Intelligent agricultural information management method and system based on cloud computing

Abstract

This invention relates to the field of cloud computing technology, solving the technical problems of existing technologies having weak ability to capture short-term fluctuations, being prone to getting trapped in local optima, and lacking feedback evaluation of strategy implementation effects. In particular, it relates to a cloud-based smart agriculture information management method and system. The method includes the following steps: preprocessing the collected raw environmental physical signals and raw crop physiological images to obtain a multi-source agricultural cloud dataset; performing anomaly diagnosis based on the multi-source agricultural cloud dataset to obtain crop health reports and abnormal data; providing interpretable physical feedback for model updates; using online learning to achieve incremental parameter updates; and performing real-time feedback evaluation to avoid getting trapped in local optima and affecting the model's analytical capabilities, thus improving the accuracy of model analysis results; balancing the efficiency of optimal parameter search and utilization through Bayesian optimization of dynamic hyperparameter search; and finally improving model robustness through cross-validation.

Description

Technical Field

[0001] This invention relates to the field of cloud computing technology, and in particular to a cloud-based smart agriculture information management method and system. Background Technology

[0002] Cloud computing, a type of distributed computing, refers to breaking down massive data processing programs into countless smaller programs via a network cloud. These smaller programs are then processed and analyzed using various models to obtain results, which are then returned to the user. In smart agricultural information management, a large amount of computing resources are often required. Cloud computing can provide high-performance computing server clusters with powerful computing and prediction capabilities. It can accurately predict crop yields, the timing and extent of pests and diseases, and abnormal crop conditions. Furthermore, based on real-time monitoring data, it can intelligently regulate agricultural operations such as irrigation and fertilization, achieving efficient resource utilization and precision agricultural management.

[0003] Existing technologies typically have a weak ability to capture short-term environmental data fluctuations. When adjusting hyperparameters, they rely on simple heuristic adjustment methods, which can easily lead to local optima. They lack feedback and evaluation of the effectiveness of strategy implementation, and the optimization direction can easily deviate from actual needs. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a cloud-based smart agriculture information management method and system, which solves the technical problems of existing technologies having a weak ability to capture short-term fluctuations, being prone to getting trapped in local optima, and lacking feedback evaluation of the effectiveness of strategy implementation.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a smart agriculture information management method based on cloud computing, the steps of which are: preprocessing the collected raw environmental physical signals and raw crop physiological images to obtain a multi-source agricultural cloud dataset; Anomaly diagnosis is performed based on multi-source agricultural cloud datasets to obtain crop health reports and abnormal data; Crop data with abnormal data and abnormal states are selected from crop health reports and abnormal data, and agricultural implementation data sequences are obtained through fuzzy logic and genetic optimization. Based on the agricultural implementation data sequence, agricultural information is regulated and controlled, and environmental control optimization instructions are obtained to obtain an integrated agricultural management model; Based on environmental control optimization instructions, an optimized multi-source agricultural cloud dataset is obtained, and the agricultural integrated management model is dynamically optimized to obtain the agricultural optimized management model and environmental control update instructions.

[0006] Preferably, the acquired raw environmental physical signals and raw crop physiological images are preprocessed, including: Preliminary environmental data and preliminary image data were obtained through preliminary calibration based on the original environmental physical signals and original crop physiological images. Preliminary environmental data and preliminary image data are spatiotemporally aligned and encapsulated to obtain spatiotemporally structured data packets. Multi-source agricultural cloud datasets are obtained by storing spatiotemporal structured data packets in an agricultural cloud database.

[0007] Preferably, crop anomaly diagnosis is performed on multi-source agricultural cloud datasets, including: A feature extraction layer is constructed based on a multi-source agricultural cloud dataset, and temporal feature vectors and crop image depth feature vectors are obtained based on the feature extraction layer. Based on temporal features, a Kalman filter fusion layer is constructed to obtain the optimal environmental state vector and environmental covariance matrix; Anomaly diagnosis and confidence assessment are performed on the optimal environmental state vector and crop image depth feature vector to obtain crop health reports and anomaly data.

[0008] Preferably, anomaly diagnosis and confidence assessment are performed on the optimal environmental state vector and the crop image depth feature vector, including: An attention layer is constructed to process the optimal environment state vector and the crop image depth feature vector using an attention mechanism, thereby obtaining enhanced joint features. Based on the enhanced joint features, a preliminary probability distribution of anomaly types is obtained through a fully connected neural network layer; Crop health reports and anomalous data are obtained by calculating covariance confidence based on the preliminary anomaly type probability distribution and environmental covariance matrix.

[0009] Preferably, the agricultural implementation data sequence is obtained through fuzzy logic and genetic optimization processing, including: Based on crop data with abnormal data and abnormal states, multi-objective functions and constraints are obtained through fuzzy logic processing. The initial population is obtained by initializing the genetic population for the multi-objective function and constraints. Agricultural implementation data sequences were obtained based on the initial population through genetic optimization.

[0010] Preferably, genetic optimization is performed based on the initial population, including: Based on the initial population, fitness vectors and Pareto ranks are generated using Pareto sorting. Simulated chromosome crossover is performed on each fitness vector and Pareto rank to obtain an updated population. Agricultural implementation data sequences are generated based on Pareto solutions selected by fuzzy preferences from updated populations.

[0011] Preferably, the regulation of agricultural information based on agricultural implementation data sequences includes: Based on agricultural implementation data sequences, zoned and time-based control data are obtained through spatiotemporal grid decomposition. Conflict detection is performed on the time-series control data for each zone, and a conflict-free instruction set is obtained. Based on the conflict-free instruction set, environmental control optimization instructions are obtained through energy consumption scheduling optimization.

[0012] Preferably, the agricultural integrated management model is dynamically optimized by acquiring optimized multi-source agricultural cloud datasets based on environmental control optimization instructions, including: Based on the optimization of multi-source agricultural cloud datasets, an effect indicator vector is obtained through indicator difference processing; The model parameters of the integrated agricultural management model are obtained, and the updated model parameters are obtained through online stochastic gradient descent processing based on the performance index vector, the optimized multi-source agricultural cloud dataset, and the model parameters. Based on updating the model parameters, the optimized hyperparameters are obtained through Bayesian optimization. Based on the updated model parameters and optimized hyperparameters, cross-validation was performed to obtain the agricultural optimization management model; The newly acquired raw environmental physical signals and raw crop physiological images are input into the agricultural optimization management model to obtain environmental control update instructions.

[0013] This technical solution also provides a system applied to the aforementioned cloud-based smart agriculture information management method, the system comprising: The preprocessing module is used to preprocess the collected raw environmental physical signals and raw crop physiological images to obtain a multi-source agricultural cloud dataset. The anomaly diagnosis module is used to perform anomaly diagnosis based on multi-source agricultural cloud datasets and obtain crop health reports and anomaly data. The implementation data module is used to select crop data with abnormal data and abnormal states from crop health reports and abnormal data, and obtain agricultural implementation data sequences through fuzzy logic and genetic optimization processing; The preliminary management module is used to regulate agricultural information based on agricultural implementation data sequences and obtain environmental control optimization instructions to obtain an integrated agricultural management model. The model optimization module is used to obtain optimized multi-source agricultural cloud datasets based on environmental control optimization instructions, and to dynamically optimize the agricultural integrated management model to obtain the agricultural optimized management model and environmental control update instructions.

[0014] By employing the above technical solution, the present invention provides a smart agriculture information management method and system based on cloud computing, which has at least the following beneficial effects: 1. This invention reduces noise interference through dynamic data fusion using Kalman filtering, and further enhances key features through multimodal feature fusion guided by attention mechanism. It calculates confidence based on uncertainty and generates detailed reports, which can more accurately and comprehensively reflect the health status and causes of crops, and has high real-time performance and reliability.

[0015] 2. This invention effectively handles the fuzziness and noise interference in agricultural data through a fuzzy logic controller, improving the robustness of the model. By using constraint-guided population initialization and simulated binary crossover operations, it achieves more efficient search and avoids the defect of existing technologies that are prone to getting trapped in local optima. Through fuzzy preference weighted scoring, it takes into account both actual needs and multi-objective balance, improving the practicality and efficiency of agricultural implementation data sequences.

[0016] 3. This invention effectively solves the problems of insufficient data allocation and frequent equipment command conflicts in traditional methods through the spatiotemporal grid decomposition algorithm. Furthermore, it dynamically optimizes the command execution order through the energy-optimal scheduling model, which significantly reduces the overall energy consumption and equipment wear of the agricultural Internet of Things system, while ensuring the temporal coordination of environmental control and water and fertilizer management.

[0017] 4. This invention provides interpretable physical feedback for model updates, uses online learning to achieve incremental parameter updates, avoids the repeated use of old data which affects the model's analytical capabilities, and improves the accuracy of model analysis results. By using Bayesian optimization to dynamically search for hyperparameters, it balances the efficiency of optimal parameter search and utilization, and finally improves the robustness of the model through cross-validation. Attached Figure Description

[0018] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart of the cloud computing-based smart agriculture information management method of the present invention; Figure 2 This is a structural block diagram of the cloud computing-based smart agriculture information management system of the present invention. Detailed Implementation

[0019] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. This will allow for a full understanding of how the present application uses technical means to solve technical problems and achieve technical effects, and to facilitate its implementation.

[0020] Example 1: Due to the limited ability of existing technologies to capture short-term fluctuations, their tendency to get trapped in local optima, and the lack of feedback and evaluation of strategy implementation effectiveness, please refer to [the relevant documentation / reference]. Figure 1 This embodiment provides a cloud computing-based smart agriculture information management method, which can avoid getting trapped in local optima, thus affecting the model's analytical capabilities and improving the accuracy of the model's analysis results. By using Bayesian optimization to dynamically search for hyperparameters, it balances the efficiency of optimal parameter search and utilization. Finally, cross-validation improves the model's robustness. The method includes the following steps: S1. Preprocess the collected raw environmental physical signals and raw crop physiological images to obtain a multi-source agricultural cloud dataset. Existing technologies for agricultural data collection and processing are prone to creating information silos due to inconsistent agricultural data protocols. Directly uploading raw data consumes bandwidth and contains invalid noise. To address these issues, the specific steps are as follows: S11. Preliminary environmental data and preliminary image data are obtained through preliminary calibration based on the original environmental physical signals and original crop physiological images. The original environmental physical signals include temperature data, humidity data, light data, wind speed data, and gas data in agriculture. The original crop physiological images include visible light images, multispectral images, and thermal infrared images. In practice, known physical quantities are first generated using high-precision standard equipment, such as standard temperature sources and voltage sources, and original sensor signals, such as temperature values, are collected. The calibration coefficients and calibration addends are obtained by fitting linear equations using the least squares method. The calibration coefficients are then added to the calibration addends using the original environmental physical signals to obtain the calibrated preliminary environmental data. Preliminary image data is obtained by color balance processing and median filtering based on the original crop physiological images. Color balance processing and median filtering are commonly used image preprocessing methods, which will not be elaborated here.

[0021] S12. The preliminary environmental data and preliminary image data are spatiotemporally aligned and encapsulated to obtain spatiotemporally structured data packets. The time offset is calculated according to the Network Time Protocol (NPT) to obtain the time offset. The local time of the preliminary environmental data and preliminary image data is obtained. The time-aligned data is obtained by adding the time offset to the local time. The rotation matrix and translation matrix are obtained according to the checkerboard calibration method. The sensors that collect the original environmental physical signals and the original crop physiological images are spatially aligned. The spatially aligned data is obtained by multiplying the local position vector of the sensor by the rotation matrix and adding the translation matrix. The checkerboard calibration method is a commonly used method for spatial calibration, which will not be elaborated here. In this invention, a plain farmland is used as an example, so the images are all two-dimensional images. The time-aligned data and spatially aligned data are converted into binary data through Protobuf and encapsulated to obtain spatiotemporally structured data packets. Protobuf is a binary protocol that can organize time-aligned data and spatially aligned data, such as timestamps, spatial coordinates, and sensor values, into standardized data packets that can be transmitted and parsed. It will not be elaborated here.

[0022] S13. Storing the spatiotemporal structured data packets in the agricultural cloud database yields a multi-source agricultural cloud dataset. The spatiotemporal structured data packets are then uploaded to and stored in the cloud database via TLS encryption. The cloud database batch processes and writes the spatiotemporal structured data packets to obtain the multi-source agricultural cloud dataset stored in the cloud database. TLS encryption is an encryption protocol used for secure data transmission over the Internet, widely applied to protect the privacy and integrity of network communications. It is a commonly used encryption method and will not be elaborated upon here. This invention improves data quality and reduces cloud transmission load through preliminary cleaning and standardization using edge computing, laying the foundation for data fusion in subsequent steps.

[0023] S2. Perform anomaly diagnosis on multi-source agricultural cloud datasets to obtain crop health reports and anomaly data. Existing technologies often suffer from insufficient data fusion when dealing with crop health diagnosis, making it difficult to effectively integrate data from different sources. This leads to incomplete information utilization, poor robustness to data noise, and susceptibility to environmental factors, affecting diagnostic accuracy. To address these issues, the specific implementation steps are as follows: S21. Construct a feature extraction layer based on the multi-source agricultural cloud dataset. Obtain temporal feature vectors and crop image depth feature vectors based on the feature extraction layer. Construct a sliding window with a window size of e and a stride of a based on the multi-source agricultural cloud dataset. Calculate the temporal features of the multi-source agricultural cloud data within the sliding window, including mean, variance, and standard deviation, or temporal features such as Fourier coefficients. Concatenate the temporal features into a temporal feature vector. Perform convolutional feature processing on the multi-source agricultural cloud dataset using CNN feature extraction. After two-dimensional convolution operation, this operation multiplies and sums the weight parameters of the convolution layer with the local image region, and adds a bias term to adjust the numerical distribution. Then, perform a non-linear transformation on the operation result through an activation function. The activation function can use the ReLU function to enhance the feature expression ability. Finally, output a crop image depth feature vector that can characterize the deep semantic information of the crop image. Among them, two-dimensional convolution operation is a commonly used convolution processing method, often used to process image data, and will not be elaborated here.

[0024] S22. Based on the temporal feature vector, a Kalman filter fusion layer is constructed to obtain the optimal environmental state vector and environmental covariance matrix. Based on the temporal feature vector, a prediction layer and an update layer of the Kalman filter fusion layer are constructed respectively. In the prediction layer, the optimal state estimate of the temporal feature vector at the previous moment is used, combined with the dynamic characteristics of the system, such as the change law of environmental parameters over time, to infer the prior state prediction value at the current moment. At the same time, according to the uncertainty caused by process noise, such as sensor error or model simplification, the uncertainty range of the prediction is adjusted. In the update layer, by fusing the measured data at the current moment, the difference weight between the prior state prediction value and the observed value is calculated, the output state estimate value is dynamically adjusted to reduce the error, and the uncertainty range is updated synchronously. Finally, the optimal environmental state vector that reflects the characteristics of multi-source data and the dynamic changes of the environment, and its uncertainty measure, i.e., the covariance matrix, are output.

[0025] S23. Perform anomaly diagnosis and confidence assessment on the optimal environmental state vector and crop image depth feature vector to obtain crop health reports and abnormal data; S231. Construct an attention layer to process the optimal environmental state vector and the crop image depth feature vector using an attention mechanism to obtain enhanced joint features. First, linear projections are performed on the optimal environmental state vector and the crop image depth feature vector using a learnable transformation matrix to map them to the same semantic space to enhance correlation. Then, the similarity score between the two modal features is calculated, such as by dot product operation, and the score is normalized using a scaling factor, such as the square root of the feature dimension. Then, the similarity is converted into attention weights in the form of a probability distribution using the softmax function. Finally, the crop image depth feature vector is weighted and summed according to the attention weights, and the result is fused with the environmental state features to generate enhanced joint features that simultaneously contain environmental dynamic information and crop visual semantic information. The learnable transformation matrix is ​​such as the weight matrix of the fully connected layer, the environmental dynamic information is such as temperature and humidity in multi-source sensor data, and the crop visual semantic information includes crop morphology, color, and other information.

[0026] S232. Based on the enhanced joint features, a preliminary probability distribution of anomaly types is obtained through a fully connected neural network layer. In the fully connected neural network classification process, the input enhanced joint features are first linearly transformed, and the enhanced joint features are mapped to the classification space through a learnable weight matrix. At the same time, a bias term is superimposed to adjust the baseline position of the classification decision boundary. Then, the softmax function is used to normalize the linear transformation result, converting the original output value into a probability distribution form, so that the sum of the probabilities of all anomaly categories is 1. The final output preliminary probability distribution intuitively reflects the probability of the current crop state belonging to health, disease, or environmental stress, such as water shortage and other anomalies.

[0027] S233. Based on the preliminary anomaly type probability distribution and the environmental covariance matrix, crop health reports and anomaly data are obtained by calculating the covariance confidence score. In this step, the overall error of environmental state estimation is first extracted from the covariance matrix output by the Kalman filter. The uncertainty of the prediction can be reflected by calculating the trace of the covariance matrix. At the same time, the information entropy is calculated from the preliminary anomaly type probability distribution to quantify the ambiguity of the classification results. Then, the overall error and information entropy are weighted and combined, and the comprehensive uncertainty is mapped to a confidence score through a normalization formula. The larger the environmental error or the more dispersed the classification probability distribution, the lower the confidence score. Finally, crop health reports containing anomaly labels are generated. This invention generates crop health reports, highlighting abnormal data such as water shortage or disease (A), with related abnormal humidity data (B) significantly lower than the commonly used humidity threshold for crops, along with a confidence score. This provides interpretable quantitative evidence for agricultural decision-making. The invention effectively integrates data from different sources and mines deep image features through multi-step collaborative processing. Dynamic data fusion based on Kalman filtering reduces noise interference, and multimodal feature fusion guided by an attention mechanism further strengthens key features. Confidence is calculated based on uncertainty to generate detailed reports, enabling a more accurate and comprehensive reflection of crop health status and causes, while also possessing high real-time performance and reliability.

[0028] S3. Select crop data with abnormal data and abnormal states from crop health reports and abnormal data, and obtain agricultural implementation data sequences through fuzzy logic and genetic optimization. Existing technologies have difficulty handling fuzzy and uncertain data such as crop health status and weather forecasts, resulting in a lack of flexibility in setting constraints. Furthermore, conventional genetic algorithms have low search efficiency and are prone to premature convergence, which may cause agricultural analysis results to deviate from actual production needs. To solve the above problems, the specific implementation steps are as follows: S31. Based on crop data with abnormal data and abnormal states, a multi-objective function and constraints are obtained through fuzzy logic processing. First, the causes of abnormalities in the crop health status report are integrated, such as water shortage level, disease type, market price fluctuations, resource usage costs, and weather forecast data, such as rainfall probability and temperature range. The integration method can be obtained through normalization processing, fuzzy coding, and feature concatenation. The discrete numerical or categorical inputs are converted into continuous fuzzy membership degrees through a fuzzy logic controller. For example, regarding water shortage level, based on the actual value of the water stress index, the fuzzy membership degree of belonging to the low water shortage level is calculated using a trapezoidal membership function. If the current stress index is close to the preset low water shortage upper limit, the membership degree decreases linearly as the index increases. The membership degree decreases as the index falls below the upper limit, and then returns to zero. If the index approaches the lower limit of water scarcity, the membership degree increases linearly as the index decreases. Once it falls below the lower limit, the membership degree is normalized to 1. Similarly, other input variables, such as high or low resource costs and suitable or unsuitable weather, all complete the membership degree calculation through their respective preset fuzzy set parameters. Finally, fuzzy constraints reflecting the tightness of each constraint and the priority of the objective are generated, such as the range of relevant agricultural variables, such as the allowable range of irrigation water volume between 100 and 200 cubic meters, and multi-objective functions, such as fuzzy weight combinations that maximize revenue and minimize cost. The result of the trapezoidal membership degree function is the membership degree. The formula for calculating the trapezoidal membership degree function is a commonly used formula for calculating membership degree, which will not be elaborated here.

[0029] S32. Genetic population initialization is performed on the multi-objective function and constraints to obtain the initial population. Based on the fuzzified constraints, such as the allowable range of irrigation amount and the upper and lower limits of fertilizer application, and the multi-objective function, such as the priority of simultaneously optimizing yield, cost, and environmental impact, a constraint-guided initialization strategy is designed. Specifically, each decision variable, such as irrigation amount, fertilizer application amount, and pesticide application amount, is encoded into a gene segment within each population, i.e., a chromosome. The value range of the gene is strictly limited to the feasible domain defined by the fuzzy constraints. For example, the initial value of the irrigation amount gene segment is randomly generated within the fuzzified low, medium, and high irrigation demand range. By combining the boundary information of the constraints and the preference weights of the multi-objective function, the system uses hierarchical sampling or random perturbation to generate the initial population, ensuring that each chromosome, i.e., a combination of decision variables, satisfies both the physical constraints of agricultural operations, such as the total water resource limit, and initially meets the fuzzy requirements of economic and environmental goals, providing an initial population set that conforms to the actual scenario for subsequent iterative optimization of the genetic algorithm.

[0030] S33. Obtain agricultural implementation data sequences based on the initial population through genetic optimization. S331. Generate fitness vectors and Pareto ranks based on the initial population using Pareto sorting. In this step, each chromosome in the initial population, for example, a set of agricultural schemes including combinations of irrigation and fertilization, undergoes multi-objective evaluation. For each scheme, the corresponding target values ​​such as yield, cost, and water consumption are predicted using an agricultural model, such as the WheatSM V5 wheat simulation model. The target to be maximized, such as yield, is negatively valued to unify it as a minimization problem, forming a fitness vector. For example, the fitness vector may contain multiple components such as negative predicted yield, total cost, and water consumption. Subsequently, the population is stratified using a fast non-dominated sorting algorithm based on the NSGA II framework. The steps include first identifying all solutions not dominated by other schemes, i.e., no other scheme is better than the others in all objectives, and assigning them the first Pareto rank. After removing these solutions, the process is repeated for the remaining schemes, assigning the second rank, the third rank, and so on, until all chromosomes are sorted. Finally, the fitness vector of each chromosome is output, reflecting multi-objective performance and the corresponding Pareto rank, used to indicate the hierarchy of the solution in the non-dominated relation. NSGA... The fast non-dominated sorting algorithm in the II framework is a classic multi-objective optimization algorithm framework. It seeks a set of balanced solutions among multiple conflicting objectives through non-dominated sorting and crowding distance mechanism, which is the Pareto optimal solution, and will not be elaborated here.

[0031] S332. Perform simulated chromosome crossover on each fitness vector and Pareto level to obtain an updated population. In this step, based on the input chromosome (equivalent to an individual in the initial population), its fitness vector and Pareto level are used to prioritize retaining individuals with lower Pareto levels (i.e., better non-dominated levels) to enter the new generation population, ensuring that excellent solutions are not eliminated. For individuals requiring genetic operations, simulated binary crossover (SBX) is used to generate offspring. The process is as follows: two parent chromosomes are randomly selected, and the gene exchange ratio is dynamically adjusted using a random perturbation factor controlled by a distribution index. If the perturbation factor is biased towards a positive value, the offspring genes will inherit more of the larger value from the parents; if it is biased towards a negative value, the offspring genes will inherit more of the larger value from the parents. If the value is negative, the smaller value is inherited more, eventually generating two offspring chromosomes. Then, polynomial mutation is performed on the offspring chromosomes to introduce a controllable random shift within the gene value range, enhancing population diversity. Finally, the elite individuals are merged with the offspring generated by the genetic operation to form a new generation of decision variable population, which is the updated population. Among them, the distribution index control is the core mechanism of simulating the binary crossover operator in the genetic algorithm. The distribution index η is used to adjust the similarity between the offspring and parent genes, thereby controlling the balance between global search and local search in the search process. Polynomial mutation is a method commonly used in genetic algorithms to make chromosomes more diverse, which will not be elaborated here.

[0032] S333. Based on the updated population, agricultural implementation data sequences are generated through Pareto solution selection using fuzzy preferences. In this step, the updated population, which has undergone multiple generations of evolution, i.e., the Pareto solution set, is comprehensively evaluated. Based on actual needs, such as prioritizing yield, controlling costs, or conserving water resources, fuzzy preference weights are assigned to each objective, such as yield, cost, and water consumption, reflecting the relative importance of different objectives. Then, the objective values ​​of each candidate solution (i.e., a combination of decision variables) are normalized to eliminate dimensional differences. Finally, a weighted sum is obtained to obtain a comprehensive score. The solution with the highest score is selected as the optimized agricultural implementation data sequence, which includes specific data such as irrigation amounts, fertilizer amounts, and pesticide usage amounts. The dosage parameters of the organism can directly guide actual production, balance multiple objective conflicts, and adapt to dynamic environmental changes. This invention transforms uncertain inputs such as crop health status and market price into fuzzy constraints through a fuzzy logic controller, effectively handling the fuzziness and noise interference in agricultural data, improving model robustness. By utilizing constraint-guided population initialization and simulated binary crossover operations, it achieves more efficient search capabilities in multi-objective optimization, avoiding the shortcomings of existing technologies that are prone to getting trapped in local optima. Finally, it dynamically selects the optimal agricultural implementation data sequence from the Pareto solution set through fuzzy preference weighted scoring, taking into account both actual needs and multi-objective balance, and improving the practicality and efficiency of agricultural implementation data sequences.

[0033] S4. Based on the agricultural implementation data sequence, regulate agricultural information and obtain environmental control optimization instructions to obtain an integrated agricultural management model. Existing technologies struggle to flexibly adjust decisions based on soil fertility differences and sudden weather events, lack conflict detection and dynamic rescheduling capabilities, easily leading to resource contention, neglecting energy consumption optimization, resulting in high long-term operating costs and shortened equipment lifespan. To address these issues, the specific implementation steps are as follows: S41. Based on agricultural implementation data sequences, spatiotemporal grid decomposition is used to obtain zoned and time-specific control data. The input agricultural implementation data sequence, such as the overall plan for irrigation, fertilization, and pesticide spraying, is finely decomposed according to spatial regions and time stages. Specifically, a spatiotemporal grid decomposition algorithm is used to divide the entire farmland into multiple control zones, such as sub-regions based on soil type and crop growth. Simultaneously, the implementation cycle is divided into multiple continuous time windows, such as time periods based on growth stages or weather changes. Then, each operation in the overall decision sequence is matched with a specific zone and a specific time window to generate specific control data for each zone within the corresponding time window. Example: For example, if the overall decision sequence requires a total irrigation volume of 300 cubic meters for the entire field in the next two weeks, after spatiotemporal decomposition, it may generate: the first zone is an area with poor soil water retention, which will be irrigated with 80 cubic meters during the seedling stage in the first week; the second zone is an area with high soil fertility, which will be irrigated with 50 cubic meters during the jointing stage in the second week; and the remaining water volume will be allocated to the other zones according to similar logic, ultimately forming a precise control dataset divided by zone and time, which provides a control basis for subsequent steps. Among them, spatiotemporal grid decomposition is a method to finely break down complex agricultural control strategies according to spatial regions and time stages. By constructing a two-dimensional grid model, the overall decision is transformed into executable local control instructions, which will not be elaborated here.

[0034] S42. Perform conflict detection on the zone-based time-based control data and obtain a conflict-free instruction set. In this step, the input zone-based time-based control data, such as the first zone needing 50 cubic meters of irrigation between 8 and 10 am, is first converted into executable instructions for specific equipment. Specifically, this is done through equipment driver mapping rules, based on task type (e.g., irrigation, fertilization, target zone number), equipment identifier (e.g., valve number, nozzle number), and resource parameters (e.g., water flow rate, pesticide concentration), automatically generating equipment operation instructions, such as valve opening degree and opening duration. Furthermore, resource conflicts between instructions are detected in real time. For example, if multiple tasks simultaneously occupy the same valve or pump, they are sorted by priority or time. Staggered scheduling ensures conflict-free instruction sets. For example, if a task requires irrigating 50 cubic meters in the first zone (valve V1, flow rate 10 cubic meters / hour) and irrigating 30 cubic meters in the second zone (valve V1, flow rate 10 cubic meters / hour) to be executed within the same time period, and a resource conflict with valve V1 is detected, the task for the second zone will be automatically adjusted to the next time period or assigned to the backup valve V2. The final output will be a conflict-free preliminary instruction set, such as valve V1 opening between 8-10 am with a flow rate of 50 cubic meters, and valve V2 opening between 10-11 am with a flow rate of 30 cubic meters. Conflict detection is a common method in equipment control to avoid instruction conflicts, and will not be elaborated upon here.

[0035] S43. Based on the conflict-free instruction set, environmental control optimization instructions are obtained through energy consumption scheduling optimization. All instructions within the conflict-free instruction set are traversed, and the energy consumption per unit time for each instruction is calculated, such as the ratio of pump power to flow rate. The execution order is dynamically adjusted, prioritizing instructions with low energy consumption or short execution time, while avoiding additional energy consumption caused by frequent equipment start-ups and shutdowns. Ultimately, the total energy consumption, i.e., the sum of the products of the energy consumption rate and execution time of all instructions, is minimized. For example, if the initial instruction set includes valve A starting irrigation at 8:00 for 2 hours with an energy consumption rate of 500 J / hour and fan B starting ventilation at 8:30 for 1.5 hours with an energy consumption rate of 800 J / hour, and the system detects high energy consumption at the moment of fan start-up and a potential conflict with the irrigation task due to shared power supply, the optimized instruction set might be adjusted so that valve A starts ventilation earlier at 7:50. By starting the fan for 2 hours and utilizing off-peak electricity prices, fan B can avoid peak irrigation times by starting for 1 hour at 10:00 AM. Total energy consumption is reduced from the original 2300 J to 1800 J, while ensuring that environmental control and water / fertilizer management tasks do not interfere with each other. The integrated agricultural management model integrates the S1-S4 algorithm models into a single, unified model. This invention uses a spatiotemporal grid decomposition algorithm to accurately map global agricultural decisions to local tasks at different time zones. Combined with equipment-driven mapping and conflict detection mechanisms, it effectively solves the problems of coarse task allocation and frequent equipment command conflicts in traditional methods. Furthermore, by dynamically optimizing the command execution order through an energy-optimal scheduling model, it significantly reduces the overall energy consumption and equipment wear of the agricultural IoT system, while ensuring the temporal coordination of environmental control and water / fertilizer management.

[0036] S5. Based on environmental control optimization instructions, obtain optimized multi-source agricultural cloud datasets, dynamically optimize the agricultural integrated management model, and obtain the agricultural optimized management model and environmental control update instructions. Existing technologies typically have weak ability to capture short-term environmental fluctuations, rely on simple heuristic rules for hyperparameter adjustment, are prone to getting trapped in local optima, lack feedback evaluation of strategy implementation effects, and the optimization direction is prone to deviating from actual needs. To solve the above problems, the specific implementation steps are as follows: S51. Based on the optimized multi-source agricultural cloud dataset, an effect indicator vector is obtained through indicator difference processing. After the environmental control optimization command has been implemented for a period of time, the optimized multi-source agricultural cloud dataset is obtained. In practice, this period can be set to 15 minutes or half an hour. Core indicators reflecting crop health or resource utilization efficiency are selected, such as the Normalized Difference Vegetation Index (NDVI) to represent growth status and soil electrical conductivity to represent salinity changes. The relative change rate is calculated by subtracting the indicator value before execution from the indicator value after execution and dividing by the indicator value before execution. This generates a vector containing indicators such as water improvement degree and nutrient absorption efficiency. The effect index vector includes dimensions such as changes in growth rate. For example, if the instruction is to increase irrigation in an arid region, and the NDVI value of the region before execution is 0.45 with low vegetation coverage, and the NDVI value collected after execution rises to 0.58, then the water improvement is approximately 28.9% (0.58 minus 0.45 divided by 0.45), indicating that the irrigation decision significantly improved crop growth. At the same time, if soil moisture sensor data shows that the soil moisture content increased from 12% to 20% after execution, the water use efficiency index can be further calculated to form a multi-dimensional effect evaluation result.

[0037] S52. Obtain the model parameters of the integrated agricultural management model. Update the model parameters using online stochastic gradient descent based on the performance indicator vector, optimized multi-source agricultural cloud dataset, and the existing model parameters. Starting with the current model parameters, calculate the prediction error based on the performance indicator vector (e.g., growth rate improvement percentage, water use efficiency) and a new round of multi-source agricultural dataset (e.g., real-time soil moisture, meteorological parameters). Then, gradually correct the model parameters by controlling the step size according to the learning rate in the opposite direction of the error gradient. Repeat the above steps to continuously reduce the deviation between the prediction results and the actual results. The final output is the incrementally optimized model parameters. For example, the current irrigation decision model predicts that a certain area needs 60 minutes of irrigation, but after execution, the performance indicators show that the crop growth rate only increases by 10%, failing to reach the expected 15%. Furthermore, the soil moisture sensor indicated that the actual water content was higher than the target value. By calculating the gradient of the loss function through online SGD, such as the deviation between the predicted value and the actual growth rate, it was found that the model underestimated the soil water retention capacity parameter. Therefore, the parameters were adjusted by a learning rate of 0.01, increasing the soil water retention coefficient from 0.75 to 0.78. At the same time, the evaporation compensation factor was fine-tuned to reduce the prediction of over-irrigation. In the next decision, the updated model will generate a more accurate 50-minute irrigation instruction, forming an iterative optimization process of execution, feedback, and optimization, and finally obtaining the updated model parameters that achieve the best effect. Among them, online SGD is an optimization algorithm based on gradient descent, designed for dynamic data environments. It incrementally updates the model parameters by processing single samples or small batches of data in real time, achieving efficient online learning, which will not be elaborated here.

[0038] S53. Obtain optimized hyperparameters through Bayesian optimization based on updated model parameters. This step first divides the updated model parameters into training, validation, and test sets. This step primarily utilizes data from the validation set for processing and optimization. The current updated model parameters are obtained from the validation set. Using the best performance on the current validation set as a benchmark, the expected performance improvement of each candidate hyperparameter combination is calculated. If the prediction performance of a parameter combination is higher than the current best, its improvement is the actual difference; if it is lower than the current best, the improvement is zero. Finally, the parameter combination with the largest expected improvement is selected as the next evaluation point, which can efficiently approximate the globally optimal hyperparameters. For example, if the current model's best accuracy on the validation set is 85%, the corresponding hyperparameter combination is a learning rate of 0.01 and a sample size of 32. A better parameter needs to be searched. Bayesian optimization first constructs a Gaussian process model based on historical evaluation data, such as 82% accuracy with a learning rate of 0.005 and 83% accuracy with a learning rate of 0.02, to predict the performance distribution of different parameter combinations. For candidate combinations... For example, with a learning rate of 0.015 and a sample size of 64, the model predicts that its performance may follow a normal distribution with a mean of 86% and a standard deviation of 2%. In this case, the EI function calculates the expected improvement of this combination: if its predicted mean exceeds the current best 85%, the improvement is 1% multiplied by the difference between 86% and 85%, and the product is multiplied by the probability that the value exceeds 85%, such as 97.7%, resulting in a higher EI value. If another combination predicts a mean of only 84%, the EI value is zero. The combination with the highest EI value is evaluated first, such as the combination with a learning rate of 0.015. If its actual measured accuracy reaches 87%, the current best is updated, and iterative optimization is performed until convergence to obtain the optimized hyperparameters. Among them, Bayesian optimization is a commonly used hyperparameter optimization model. The Gaussian process model is a probability-based nonparametric Bayesian method used to model and predict unknown functions, which will not be elaborated here. Updating model parameters is used to quickly adapt to real-time changes in data distribution and correct model predictions. Optimizing hyperparameters is used to search for the optimal hyperparameter combination globally to improve the model's generalization ability on the validation set.

[0039] S54. Based on the updated model parameters and optimized hyperparameters, cross-validation is performed to obtain the agricultural optimization management model. In this step, the updated model parameters are first combined with the optimized hyperparameters to generate a candidate model to be validated. Then, on a validation set independent of the training data, the model's classification results are statistically analyzed. The sum of the number of correctly predicted positive samples (true positives) and the number of correctly predicted negative samples (true negatives) is divided by the total number of all samples, including the number of misclassified positive samples (false positives) and the number of misclassified negative samples (false negatives), to obtain the accuracy index. A detection threshold is set, for example, by using 1.15 times the average of historical data, such as a detection threshold of 90%. If the accuracy index is greater than or equal to the detection threshold, the current model parameters and hyperparameters are used as the model hyperparameters, and the agricultural optimization management model is obtained. Otherwise, the hyperparameter combination is adjusted and re-validated until the agricultural optimization management model is obtained. For example, in an agricultural disease identification task, the updated model parameters have adapted to recent humidity changes. In this case, Bayesian optimization suggests optimizing hyperparameters, such as adjusting the learning rate from 0.01 to 0.005 and the regularization coefficient from 0.1 to 0.05. After combining these hyperparameters with the updated model parameters to generate a candidate model, it is tested on an independent validation set containing 1000 crop leaf images, of which 200 are positive disease images and 800 are negative healthy images. If the model correctly identifies 180 disease images and 760 healthy images, while mistakenly identifying 20 healthy images as disease (false positives) and 40 disease images as healthy (false negatives), then the accuracy is calculated as follows: the sum of the number of correctly identified diseases (180) and the number of correctly identified healthy images (760), divided by the total number of samples (1000), equals 94%. Since 94% is higher than the preset 90% threshold, the current model is adopted as the agricultural optimization management model.

[0040] S55. The newly acquired raw environmental physical signals and raw crop physiological images are input into the agricultural optimization management model to obtain environmental control update instructions. This invention provides interpretable physical feedback for model updates, uses online learning to achieve incremental parameter updates, performs real-time feedback evaluation, avoids getting trapped in local optima, which affects the model's analytical capabilities, and improves the accuracy of model analysis results. By using Bayesian optimization to dynamically search for hyperparameters, the efficiency of optimal parameter search and utilization is balanced, and finally, cross-validation improves the model's robustness.

[0041] Example 2: Due to the limited ability of existing technologies to capture short-term fluctuations, their tendency to get trapped in local optima, and the lack of feedback evaluation for the effectiveness of strategy implementation, please refer to [link to relevant documentation]. Figure 2 The diagram shown is a structural block diagram of the cloud computing-based smart agriculture information management system provided in this embodiment. The system includes a preprocessing module, an anomaly diagnosis module, an implementation data module, a preliminary management module, and a model optimization module.

[0042] The preprocessing module is used to preprocess the collected raw environmental physical signals and raw crop physiological images to obtain a multi-source agricultural cloud dataset. The anomaly diagnosis module is used to perform anomaly diagnosis based on multi-source agricultural cloud datasets and obtain crop health reports and anomaly data. The implementation data module is used to select crop data with abnormal data and abnormal states from crop health reports and abnormal data, and obtain agricultural implementation data sequences through fuzzy logic and genetic optimization processing; The preliminary management module is used to regulate agricultural information based on agricultural implementation data sequences and obtain environmental control optimization instructions to obtain an integrated agricultural management model. The model optimization module is used to obtain optimized multi-source agricultural cloud datasets based on environmental control optimization instructions, and to dynamically optimize the agricultural integrated management model to obtain the agricultural optimized management model and environmental control update instructions.

[0043] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, this application can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code, including but not limited to disk storage, CD-ROM, optical storage, etc.

[0044] The above embodiments provide a detailed description of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A cloud computing-based smart agriculture information management method, characterized in that, The method involves the following steps: preprocessing the collected raw environmental physical signals and raw crop physiological images to obtain a multi-source agricultural cloud dataset; Anomaly diagnosis is performed based on multi-source agricultural cloud datasets to obtain crop health reports and abnormal data; Crop data with abnormal data and abnormal states are selected from crop health reports and abnormal data, and agricultural implementation data sequences are obtained through fuzzy logic and genetic optimization. Based on the agricultural implementation data sequence, agricultural information is regulated and controlled, and environmental control optimization instructions are obtained to obtain an integrated agricultural management model; Based on environmental control optimization instructions, an optimized multi-source agricultural cloud dataset is obtained, and the agricultural integrated management model is dynamically optimized to obtain the agricultural optimized management model and environmental control update instructions.

2. The cloud computing-based smart agriculture information management method according to claim 1, characterized in that, Preprocessing of the acquired raw environmental physical signals and raw crop physiological images includes: Preliminary environmental data and preliminary image data were obtained through preliminary calibration based on the original environmental physical signals and original crop physiological images. Preliminary environmental data and preliminary image data are spatiotemporally aligned and encapsulated to obtain spatiotemporally structured data packets. Multi-source agricultural cloud datasets are obtained by storing spatiotemporal structured data packets in an agricultural cloud database.

3. The cloud computing-based smart agriculture information management method according to claim 1, characterized in that, Crop anomaly diagnosis based on multi-source agricultural cloud datasets, including: A feature extraction layer is constructed based on a multi-source agricultural cloud dataset, and temporal feature vectors and crop image depth feature vectors are obtained based on the feature extraction layer. Based on temporal features, a Kalman filter fusion layer is constructed to obtain the optimal environmental state vector and environmental covariance matrix; Anomaly diagnosis and confidence assessment are performed on the optimal environmental state vector and crop image depth feature vector to obtain crop health reports and anomaly data.

4. The cloud computing-based smart agriculture information management method according to claim 3, characterized in that, Anomaly diagnosis and confidence assessment are performed on the optimal environmental state vector and crop image depth feature vector, including: An attention layer is constructed to process the optimal environment state vector and the crop image depth feature vector using an attention mechanism, thereby obtaining enhanced joint features. Based on the enhanced joint features, a preliminary probability distribution of anomaly types is obtained through a fully connected neural network layer; Crop health reports and anomalous data are obtained by calculating covariance confidence based on the preliminary anomaly type probability distribution and environmental covariance matrix.

5. The cloud computing-based smart agriculture information management method according to claim 1, characterized in that, The agricultural implementation data sequence was obtained through fuzzy logic and genetic optimization processing, including: Based on crop data with abnormal data and abnormal states, multi-objective functions and constraints are obtained through fuzzy logic processing. The initial population is obtained by initializing the genetic population for the multi-objective function and constraints. Agricultural implementation data sequences were obtained based on the initial population through genetic optimization.

6. The cloud computing-based smart agriculture information management method according to claim 5, characterized in that, Based on the initial population, genetic optimization treatments were performed, including: Based on the initial population, fitness vectors and Pareto ranks are generated using Pareto sorting. Simulated chromosome crossover is performed on each fitness vector and Pareto rank to obtain an updated population. Agricultural implementation data sequences are generated based on Pareto solutions selected by fuzzy preferences from updated populations.

7. The cloud computing-based smart agriculture information management method according to claim 1, characterized in that, Regulating agricultural information based on agricultural implementation data sequences includes: Based on agricultural implementation data sequences, zoned and time-based control data are obtained through spatiotemporal grid decomposition. Conflict detection is performed on the time-series control data for each zone, and a conflict-free instruction set is obtained. Based on the conflict-free instruction set, environmental control optimization instructions are obtained through energy consumption scheduling optimization.

8. The cloud computing-based smart agriculture information management method according to claim 1, characterized in that, Based on environmental control optimization instructions, multi-source agricultural cloud datasets are obtained for dynamic optimization of the integrated agricultural management model, including: Based on the optimization of multi-source agricultural cloud datasets, an effect indicator vector is obtained through indicator difference processing; The model parameters of the integrated agricultural management model are obtained, and the updated model parameters are obtained through online stochastic gradient descent processing based on the performance index vector, the optimized multi-source agricultural cloud dataset, and the model parameters. Based on updating the model parameters, the optimized hyperparameters are obtained through Bayesian optimization. Based on the updated model parameters and optimized hyperparameters, cross-validation was performed to obtain the agricultural optimization management model; The newly acquired raw environmental physical signals and raw crop physiological images are input into the agricultural optimization management model to obtain environmental control update instructions.

9. A system applied to the cloud computing-based smart agriculture information management method according to any one of claims 1-8, characterized in that, The system includes: The preprocessing module is used to preprocess the collected raw environmental physical signals and raw crop physiological images to obtain a multi-source agricultural cloud dataset. The anomaly diagnosis module is used to perform anomaly diagnosis based on multi-source agricultural cloud datasets and obtain crop health reports and anomaly data. The implementation data module is used to select crop data with abnormal data and abnormal states from crop health reports and abnormal data, and obtain agricultural implementation data sequences through fuzzy logic and genetic optimization processing; The preliminary management module is used to regulate agricultural information based on agricultural implementation data sequences and obtain environmental control optimization instructions to obtain an integrated agricultural management model. The model optimization module is used to obtain optimized multi-source agricultural cloud datasets based on environmental control optimization instructions, and to dynamically optimize the agricultural integrated management model to obtain the agricultural optimized management model and environmental control update instructions.

Images