through multi-source sensors, and preprocessing the farmland multi-source data to obtain preprocessed data; According to the D-S evidence theory, the preprocessed data is subjected to multi-factor distributed step-by-step fusion to obtain a multi-factor fusion evaluation index; According to the preprocessed data and the multi-factor fusion evaluation index, a long short-term memory network (LSTM) is used to capture a time sequence dependency relationship, and a grain-oil crop environment-growth state comprehensive model is trained; The preprocessed data and the grain-oil crop environment-growth state comprehensive model are fused to construct a farmland digital twin system; The farmland digital twin system is used to perform real-time analysis on the grain-oil crop, and a grain-oil crop growth deviation index is obtained, and the farmland digital twin system is feedback-optimized according to the grain-oil crop growth deviation index.

Images