Precise analysis method based on smart agriculture multi-modal data

Abstract

The invention discloses an accurate analysis method based on smart agriculture multi-modal data, and relates to the technical field of smart agriculture, comprising the following steps: collecting and preprocessing soil humidity and meteorological data, extracting features to construct an environment sequence table, establishing a prediction and anomaly detection model, and outputting a soil humidity prediction value and an environment anomaly probability. And constructing a decision model to generate a regulation index, dynamically adjusting an irrigation strategy, visualizing farmland data distribution and trend in combination with a thermodynamic diagram and a broken line diagram, synchronously recording an operation log, and realizing a multi-modal data driven precise agricultural decision closed loop. Through multi-modal data fusion and a dynamic decision-making mechanism, the problems of traditional agricultural data island and disaster response lag are effectively solved, the regulation and control index is optimized, the water resource utilization rate is improved, the invalid irrigation risk is reduced, the farmland state is fed back in real time, precise partition management is assisted, the disaster response efficiency is improved, and abnormal losses are reduced. And agricultural intelligent upgrading is promoted.

Description

Technical Field

[0001] The present invention relates to the technical field of smart agriculture, and specifically relates to a precise analysis method based on multi-modal data of smart agriculture. Background Art

[0002] Traditional agriculture refers to an agricultural form that, under natural economic conditions, mainly uses human and animal power, manual tools, and iron tools as production methods and relies on generations of inherited experience for farming. Its core characteristics include: technological stagnation, self-sufficiency, low-efficiency production, and experience dominance. Compared with modern agriculture, traditional agriculture has a small scale, less specialized division of labor, and does not rely on external material inputs. Although it has less impact on the ecosystem, it is difficult to break through the low-efficiency bottleneck. For example, the yield per unit area of traditional farmland is usually only 30%-50% of that of modern agriculture and is vulnerable to climate fluctuations. This agricultural form has long dominated agricultural production in history, but with population growth and technological progress, it is gradually transforming towards intensification and scientification.

[0003] In order to solve the problems of data islands and lagging agricultural disaster response in traditional agriculture, the existing technology uses a method of single-sensor monitoring and static threshold judgment for processing. However, there will still be a situation where multi-source data is fragmented and cannot dynamically adapt to environmental changes, which in turn leads to misjudgment of irrigation decisions and lagging disaster response. For example, the traditional method only relies on a single index of soil humidity to set a fixed irrigation threshold, but ignores the combined effects of meteorological data such as light intensity and air humidity. In high-temperature weather, even if the soil humidity meets the standard, the crops may still lack water due to excessive evaporation. In addition, the static model cannot predict the future trend of soil humidity and is difficult to timely warn of droughts and floods. To solve the above problems, a precise analysis method based on multi-modal data of smart agriculture is proposed. Summary of the Invention

[0004] The purpose of the present invention is to provide a precise analysis method based on multi-modal data of smart agriculture to solve the problems raised in the above background art.

[0005] To solve the above technical problems, the technical solution adopted by the present invention is: a precise analysis method based on multi-modal data of smart agriculture, including the following steps: S1. Collect and preprocess multi-modal data in smart agriculture analysis, where the multi-modal data includes soil humidity data and meteorological data; S2. Extract features from the collected multi-modal data and construct an agricultural environment feature sequence table; S3. Combine the agricultural environment feature sequence table to construct an agricultural environment prediction model and an agricultural environment anomaly detection model to predict soil humidity and output the probability of environmental anomalies; S4. Based on the soil humidity and the output probability of environmental anomalies, construct an agricultural analysis decision model and output a regulation index; S5. Set the linkage determination logic for irrigation decisions according to the regulation index; S6. Use a heat map to display the farmland humidity distribution and abnormal areas, and a line chart to display the soil humidity and meteorological changes, and record the decision log.

[0006] A further improvement of the technical solution of the present invention lies in that: in the S1, the process of collecting and preprocessing the soil humidity data and meteorological data includes: Lay soil humidity sensors in a grid layout in the farmland. The soil humidity sensors are based on the principle of dielectric constant, and invert the soil water content through the change of the electromagnetic field between the electrodes to collect soil humidity data in real time; The meteorological data includes light intensity data and air humidity data. Install a light intensity sensor and an air humidity sensor in the central area of the farmland. The light intensity sensor is based on the principle of the photoelectric effect, and its internal photosensitive element converts the optical signal into an electric current signal, and outputs the light intensity data through an amplifier circuit. Based on the change of the humidity-sensitive capacitor inside the air humidity sensor, the air humidity data is output through a capacitance-frequency conversion circuit; Among them, the burial depth of the soil humidity sensor is the main distribution layer of the crop roots, and its spacing is evenly distributed according to the farmland area. The light intensity sensor is installed at a height of 2 meters from the ground, and the air humidity sensor is equipped with a radiation protection cover; The soil humidity data, light intensity data and air humidity data fill in the missing values through linear interpolation, restore the continuity by using adjacent time series data, eliminate the outliers based on the 3σ principle, and map the three types of data to the [0,1] interval through min-max normalization, and retain the fluctuation characteristics of the light accumulation effect, soil humidity and air humidity.

[0007] A further improvement of the technical solution of the present invention lies in that: in the S2, the process of constructing the agricultural environment characteristic sequence table includes: Use the ratio of the difference between the preprocessed soil humidity data at two adjacent time points to the time interval as the soil humidity change rate; Statistically sum the preprocessed light intensity data per hour within a single day and output the daily light accumulation; Divide the standard deviation of the preprocessed air humidity data within a single day by the mean value of the air humidity data within a single day to obtain the air humidity fluctuation coefficient; Integrate the obtained soil humidity change rate, daily light accumulation and air humidity fluctuation coefficient to form an agricultural environment characteristic sequence table.

[0008] A further improvement of the technical solution of the present invention lies in that: in the S3, the process of constructing an agricultural environment prediction model to predict the soil humidity includes: A stacked double memory unit layer based on the long short - term memory network architecture is adopted. The first layer contains 16 memory units, and the second layer contains 8 memory units. Taking seven consecutive days as the time window, the soil moisture change rate and the cumulative light amount are combined into a 7×2 prediction matrix to construct an agricultural environment prediction model; The prediction matrix is sequentially input into the agricultural environment prediction model according to time steps. The information flow is controlled by the internal forget gate and input gate. The memory units store long - term dependencies, and the hidden layer state output at the last time step of the output gate is mapped to the soil moisture prediction value through a fully - connected layer; The mean square error is used to measure the difference between the soil moisture prediction value and the true value. The partial derivatives of the loss function with respect to the weights and biases of each layer are calculated by the gradient descent method, and the maximum number of iterations is set. The training is terminated when the maximum number of iterations is reached.

[0009] A further improvement of the technical solution of the present invention lies in that: in S3, the process of constructing an agricultural environment anomaly detection model and outputting the environmental anomaly probability includes: Taking seven consecutive days as the time window, the soil moisture change rate and the air humidity fluctuation coefficient are combined into a 7×2 input matrix, and each row contains an input feature vector ; The radial basis function of the support vector machine algorithm is adopted to construct a non - linear classification boundary using the input matrix. The classification hyperplane is determined by the support vector weights , sample labels and bias together to construct a decision function , and the classification distance is converted into the environmental anomaly probability through the sigmoid function . The sample labels include normal environment = 1 and abnormal environment = - 1, and its calculation process is as follows: ; ; ; where, is the input feature vector, is the radial basis function parameter, and are the parameters fitted through cross - validation; The abnormal events are labeled according to the historical soil moisture data and the air humidity fluctuation coefficient. The optimal radial basis function parameter and penalty factor are determined through the grid search method to minimize the loss function. Based on the validation set data, the parameters A and B are fitted to make the environmental anomaly probability consistent with the true environmental anomaly probability.

[0010] A further improvement of the technical solution of the present invention lies in that: in S4, the process of determining the soil moisture weight and the environmental anomaly weight includes: Randomly generate N groups of combinations of soil moisture weights and environmental anomaly weights through a genetic algorithm , so that ; Taking the irrigation decision-making effect in historical data as the optimization objective, define the crop yield increase ratio as the irrigation benefit G, and define the reduction ratio caused by abnormal events as the risk loss S. Calculate the fitness F based on the irrigation benefit and risk loss, sort from high to low according to the fitness, and retain the top 30% of the individuals as the parent generation; For the parent generation weights and perform linear crossover to generate offspring , apply a small random perturbation to the offspring weights to generate mutant offspring , when the fitness has not been significantly improved for 20 consecutive times, output the optimal weights , and its calculation process is as follows: ; ; ; ; ; Among them, M is the number of historical data samples, is a crossover coefficient between 0 and 1, is a random perturbation value between -0.1 and 0.1.

[0011] A further improvement of the technical solution of the present invention is that in S4, the process of constructing an agricultural analysis and decision-making model and outputting a regulation index includes: Based on the predicted value of soil moisture, the probability of environmental anomalies, the output optimal soil moisture weight and environmental anomaly weight, construct an agricultural analysis and decision-making model to generate a dynamic regulation index Y, and its calculation process is as follows: ; Manually integrate the actual soil moisture data and the probability of environmental anomalies as feedback data, recalculate the fitness value based on the feedback data, start a new round of genetic algorithm to optimize the soil moisture weight and environmental anomaly weight, and correct the risk loss.

[0012] A further improvement of the technical solution of the present invention is that in S5, the process of setting the linkage determination logic of irrigation decisions includes: Based on the agricultural situation regulation dynamic index, formulate a linkage determination rule, set the dynamic regulation index threshold and the environmental anomaly probability threshold. When Y is greater than the dynamic regulation index threshold and is less than the environmental anomaly probability threshold, trigger an irrigation instruction. When Y is greater than the dynamic regulation index threshold and When the environmental anomaly probability threshold is exceeded, irrigation is suspended and a soil anomaly warning is triggered synchronously. When Y is less than the dynamic regulation index threshold and the predicted soil humidity value is lower than 30%, an artificial verification instruction is pushed, requiring verification of the environmental status and feedback of the results to the agricultural analysis and decision-making model.

[0013] A further improvement of the technical solution of the present invention lies in: in S6, the process of using a heat map to display the farmland humidity distribution and abnormal areas includes: Based on the soil humidity sensor data of the grid layout, the farmland is divided into uniform grids, and the position data of the soil humidity sensors are mapped to the center of each grid. Different degrees of soil humidity levels are represented by the green-yellow-red color scale, and the environmental anomaly areas are superimposed with black borders, and the transparency increases with the increase of the environmental anomaly probability. The latest data is obtained every 5 minutes to refresh the display of the heat map.

[0014] A further improvement of the technical solution of the present invention lies in: in S6, the process of using a line chart to display the soil humidity and meteorological changes and record the decision log includes: Taking 7 consecutive days as the time window, the horizontal axis is the date, the left vertical axis shows the soil humidity change rate and the predicted soil humidity value, and the right vertical axis shows the daily light cumulative amount and the air humidity fluctuation coefficient. The blue solid line is the soil humidity change rate, the orange solid line is the daily light cumulative amount, and the green solid line is the air humidity fluctuation coefficient. Abnormal warnings are marked with red triangles at the corresponding time points, and clicking on the nodes is supported to view detailed data; Record the trigger time, type, regulation index, environmental anomaly probability, predicted soil humidity and operation details of each operation, store them in the agricultural multi-modal database according to structured fields, support filtering and querying by combining the trigger time, type, and sensor device number, write the log data in real time, and can be exported as a standard format file.

[0015] Due to the adoption of the above technical solution, the technical progress achieved by the present invention compared with the prior art is: 1. The present invention provides a precise analysis method based on multi-modal data of smart agriculture. By integrating multi-modal data such as soil humidity, light, and air humidity, a dynamic decision-making closed loop is constructed, breaking the traditional agricultural data island, realizing the adaptive optimization of irrigation strategies, increasing the water resource utilization rate by 20%-30%, and reducing the risk of ineffective irrigation such as high-temperature evaporation by 50%.

[0016] 2. The present invention provides a precise analysis method based on multi-modal data of smart agriculture. Combining the LSTM prediction model and the support vector machine anomaly detection, it accurately predicts the future soil humidity and quantifies the environmental anomaly probability, improves the disaster response efficiency by 60%, reduces the loss of abnormal events by more than 50%, and ensures the growth stability of crops.

[0017] 3. The present invention provides a precise analysis method based on multi-modal data of smart agriculture. Relying on heat maps, line chart visualization, and full-process log records, it provides real-time feedback on farmland humidity distribution, meteorological trends, and operation basis, supports zonal precise management and disaster retrospective analysis, and promotes the transparency and intelligent upgrade of agricultural production. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0019] Figure 1 It is a flowchart of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0021] Embodiment, as Figure 1 shown, the present invention provides a precise analysis method based on multi-modal data of smart agriculture, including the following steps: S1. Collect and preprocess multi-modal data in smart agriculture analysis. Among them, the multi-modal data includes soil humidity data and meteorological data. Soil humidity sensors are buried in the farmland according to a grid layout. The soil humidity sensors are based on the principle of dielectric constant, and the soil water content is inverted through the change of the electromagnetic field between the electrodes to collect soil humidity data in real time. The meteorological data includes light intensity data and air humidity data. Light intensity sensors and air humidity sensors are installed in the central area of the farmland. The light intensity sensors are based on the principle of the photoelectric effect, and the internal photosensitive elements convert the optical signal into an electric current signal, and the light intensity data is output through an amplifier circuit. Based on the change of the humidity-sensitive capacitor inside the air humidity sensor, the air humidity data is output through a capacitance-frequency conversion circuit. Among them, the burial depth of the soil humidity sensor is the main distribution layer of the crop roots, and its spacing is evenly distributed according to the farmland area. The light intensity sensor is installed at a height of 2 meters from the ground, and the air humidity sensor is equipped with a radiation protection cover. The missing values of the soil humidity data, light intensity data, and air humidity data are filled by linear interpolation, and the continuity is restored by using adjacent time series data. The outliers are removed based on the 3σ principle, and the three types of data are mapped to the [0,1] interval through min-max normalization, and the fluctuation characteristics of the light accumulation effect, soil humidity, and air humidity are retained; S2. Extract features from the collected multi-modal data and construct an agricultural environment feature sequence table. Take the ratio of the difference between the preprocessed soil humidity data at two adjacent time points to the time interval as the soil humidity change rate. Statistically calculate the sum of the preprocessed light intensity data per hour within a single day and output the daily light accumulation. Divide the standard deviation of the preprocessed air humidity data within a single day by the mean value of the air humidity data within a single day to obtain the air humidity fluctuation coefficient. Integrate the obtained soil humidity change rate, daily light accumulation, and air humidity fluctuation coefficient to form an agricultural environment feature sequence table; S3. Combine the agricultural environment feature sequence table to construct an agricultural environment prediction model and an agricultural environment anomaly detection model to predict soil humidity and output the environmental anomaly probability. Adopt a stacked dual memory cell layer based on the long short-term memory network architecture. The first layer contains 16 memory cells, and the second layer contains 8 memory cells. Take seven consecutive days as the time window, combine the soil humidity change rate and the light accumulation to form a 7×2 prediction matrix, and construct an agricultural environment prediction model. The prediction matrix is input into the agricultural environment prediction model step by step according to the time steps. The information flow is controlled by the internal forget gate and input gate. The memory cells store the long-term dependence relationship. The hidden layer state output at the end time step of the output gate is mapped to the soil humidity prediction value through a fully connected layer. Use the mean square error to measure the difference between the soil humidity prediction value and the true value, and calculate the partial derivatives of the loss function with respect to the weights and biases of each layer through the gradient descent method. Set the maximum number of iterations and terminate the training when the maximum number of iterations is reached. Take seven consecutive days as the time window, combine the soil humidity change rate and the air humidity fluctuation coefficient to form a 7×2 input matrix, and each row contains an input feature vector , the radial basis function using the support vector machine algorithm , constructs a non-linear classification boundary using the input matrix, and the classification hyperplane is determined by the support vector weights , sample labels and bias jointly determine the classification hyperplane to construct the decision function , converts the classification distance into the environmental anomaly probability through the sigmoid function , the sample labels include normal environment = 1 and abnormal environment = -1, and the calculation process is as follows: ; ; ; wherein, is the input feature vector, is the radial basis function parameter, and are the parameters fitted through cross-validation. Abnormal events are labeled based on historical soil moisture data and air humidity fluctuation coefficients. The optimal radial basis function parameters and penalty factors are determined through the grid search method, the loss function is minimized, and based on the validation set data, the parameters A and B are fitted to make the environmental anomaly probability consistent with the true environmental anomaly probability; S4. Based on the soil moisture and the output environmental anomaly probability, construct an agricultural analysis decision model, output the regulation index, and randomly generate N groups of combinations of soil moisture weights and environmental anomaly weights through the genetic algorithm , make , with the irrigation decision effect in historical data as the optimization goal, define the crop yield increase ratio as the irrigation benefit G, and define the reduction ratio caused by abnormal events as the risk loss S. Calculate the fitness F based on the irrigation benefit and the risk loss, sort from high to low according to the fitness, retain the top 30% of the individuals as the parent generation, and perform linear crossover on the parent generation weights and to generate offspring , apply a small random perturbation to the offspring weights to generate mutant offspring , when the fitness has not been significantly improved for 20 consecutive times, output the optimal weights , and the calculation process is as follows: ; ; ; ; ; where M is the number of historical data samples, is the crossover coefficient between 0 and 1, is the random perturbation value between -0.1 and 0.1. Based on the predicted soil moisture value, the probability of environmental anomalies, the optimal soil moisture weight and the environmental anomaly weight output, an agricultural analysis and decision-making model is constructed to generate the dynamic regulation index Y. The calculation process is as follows: ; Manually integrate the actual soil moisture data and the probability of environmental anomalies as feedback data, recalculate the fitness value based on the feedback data, start a new round of genetic algorithm to optimize the soil moisture weight and the environmental anomaly weight, and correct the risk loss; S5. According to the regulation index, set the linkage determination logic of the irrigation decision. Based on the agricultural situation regulation dynamic index, formulate the linkage determination rule, set the dynamic regulation index threshold and the environmental anomaly probability threshold. When Y is greater than the dynamic regulation index threshold and is less than the environmental anomaly probability threshold, trigger the irrigation instruction. When Y is greater than the dynamic regulation index threshold and is greater than the environmental anomaly probability threshold, suspend irrigation and synchronously trigger the soil anomaly alarm. When Y is less than the dynamic regulation index threshold and the predicted soil moisture value is lower than 30%, push the manual verification instruction to require verifying the environmental status and feedback the result to the agricultural analysis and decision-making model; S6. Use a heat map to display the farmland humidity distribution and abnormal areas, and a line chart to display the soil moisture and meteorological changes. Record the decision log. Based on the soil moisture sensor data with a grid layout, divide the farmland into uniform grids, map the soil moisture sensor position data at the center of each grid, and represent the high and low soil moisture levels of different degrees through a green-yellow-red color scale. The environmental abnormal areas are overlaid with a black border, and the transparency increases with the increase of the environmental anomaly probability. Obtain the latest data every 5 minutes and refresh the heat map display. Take 7 consecutive days as the time window, with the horizontal axis being the date, the left vertical axis showing the soil moisture change rate and the predicted soil moisture value, and the right vertical axis showing the daily light accumulation and the air humidity fluctuation coefficient. Use a blue solid line for the soil moisture change rate, an orange solid line for the daily light accumulation, and a green solid line for the air humidity fluctuation coefficient. The abnormal alarms are marked with red triangles at the corresponding time points, supporting clicking on the nodes to view the detailed data. Record the trigger time, type, regulation index, environmental anomaly probability, predicted soil moisture, and operation details of each operation, store them in the agricultural multi-modal database according to the structured fields, support filtering and querying by combining the trigger time, type, and sensor device number, write the log data in real time, and can be exported as a standard format file.

[0022] As described above, it is only the specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application can easily think of changes or substitutions, which should all be covered within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims described above.

Claims

1. A precise analysis method based on multi-modal data of smart agriculture, characterized in that, It includes the following steps: S1. Collect and preprocess multi-modal data in smart agriculture analysis, where the multi-modal data includes soil humidity data and meteorological data; S2. Extract features from the collected multi-modal data and construct an agricultural environment feature sequence list; S3. Combine the agricultural environment feature sequence list to construct an agricultural environment prediction model and an agricultural environment anomaly detection model to predict soil humidity and output the environmental anomaly probability; S4. Based on the soil humidity and the output environmental anomaly probability, construct an agricultural analysis decision model and output a regulation index; S5. Set the linkage determination logic of irrigation decision according to the regulation index; S6. Use a heat map to display the farmland humidity distribution and abnormal areas, and a line chart to display the soil humidity and meteorological changes, and record the decision log.

2. The precise analysis method based on multi-modal data of smart agriculture according to claim 1, wherein: In the above S1, the process of collecting and preprocessing the soil humidity data and meteorological data includes: Lay soil humidity sensors in the farmland according to a grid layout. The soil humidity sensors are based on the principle of dielectric constant, and invert the soil water content through the change of the electromagnetic field between the electrodes to collect soil humidity data in real time; The meteorological data includes light intensity data and air humidity data. Install a light intensity sensor and an air humidity sensor in the central area of the farmland. The light intensity sensor is based on the photoelectric effect principle, and its internal photosensitive element converts the optical signal into an electric current signal, and outputs the light intensity data through an amplifier circuit. Based on the change of the humidity-sensitive capacitor inside the air humidity sensor, the air humidity data is output through a capacitance-frequency conversion circuit; The soil humidity data, light intensity data and air humidity data fill in the missing values through linear interpolation, use adjacent time series data to restore continuity, eliminate outliers based on the 3σ principle, and map the three types of data to the [0,1] interval through min-max normalization, and retain the fluctuation characteristics of light accumulation effect, soil humidity and air humidity.

3. The precise analysis method based on multi-modal data of smart agriculture according to claim 2, wherein: In the above S2, the process of constructing the agricultural environment feature sequence list includes: Take the ratio of the difference between the preprocessed soil humidity data at two adjacent time points to the time interval as the soil humidity change rate; Statistically sum the preprocessed light intensity data per hour within a single day and output the daily light accumulation; Divide the standard deviation of the preprocessed air humidity data within a single day by the mean value of the air humidity data within a single day to obtain the air humidity fluctuation coefficient; Integrate the obtained soil humidity change rate, daily light accumulation and air humidity fluctuation coefficient to form an agricultural environment feature sequence list.

4. The precise analysis method based on multi-modal data of smart agriculture according to claim 3, characterized in that: In the above S3, the process of constructing the agricultural environment prediction model to predict soil humidity includes: Adopt a stacked double memory cell layer based on the long short-term memory network architecture. The first layer contains 16 memory cells, and the second layer contains 8 memory cells. Take seven consecutive days as the time window, combine the soil humidity change rate and the light accumulation into a 7×2 prediction matrix, and construct an agricultural environment prediction model; The prediction matrix is sequentially input into the agricultural environment prediction model according to the time step. The information flow is controlled by the internal forget gate and input gate. The memory cells store the long-term dependence relationship, and the hidden layer state output at the end time step of the output gate is mapped to the soil humidity prediction value through a fully connected layer; The mean square error is used to measure the difference between the predicted value and the true value of soil humidity. The partial derivatives of the loss function with respect to the weights and biases of each layer are calculated by the gradient descent method, and the maximum number of iterations is set. The training is terminated when the maximum number of iterations is reached.

5. The precise analysis method based on multi-modal data of smart agriculture according to claim 4, characterized in that: In S3, the process of constructing an agricultural environment anomaly detection model and outputting the environmental anomaly probability includes: Taking seven consecutive days as the time window, the soil moisture change rate and the air humidity fluctuation coefficient are combined into a 7×2 input matrix, and each row contains the input feature vector ; Radial basis function using the support vector machine algorithm , construct a non-linear classification boundary using the input matrix, and jointly determine the classification hyperplane by the support vector weights , sample labels and bias to construct a decision function , convert the classification distance into the environmental anomaly probability through the sigmoid function , where the sample labels include normal environment = 1 and abnormal environment = -1; Abnormal events are labeled based on historical soil humidity data and air humidity fluctuation coefficients. The optimal radial basis function parameters and penalty factors are determined by the grid search method to minimize the loss function. Based on the validation set data, parameters A and B are fitted to make the environmental anomaly probability consistent with the true environmental anomaly probability.

6. The precise analysis method based on multi-modal data of smart agriculture according to claim 5, characterized in that: In S4, the process of determining the soil humidity weight and environmental anomaly weight includes: Randomly generate N groups of combinations of soil moisture weights and environmental anomaly weights through a genetic algorithm , so that ; Taking the irrigation decision effect in historical data as the optimization goal, defining the crop yield increase ratio as the irrigation benefit G, and defining the reduction ratio caused by abnormal events as the risk loss S. The fitness F is calculated based on the irrigation benefit and risk loss, and the individuals are sorted from high to low according to the fitness. The top 30% of the individuals are retained as the parents. For the parental weights and perform linear crossover to generate offspring , apply a small random perturbation to the offspring weights to generate mutated offspring , when the fitness has not improved significantly for 20 consecutive times, output the optimal weights .

7. The precise analysis method based on multi-modal data of smart agriculture according to claim 6, characterized in that: In S4, the process of constructing an agricultural analysis and decision-making model and outputting the regulation index includes: Based on the predicted value of soil humidity, the environmental anomaly probability, the output optimal soil humidity weight and environmental anomaly weight, an agricultural analysis and decision-making model is constructed to generate a dynamic regulation index Y. The actual soil humidity data and environmental anomaly probability are manually integrated as feedback data. Based on the feedback data, the fitness value is recalculated, a new round of genetic algorithm is started to optimize the soil humidity weight and environmental anomaly weight, and the risk loss is corrected.

8. A precise analysis method based on multi-modal data of smart agriculture according to claim 7, characterized in that: In S5, the process of setting the linkage determination logic of irrigation decisions includes: Based on the dynamic index of agricultural situation regulation, formulate linkage judgment rules, set the threshold of the dynamic regulation index and the threshold of the environmental anomaly probability. When Y is greater than the threshold of the dynamic regulation index and less than the threshold of the environmental anomaly probability, trigger the irrigation instruction. When Y is greater than the threshold of the dynamic regulation index and greater than the threshold of the environmental anomaly probability, suspend irrigation and simultaneously trigger a soil anomaly alarm. When Y is less than the threshold of the dynamic regulation index and the predicted soil humidity value is lower than 30%, push an artificial verification instruction, requiring verification of the environmental status and feedback of the results to the agricultural analysis and decision-making model.

9. The precise analysis method based on multi-modal data of smart agriculture according to claim 8, characterized in that: In S6, the process of using a heat map to display the farmland humidity distribution and abnormal areas includes: Based on the soil humidity sensor data with a grid layout, the farmland is divided into uniform grids, and the position data of the soil humidity sensors are mapped to the center of each grid. Different levels of soil humidity are represented by a green-yellow-red color scale, and the abnormal areas are overlaid with a black border, and the transparency increases with the increase of the environmental anomaly probability. The latest data is obtained every 5 minutes to refresh the heat map display.

10. A precise analysis method based on multi-modal data of smart agriculture according to claim 9, characterized in that: In S6, the process of using a line chart to display the soil humidity and meteorological changes and record the decision-making log includes: Taking 7 consecutive days as the time window, the horizontal axis is the date, the left vertical axis shows the soil humidity change rate and the predicted value of soil humidity, and the right vertical axis shows the daily cumulative sunlight amount and the air humidity fluctuation coefficient. The blue solid line is the soil humidity change rate, the orange solid line is the daily cumulative sunlight amount, and the green solid line is the air humidity fluctuation coefficient. Abnormal alarms are marked with red triangles at the corresponding time points, and detailed data can be viewed by clicking on the nodes. The trigger time, type, regulation index, environmental anomaly probability, predicted soil humidity and operation details of each operation are recorded and stored in the agricultural multi-modal database according to structured fields. Support filtering and querying by the combination of trigger time, type, and sensor device number. The log data is written in real time and can be exported as a standard format file.

Images