Intelligent agricultural management system and method based on digital technology

Abstract

The invention provides an intelligent agricultural management system and method based on a digital technology, and the method comprises the steps: analyzing the correlation between historical operation data and a yield change trend for the historical operation data, tracing the impact path of the operation on the yield through a time sequence analysis method, and determining key operation factors causing the symptoms of diseases and pests; generating an index tag according to the key operation factor, integrating the key operation factor and the influence degree by the index tag, binding the index tag with preset reference information, and obtaining comprehensive processing suggestion data for the current disease and pest symptom; and generating a structured retrieval result through the comprehensive processing suggestion data, and integrating the natural description of the disease and pest symptoms, the historical case records and the comprehensive processing suggestion data to obtain a complete information view presented to a terminal.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to a smart agricultural management system and method based on digital technology. Background Technology

[0002] Smart agriculture, as a crucial field of deep integration between digital technology and agricultural production, has become an important pathway to improve agricultural efficiency, ensure food security, and achieve sustainable development. In modern agriculture, massive amounts of data permeate the entire production process, including crop varieties, planting times, plot information, agricultural input usage, irrigation records, and harvest yields. The effective management of this data directly affects the accuracy and timeliness of farmers' decision-making.

[0003] While many agricultural management systems have achieved data storage and basic query capabilities, their retrieval functions often struggle to adapt to the complex and ever-changing real-world needs. When searching for information, farmers typically need to consider multiple conditions simultaneously, such as the combination of crop variety and plot number, or the correlation between agricultural input type and time period, yet they often find it difficult to quickly obtain complete and relevant information. Especially when facing sudden problems such as pests and diseases, the system struggles to automatically link historical cases and control experiences based on symptom descriptions, forcing farmers to rely on personal memory or external consultations, thus impacting decision-making efficiency.

[0004] In information retrieval within smart agricultural management systems, establishing multi-dimensional connections and semantic understanding has become a core challenge. Agricultural production data exhibits complex internal relationships; for example, the same batch of seeds may involve planting records across different plots, multiple applications of agricultural inputs, and final yield performance. These data need to be effectively correlated to form a complete view. However, due to the dynamic and diverse nature of these data relationships, the system struggles to achieve semantic annotation and deep indexing of this information. When farmers input descriptions of pest and disease symptoms hoping to find similar historical cases, the system often fails to accurately understand the description's intent and struggles to correlate symptoms with relevant records such as historical irrigation, agricultural input usage, and yield changes. This results in fragmented and low-relevance search results, failing to provide farmers with valuable references.

[0005] Therefore, how to achieve multi-dimensional correlation and semantic retrieval of agricultural production data in a smart agricultural management system, and enable farmers to quickly obtain highly relevant historical cases and related information through combined conditions or natural descriptions, has become a key issue in improving the system's practicality and farmers' decision support capabilities. Summary of the Invention

[0007] This invention provides a smart agricultural management system and method based on digital technology, mainly including: By collecting agricultural production data, including plot planting information, agricultural input usage records, irrigation operation logs, and yield change trends, a unified database is constructed. The agricultural production data is then classified and labeled using a structured storage method to obtain a preliminarily organized data set. For the initially organized data set, a multi-dimensional association model is constructed. The multi-dimensional association model achieves dynamic linking through field mapping, and associates the plot planting information with the agricultural input usage records and the irrigation operation log to form an association network containing historical case records, and determines the intrinsic relationship graph between the data. Based on the intrinsic connection graph, a semantic retrieval framework is constructed. The semantic retrieval framework adopts a text parsing mechanism to parse the input natural description of pest and disease symptoms, extract key feature words, and obtain a set of semantic features corresponding to the natural description of pest and disease symptoms. Using the semantic feature set, similar symptom descriptions are matched in the historical case records. If the matching degree is higher than a preset threshold, the corresponding agricultural input usage records and irrigation operation logs are extracted to obtain historical operation data related to the current pest and disease symptoms. Based on the historical operational data, the correlation between the historical operational data and the yield change trend is analyzed. Time series analysis is used to trace the impact path of operations on yield and identify the key operational factors that lead to pest and disease symptoms. Based on the key operational factors, index tags are generated. The index tags integrate the key operational factors and their degree of influence. The index tags are then bound to preset reference information to obtain comprehensive treatment suggestions for the current pest and disease symptoms. The comprehensive processing suggestion data is used to generate structured search results, which integrate the natural description of the pest and disease symptoms, the historical case records, and the comprehensive processing suggestion data to obtain a complete information view presented to the terminal.

[0008] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: This invention discloses a method for generating integrated pest management recommendations based on agricultural production data analysis. Addressing the problems of low efficiency in pest and disease symptom identification and treatment, and insufficient utilization of historical data in agricultural production, this method constructs a unified database and a multi-dimensional association model. It dynamically links plot planting information, agricultural input usage records, and irrigation operation logs to form a historical case association network. Furthermore, it uses a semantic retrieval framework to analyze pest and disease symptom descriptions, extracts key features, matches similar cases, traces the impact path of operations on yield, identifies key operational factors, and ultimately generates integrated treatment recommendations. By integrating historical data with current symptoms, this invention generates structured search results and a complete information view, significantly improving the accuracy and scientific rigor of pest and disease treatment, optimizing agricultural production decision-making efficiency, and providing strong support for intelligent agricultural management. Attached Figure Description

[0009] Figure 1 This is a flowchart of a smart agricultural management system and method based on digital technology according to the present invention.

[0010] Figure 2 This is a schematic diagram of a smart agricultural management system and method based on digital technology according to the present invention.

[0011] Figure 3 This is another schematic diagram of a smart agricultural management system and method based on digital technology according to the present invention. Detailed Implementation

[0012] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] like Figures 1-3 This embodiment of a smart agricultural management system and method based on digital technology may specifically include: Step S101: By collecting agricultural production data, including plot planting information, agricultural input usage records, irrigation operation logs, and yield change trends, a unified database is constructed. The agricultural production data is then classified and labeled using a structured storage method to obtain a preliminarily organized data set.

[0014] By comprehensively collecting information on plot planting, agricultural input usage records, and irrigation operation logs, an initial dataset containing various production data was compiled. Using a structured storage method, the initial dataset was categorized and labeled according to plot planting, agricultural input usage, and irrigation operations, resulting in categorized data groups. For yield change information within these categorized data groups, key fields related to the change trend were extracted to obtain a trend feature set associated with the production data. If a field value in the trend feature set exceeds a preset threshold range, the corresponding plot planting information is marked to identify key areas of focus for abnormal changes. Based on these marked key areas of focus, relevant historical data was extracted from agricultural input usage records and irrigation operation logs to obtain specific operational combinations affecting yield changes. Through correlation analysis between operational combinations and yield changes, a logistic regression model was used to process the correspondence between data to determine the main factors influencing yield changes. After identifying the main factors, a reference basis for optimization and adjustment was generated based on the plot planting information, determining the priority direction for subsequent production data collection.

[0015] By comprehensively collecting information on plot planting, agricultural input usage records, and irrigation operation logs, an initial dataset containing various types of production data can be formed. For example...

[0016] In one possible implementation, for a rice-growing plot, the initial dataset includes the planting date 2025-05-10, variety type, fertilization records such as nitrogen fertilizer application rate of 50 kg / mu, irrigation logs such as irrigation water volume of 300 cubic meters per irrigation, etc., which are derived from the integration of field sensors and manual records.

[0017] Specifically, the initial dataset is classified and labeled using a structured storage method to obtain the sorted classification data group.

[0018] For example, data can be categorized into plot planting categories (including crop growth cycles), agricultural input usage categories (including fertilizer and pesticide types and dosages), and irrigation operation categories (including irrigation time and water volume), thus facilitating subsequent queries and analysis. This categorization helps to quickly locate specific production stages and improves data management efficiency.

[0019] In one embodiment, for the yield change information in the categorized data group, key fields related to the change trend, such as historical yield data and the amount of agricultural input in the corresponding period, are extracted to obtain a trend feature set.

[0020] For example, a trend of yield decreasing from 8 tons / hectare to 6.5 tons / hectare can be extracted from three consecutive quarters of data and correlated with a field showing an increase in nitrogen fertilizer use from 40 kg / mu to 60 kg / mu. This extraction can reveal potential correlations and help identify problems early.

[0021] For example, if the amount of nitrogen fertilizer used in the trend feature set exceeds a preset threshold such as 55 kg / mu, the planting information of the plot corresponding to that field will be marked to identify the key objects of attention for abnormal changes.

[0022] Specifically, marking the plot as a high-risk area facilitates priority monitoring by management personnel. This marking mechanism can significantly reduce the risk of abnormal spread and improve production stability.

[0023] It should be noted that, based on the marked key focus areas, relevant historical data are extracted from agricultural input usage records and irrigation operation logs to obtain specific operational combinations that affect yield changes.

[0024] For example, it was found that the combination of excessive nitrogen fertilization and untimely irrigation was directly associated with reduced yields. This finding provides causal clues and supports precise intervention.

[0025] For example, by analyzing the correlation between operational combinations and changes in output, a logistic regression model can be used to process the correspondence between data and identify the main factors.

[0026] Specifically, model analysis shows that excessive nitrogen fertilizer is the main factor causing yield decline, with a probability of 85%. This analysis quantifies the degree of impact, avoiding subjective judgment bias.

[0027] In one possible implementation, after obtaining the key factors, a reference for optimization and adjustment is generated based on the planting information of the plot, such as suggesting adjusting the nitrogen fertilizer application rate to 45 kg / mu and increasing the irrigation frequency. This reference directly guides production optimization, increasing overall yield by 5%-10%.

[0028] Specifically, this involves prioritizing subsequent production data collection, such as focusing on soil nitrogen content and moisture data. This prioritization concentrates resources on key aspects, reducing collection costs and enhancing the value of the data. Through this process, the entire system achieves closed-loop management, effectively supporting the sustainable improvement of agricultural production.

[0029] Step S102: For the initially organized data set, a multi-dimensional association model is constructed. The multi-dimensional association model achieves dynamic linking through field mapping, and associates the plot planting information with the agricultural input usage records and the irrigation operation log to form an association network containing historical case records, and determines the intrinsic relationship graph between the data.

[0030] By matching records of identical plots based on plot identifier, planting time, and irrigation time, information on planted crops, agricultural input usage records, and irrigation operation logs for the same plot is obtained. Using preset field mapping rules, planted crops are associated with agricultural input types and usage times to obtain agricultural input application sequences corresponding to crop growth stages. Similarly, using the same field mapping rules, planted crops are associated with irrigation times and irrigation amounts to obtain irrigation sequences corresponding to crop growth stages. For the same plot identifier, the agricultural input application sequences and irrigation sequences are combined to form historical case records containing crop, agricultural input type, agricultural input usage, and irrigation amount. If agricultural input usage and irrigation amount appear simultaneously in historical case records, a direct association edge is established between them, generating a pair of associated network nodes. By accumulating historical case records from multiple plots, the association network is expanded by adding association edges from crop to agricultural input type and from crop to irrigation amount, determining a multi-dimensional relationship graph. A graph neural network is used to embed the association network, obtaining implicit relationship vectors between nodes to determine the strength of potential associations in the relationship graph.

[0031] For example, in agricultural production data management, by accurately matching plot identification, planting time and irrigation time, multi-source records of the same plot can be effectively integrated to ensure the accuracy of data association.

[0032] Specifically, by using the plot identifier as the primary key and combining it with timestamps, information on planted crops, records of agricultural input usage, and irrigation operation logs are merged to form a complete historical trajectory. This helps to avoid data confusion across plots and improves the reliability of subsequent analysis.

[0033] In one embodiment, a preset field mapping rule is used to connect the planted crops with the types of agricultural inputs and the time of use, thereby generating an agricultural input application sequence corresponding to the crop growth stage.

[0034] For example, for a rice-growing plot, the mapping rule can define the tillering stage as 30 days after sowing. Records of nitrogen fertilizer application at this time can be used to form sequences such as sowing stage - basal fertilizer, seedling stage - topdressing, and tillering stage - tillering-promoting fertilizer. This sequential processing clearly reflects the temporal relationship between agricultural input application and crop growth, facilitating the identification of optimal fertilization times. Similarly, by connecting irrigation time and irrigation volume using the same mapping rule, an irrigation sequence corresponding to the crop growth stages can be obtained.

[0035] For example, rice paddies need to maintain a 5-centimeter water layer during the tillering stage. The mapped sequence shows that the cumulative irrigation amount during this stage reaches 200 millimeters, while irrigation is reduced to shallow water during the heading stage. This approach makes irrigation management more targeted and supports precise water control.

[0036] Specifically, for the same plot of land, the agricultural input application sequence and irrigation sequence are combined to form a historical case record that includes crops, types of agricultural inputs, amounts of agricultural inputs used, and amounts of irrigation.

[0037] For example, historical records from one plot show that wheat yielded 550 kg / mu when 30 kg / mu of phosphorus and potassium fertilizer was applied during the jointing stage, along with 150 mm of irrigation. Such complete case studies provide direct reference for subsequent optimization and promote the accumulation of production experience.

[0038] For example, if agricultural input usage and irrigation volume appear simultaneously in historical case records, a direct association edge is established, generating a pair of associated network nodes.

[0039] In one embodiment, when 20 kg of urea application and 120 mm of irrigation occur simultaneously during the booting stage, an edge is formed connecting the two, reflecting the fertilizer-water coupling effect and enhancing the understanding of their interactive influence. By accumulating historical case records from multiple plots, the association network is expanded by adding association edges between crops and agricultural input types, and between crops and irrigation amounts, ultimately determining a multi-dimensional relationship graph.

[0040] For example, rice nodes in a network are connected to edges with various nitrogen fertilizers and different irrigation amounts, forming a radial structure. This graph can reveal the comprehensive response pattern of crops to inputs, which helps in global optimization decisions.

[0041] In one embodiment, a graph neural network is used to embed the association network to obtain the hidden relationship vector between nodes and determine the potential association strength.

[0042] For example, through multi-layer propagation, the vector similarity between urea nodes and high irrigation nodes reached 0.85, indicating a strong positive correlation. This helps to discover the implicit fertigation mechanism, improve yield prediction accuracy, and enhance resource utilization efficiency. This method significantly improves the scientific rigor of data-driven decision-making and supports sustainable agricultural production.

[0043] Step S103: Based on the intrinsic relationship graph, construct a semantic retrieval framework. The semantic retrieval framework adopts a text parsing mechanism to parse the input natural description of pest and disease symptoms, extract key feature words, and obtain a set of semantic features corresponding to the natural description of pest and disease symptoms.

[0044] Based on the input natural language description of pest and disease symptoms, a text parsing mechanism is used to remove stop words and punctuation marks, resulting in purified text. Word segmentation is then used to purify the text, obtaining candidate word sequences. Part-of-speech tagging is applied to these candidate word sequences, resulting in word sequences with part-of-speech tags. Nouns, verbs, and adjectives are selected based on these tags; words belonging to the relevant lexicon are retained, yielding key feature word groups. A pre-trained word embedding model converts these key feature word groups into vector representations, obtaining a set of individual word group vectors. Vector average pooling is used to aggregate these individual word group vector sets, resulting in a descriptive overall semantic vector. Cosine similarity is used to calculate the similarity between the overall semantic vector and node vectors in the intrinsic relationship graph; if the similarity exceeds a preset threshold, the corresponding node is retained, resulting in a semantic feature set.

[0045] For example, when processing natural language descriptions of pest and disease symptoms, the input text can be preliminarily processed through a text parsing mechanism. Suppose the user inputs "There are yellow spots on the corn leaves and the edges are withered," the system will first remove stop words such as "have" and "and," as well as punctuation marks, resulting in the purified text "Yellow spots on the corn leaves and withered edges." The purpose of this step is to reduce interference from irrelevant information and highlight the core content.

[0046] In one possible implementation, word segmentation breaks down the cleaned text into individual words, forming a sequence of candidate words such as "corn," "leaves," "yellow," "spots," "edge," and "withered." Then, through part-of-speech tagging, the system labels each word with its part of speech; for example, "corn" is labeled as a noun, and "yellow" as an adjective. This process helps in subsequently filtering out words that are meaningful in describing pests and diseases.

[0047] For example, when filtering key feature phrases, the system retains nouns, verbs, and adjectives based on part-of-speech tags and matches them against a domain-specific vocabulary list for pests and diseases. Suppose the domain vocabulary list contains words like "corn," "spots," and "wilt," these words will be retained, while "leaves" and "edges" might be removed, ultimately resulting in the key feature phrases "corn," "yellow," "spots," and "wilt." This approach allows the system to focus on descriptive content directly related to pests and diseases.

[0048] In one possible implementation, converting key feature phrases into vector representations can be achieved using a pre-trained word embedding model. Assuming the vector corresponding to "corn" reflects its crop attributes, and the vectors corresponding to "spots" and "wilt" reflect disease characteristics, these vectors can capture the semantic relationships between words. Then, through average pooling, multiple phrase vectors are aggregated into a single semantic vector, representing the semantic information of the entire description. This approach effectively integrates scattered semantic features.

[0049] For example, when calculating similarity, the system compares the overall semantic vector with the node vectors in the intrinsic relationship graph. Suppose a node in the graph represents "corn yellow spot disease," and its vector has a cosine similarity of 0.85 with the descriptive vector, which is higher than the preset threshold of 0.7. This node will then be retained in the semantic feature set. This process helps the system quickly locate potentially related disease types.

[0050] In one possible implementation, the planting data and agricultural input usage records of the plots in the historical information can be further combined with the analysis of pest and disease symptoms.

[0051] For example, historical cases of corn cultivation in a certain plot of land show that low agricultural input usage and insufficient irrigation during a certain period may have led to decreased crop resistance and increased susceptibility to yellow spot disease. By associating semantic feature sets with historical cases, the system can infer the potential link between current symptoms and historical environmental factors. This association analysis helps farmers provide more accurate judgments on the causes of diseases.

[0052] For example, regarding the correlation of irrigation operation logs, if historical records show that irrigation volume has been consistently below normal levels, it may exacerbate disease symptoms. The system can combine this information with the current symptom description to generate a more comprehensive analysis report. This multi-dimensional correlation approach can provide data support for subsequent disease prevention and control, while also improving farmers' understanding of and ability to respond to disease causes.

[0053] Step S104: Using the semantic feature set, match similar symptom descriptions in the historical case records. If the matching degree is higher than a preset threshold, extract the corresponding agricultural input usage records and irrigation operation logs to obtain historical operation data related to the current pest and disease symptoms.

[0054] By comparing semantic features with symptom descriptions, records similar to the current pest and disease symptoms are selected from historical cases to obtain preliminary matching results. If the matching degree of the preliminary matching results is higher than a preset threshold, relevant operation records of agricultural input use and irrigation operations are extracted to obtain a historical data set associated with the current symptoms. Based on the agricultural input use records in the historical data set, the correspondence between different types of agricultural inputs and pest and disease symptoms is analyzed to determine potential agricultural input application schemes for the current symptoms. Using irrigation operation logs in the historical data set, combined with the severity of the current symptoms, the appropriate irrigation frequency and water volume configuration are determined to obtain a preliminary operation adjustment scheme. By comprehensively comparing the potential agricultural input application schemes and operation adjustment schemes, the historical operation combination that best matches the current symptoms is selected to obtain optimized operation guidance data. Based on the optimized operation guidance data and combined with the real-time changes of the current symptoms, the priority of agricultural input use and irrigation operations is dynamically adjusted to determine the final execution plan. Based on the final execution plan, automated operation instructions for pest and disease symptoms are generated and output to the relevant agricultural equipment system to complete the operation deployment.

[0055] For example.

[0056] In one possible implementation, highly similar pest and disease records can be quickly filtered out by comparing a set of semantic features with a historical case library.

[0057] Specifically, the overall semantic vector of the current symptom is compared with the pre-stored vectors in the case database using cosine similarity calculation. If the similarity exceeds 0.85, it is considered a preliminary match, thus obtaining a set of potentially relevant historical data. This high threshold setting helps avoid interference from low-relevance cases and improves the accuracy of subsequent analysis.

[0058] In one embodiment, for the preliminary matching results, records of agricultural input usage and irrigation operations are further extracted.

[0059] For example, suppose the current symptom is brown spots and curling on rice leaves. By comparing three historical records, the similarity is found to be higher than 0.88. From these, data showing that fungicide A was used in combination with specific irrigation adjustments can be extracted. This set of historical data provides a reliable basis for subsequent program development, avoiding blind attempts from scratch.

[0060] Specifically, when analyzing the relationship between agricultural input types and symptoms based on historical data sets, the alleviating effects of different agricultural inputs on similar symptoms can be statistically analyzed.

[0061] For example, in the screened historical records, fungicide A appeared most frequently, with a symptom improvement rate of 90%, while fungicide B only achieved 60%. Therefore, fungicide A was identified as the potential first-line agricultural input for addressing the current brown spot symptoms. This statistically driven analysis significantly improves the targeting and success rate of the treatment plan.

[0062] For example, when making a judgment by combining irrigation operation logs with the current severity of symptoms.

[0063] It should be noted that if the severity of symptoms is assessed as moderate, irrigation adjustment records from similar historical cases should be referenced.

[0064] For example, historical data shows that under moderate brown spot symptoms, adjusting the irrigation frequency from once a day to once every other day, with a 20% reduction in water volume, resulted in better symptom control. This leads to a preliminary operational adjustment plan, such as recommending that the water volume be configured to 80% of the original plan for the current cycle to prevent excessive moisture from exacerbating the disease's spread.

[0065] In one possible implementation, the best-matching historical operational combination is selected by comprehensively comparing potential agricultural input solutions and operational adjustment solutions.

[0066] For example, the combination of fungicide A and a 20% reduction in water usage historically showed the highest success rate (95%), and was therefore selected as the optimized operational guidance data. This comprehensive comparison ensures the synergistic effect of agricultural inputs and irrigation, maximizing the overall control effect.

[0067] Specifically, the priority is dynamically adjusted based on the optimized guidance data and real-time changes in symptoms.

[0068] For example, if monitoring shows that the affected area expands by 10% within 24 hours, the application of fungicide is prioritized, followed by adjustments to irrigation, thus determining the final implementation plan. This dynamic mechanism makes the plan more adaptable and effectively addresses the uncertainty of disease progression.

[0069] For example, the system ultimately generates automated operation instructions based on the execution plan, such as sending instructions to spraying equipment to "apply fungicide A at a dosage of 0.5 liters per acre" and to irrigation systems to "use water at 80% standard level every 48 hours," which are then directly deployed. This automated output significantly reduces human intervention errors and improves the timeliness and consistency of agricultural operations.

[0070] Step S105: For the historical operation data, analyze the correlation between the historical operation data and the yield change trend, use time series analysis to trace the impact path of operation on yield, and determine the key operation factors that lead to pest and disease symptoms.

[0071] A joint time series curve was plotted based on historical operational data and yield change trends. Lag correlation calculations were performed on the joint time series curve using time series analysis methods to obtain the lag correlation sequence between operations and yield. The set of operational change points within a significant lag window was determined based on the lag correlation sequence. Operational data subsequences within corresponding time periods were extracted from the set of operational change points. Granger causality tests were used to determine the causal direction of the operational data subsequences on yield changes, resulting in a causal operational subset. Symptom-related operational sequences were identified by aligning the causal operational subset with the occurrence time of pest and disease symptoms. Key operational factor groups were obtained by clustering the symptom-related operational sequences.

[0072] For example, in agricultural production management, the analysis of historical operational data and yield trends can be visually reflected by plotting a joint time series curve. A joint time series curve visualizes operational data such as agricultural input use and irrigation frequency alongside crop yield data for the corresponding time period on the same time axis. For instance, in a rice-growing region, data on fertilizer application and yield per hectare were recorded for the past 12 months. Fertilizer application increased by 20% and 15% in the 3rd and 7th months, respectively, while yield increased by 10% and 8% in the 5th and 9th months, respectively. The curve can initially show a delayed correlation between operational data and yield.

[0073] For example, lag correlation calculation can be understood as identifying the time delay in the impact of operational changes on output through time series analysis methods.

[0074] In one possible implementation, assuming that yield changes typically become apparent two months after an increase in fertilizer application, the lagged correlation sequence would show the highest correlation within a two-month lag window. Based on this, a set of operational changes within a significant lag window can be identified; for example, an incremental fertilizer application in the third month would be marked as a key change.

[0075] For example, for extracting subsequences of operational data.

[0076] Specifically, data on fertilizer application and irrigation water volume can be extracted from the data of the month before and after the third month to form a subset dataset for subsequent causal analysis. The Granger causality test is then used to determine whether these operations truly drove changes in yield. For example, the test results might show that increased fertilizer application was a significant cause of yield improvement, while changes in irrigation water volume had little impact, thus identifying a subset of causal operations.

[0077] For example, in determining symptom-associated operation sequences, a subset of causal operations can be aligned with the timing of pest and disease symptom appearance. Suppose that a certain leaf spot disease symptom appears simultaneously with a yield increase in the 5th month. Matching reveals that the increased fertilizer application in the 3rd month may be related to the symptom, thus forming a symptom-associated operation sequence. For sequence clustering, similar operation patterns can be categorized into key operational factor groups; for example, high fertilizer application and high incidence of leaf spot disease are grouped together.

[0078] For example, the above analysis can provide data support for subsequent agricultural input use and pest and disease control.

[0079] It should be noted that combining time series analysis and causal testing helps to accurately identify core operations affecting yield, while clustering of symptom-related operational sequences provides targeted references for pest and disease management. These methods work together to optimize agricultural operational strategies and improve crop yield and pest and disease control efficiency.

[0080] Step S106: Generate index tags based on the key operational factors. The index tags integrate the key operational factors and their degree of influence. Bind the index tags to preset reference information to obtain comprehensive treatment suggestions for the current pest and disease symptoms.

[0081] Obtain symptom description data based on current pest and disease symptoms. Extract key operational factors from the symptom description data. Determine the impact degree of each key operational factor. Generate index labels using the key operational factors and their impact degree values. Bind and match the index labels with preset reference information. Obtain comprehensive treatment suggestion data from the binding and matching results. Use a random forest algorithm to classify and sort the comprehensive treatment suggestion data to obtain priority treatment suggestions.

[0082] For example, in the field of agricultural pest and disease management, the analysis and treatment of pest and disease symptoms is a crucial step. Based on the acquisition of symptom description data, specific manifestations such as yellowing leaves, deformed fruits, or wilting plants can be collected through field observations and farmer feedback. Taking a rice-growing area as an example, farmers found that rice leaves were yellowing, accompanied by a few insect bite marks. This descriptive data provided the basis for subsequent analysis.

[0083] Specifically, when extracting key operational factors from symptom description data, we can focus on the operational steps that may lead to the symptoms. For example, in the rice case mentioned above, data analysis revealed that recent fertilizer application was excessive and irrigation frequency was insufficient. These two operational factors may be related to leaf yellowing. Excessive fertilizer may lead to soil salinity accumulation, while insufficient irrigation exacerbates crop water stress; the combined effect of these two factors induces the symptoms.

[0084] For example, when determining the numerical impact level, historical data comparison and expert evaluation can be used to define the impact of fertilizer application as 70% and the impact of irrigation frequency as 30%. This quantification method facilitates the subsequent generation of index tags, such as using "fertilizer over-70%" and "irrigation under-30%" as tags, and binding and matching them with the treatment plans in the preset reference information database to quickly find corresponding suggestions for adjusting fertilizer application and optimizing irrigation plans.

[0085] Specifically, after binding and matching, comprehensive treatment suggestions are obtained, which may include options such as reducing fertilizer use to 50 kg per acre or increasing irrigation frequency to twice a week. Such suggestions cover the adjustment directions of different operational factors, providing diverse options for subsequent optimization.

[0086] For example, when using the random forest algorithm to classify and sort suggestion data, the suggestion to reduce fertilizer use can be prioritized based on the implementation effects of each suggestion in historical cases, because its impact is higher, and historical data shows that adjusting fertilizer use significantly improves leaf yellowing symptoms. Prioritizing suggestions thus makes them more targeted and can quickly alleviate the current problem.

[0087] Specifically, in the cases mentioned above, implementing the priority treatment recommendations can bring additional benefits. For example, reducing fertilizer use not only alleviates pest and disease symptoms but also lowers production costs and environmental pollution risks. Optimizing irrigation frequency helps improve the overall growth of crops, creating a multi-faceted positive feedback loop. This logical progression from core solutions to extended solutions ensures the comprehensiveness and sustainability of the treatment measures.

[0088] Step S107: Using the comprehensive processing suggestion data, generate structured search results, integrate the natural description of the pest and disease symptoms, the historical case records, and the comprehensive processing suggestion data to obtain a complete information view presented to the terminal.

[0089] Key symptom keywords and corresponding case identifiers are extracted from natural descriptions of pest and disease symptoms and historical case records. Based on the extracted key symptom keywords, similar case records are matched against a pre-established pest and disease case database to obtain associated historical case records. These historical case records are then linked to corresponding comprehensive treatment suggestion data to determine a preliminary set of treatment suggestions. Based on the preliminary set of treatment suggestions and the natural descriptions of pest and disease symptoms, cosine similarity is used to calculate the text matching degree between the symptom descriptions and the suggestion data, determining the treatment suggestion sequence after matching degree ranking. If the similarity of the first suggestion in the ranked treatment suggestion sequence is higher than a preset threshold, that suggestion is determined as a priority treatment suggestion. Using the priority treatment suggestions and associated historical case records, a structured data object containing natural symptom descriptions, case records, and priority treatment suggestions is constructed. Based on the constructed structured data object, a complete information view in a unified format is generated and presented to the terminal.

[0090] For example, in the process of treating pest and disease symptoms, extracting key symptom keywords and case identifiers through natural descriptions and historical case records can effectively focus on the core problem. Suppose a crop exhibits yellowing leaves and withered edges. The farmer's natural description is "leaves are yellowing, edges are withered, and it recently rained heavily." Combining this with historical case records, the keywords "yellowing leaves," "withered edges," and "after rain" can be extracted and linked to the case identifier A-2022-001. This process relies on semantic analysis of the descriptive content to ensure that the extracted keywords accurately reflect the symptom characteristics.

[0091] For example, when matching similar case records in the case database, multiple historical records can be retrieved using the keywords "leaf yellowing" and "after rain." One record shows that similar symptoms are caused by fungal infection, and the treatment recommendation is to spray a specific fungicide. This matching method relies on the completeness and classification accuracy of the cases in the database to ensure that the retrieved records are highly relevant to the current symptoms.

[0092] For example, in forming the initial set of treatment suggestions, suppose three historical cases are matched, suggesting the use of disinfectant A, disinfectant B, and improved drainage, respectively. Combining the natural description of symptoms, the text matching degree is calculated using cosine similarity. The results show that the matching degree between the suggestion and description for disinfectant A is 0.85, for disinfectant B it is 0.75, and for improved drainage it is 0.6. If the preset threshold is 0.8, then disinfectant A is determined as the priority treatment suggestion. This method can filter out the solution that best fits the current situation from multiple suggestions.

[0093] For example, when constructing structured data objects, the symptom description "yellowing leaves with dry edges," case record A-2022-001, and the priority recommendation "use fungicide A" are integrated into a unified data unit. This structured approach facilitates subsequent information presentation and traceability, ensuring the organization and completeness of the data.

[0094] For example, when generating a complete information view, structured data objects can be converted into a terminal-friendly format, displayed as "Symptoms: Leaf yellowing, edge wilting; Related case: A-2022-001; Recommendation: Use fungicide A first, spray once every 7 days." This view is intuitive and clear, making it easy for farmers to quickly understand and implement. Through the coordinated efforts of these multiple stages, from symptom extraction to recommendation generation and information presentation, a closed-loop processing flow is formed, which helps improve the targeting and efficiency of pest and disease treatment, while reducing the risk of misjudgment and providing reliable support for agricultural production.

[0095] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A smart agricultural management system and method based on digital technology, characterized in that, The method includes: By collecting agricultural production data, including plot planting information, agricultural input usage records, irrigation operation logs, and yield change trends, a unified database is constructed. The agricultural production data is then classified and labeled using a structured storage method to obtain a preliminarily organized data set. For the initially organized data set, a multi-dimensional association model is constructed. The multi-dimensional association model achieves dynamic linking through field mapping, and associates the plot planting information with the agricultural input usage records and the irrigation operation log to form an association network containing historical case records, and determines the intrinsic relationship graph between the data. Based on the intrinsic connection graph, a semantic retrieval framework is constructed. The semantic retrieval framework adopts a text parsing mechanism to parse the input natural description of pest and disease symptoms, extract key feature words, and obtain a set of semantic features corresponding to the natural description of pest and disease symptoms. Using the semantic feature set, similar symptom descriptions are matched in the historical case records. If the matching degree is higher than a preset threshold, the corresponding agricultural input usage records and irrigation operation logs are extracted to obtain historical operation data related to the current pest and disease symptoms. Based on the historical operational data, the correlation between the historical operational data and the yield change trend is analyzed. Time series analysis is used to trace the impact path of operations on yield and identify the key operational factors that lead to pest and disease symptoms. Based on the key operational factors, index tags are generated. The index tags integrate the key operational factors and their degree of influence. The index tags are then bound to preset reference information to obtain comprehensive treatment suggestions for the current pest and disease symptoms. The comprehensive processing suggestion data is used to generate structured search results, which integrate the natural description of the pest and disease symptoms, the historical case records, and the comprehensive processing suggestion data to obtain a complete information view presented to the terminal.

2. The intelligent agricultural management system and method based on digital technology according to claim 1, characterized in that, The process involves collecting agricultural production data, including plot planting information, agricultural input usage records, irrigation operation logs, and yield change trends, to construct a unified database. The agricultural production data is then categorized and labeled using a structured storage method to obtain a preliminarily organized dataset, including: By comprehensively collecting information on plot planting, records of agricultural input usage, and irrigation operation logs, an initial dataset containing various types of production data was compiled. Using a structured storage method, the initial dataset is classified and labeled according to categories such as plot planting, agricultural input use, and irrigation operations, resulting in sorted classified data groups; For production change information in categorized data groups, extract key fields related to the change trend to obtain a set of trend features associated with production data; If the value of a certain field in the trend feature set exceeds the preset threshold range, the planting information of the plot corresponding to that field will be marked to identify the key objects of attention for abnormal changes. Based on the marked key focus areas, relevant historical data are extracted from agricultural input usage records and irrigation operation logs to obtain specific operational combinations that affect yield changes; By analyzing the correlation between operational combinations and output changes, a logistic regression model is used to process the correspondence between data and identify the main factors affecting output changes. After obtaining the main factors, the reference basis for optimizing and adjusting the planting information of the plots is used to determine the priority direction for subsequent production data collection.

3. The intelligent agricultural management system and method based on digital technology according to claim 1, characterized in that, For the initially organized dataset, a multi-dimensional association model is constructed. This model achieves dynamic linking through field mapping, associating the plot planting information with the agricultural input usage records and the irrigation operation logs to form an association network containing historical case records. This determines the intrinsic relationship graph between the data, including: By matching the same plot records with the plot identifier and planting time and irrigation time, information on the crops planted, agricultural input usage records and irrigation operation logs for the same plot can be obtained. By using preset field mapping rules, the crops planted are associated with the types and times of agricultural inputs used, thus obtaining the agricultural input application sequence corresponding to the crop growth stages. Using the same field mapping rules, the planted crops are associated with irrigation time and irrigation amount to obtain the irrigation sequence corresponding to the crop growth stage; For the same plot of land, combine agricultural input application sequences and irrigation sequences to form a historical case record that includes crops, types of agricultural inputs, amounts of agricultural inputs used, and amounts of irrigation. If agricultural input usage and irrigation volume appear simultaneously in historical case records, a direct association edge is established between the two, generating a pair of associated network nodes. By accumulating historical case records of multiple plots, we expand the connection network, add connection edges between crops and agricultural input types and between crops and irrigation volume, and determine a multi-dimensional connection graph. A graph neural network is used to embed the network into the association network, obtain the hidden relationship vector between nodes, and determine the strength of potential associations in the association graph.

4. The intelligent agricultural management system and method based on digital technology according to claim 1, characterized in that, The semantic retrieval framework is constructed based on the intrinsic relationship graph. This framework employs a text parsing mechanism to analyze the input natural descriptions of pest and disease symptoms, extract key feature phrases, and obtain a set of semantic features corresponding to the natural descriptions of pest and disease symptoms, including: Based on the input natural language description of pest and disease symptoms, a text parsing mechanism is used to remove stop words and punctuation marks to obtain purified text; Text is purified through word segmentation to obtain candidate word sequences; Candidate word sequences are labeled using part-of-speech tagging to obtain word sequences with part-of-speech tags; Based on part-of-speech tags, nouns, verbs, and adjectives are filtered out. If a word belongs to a lexicon related to pests and diseases, it is retained to obtain key feature word groups. By using a pre-trained word embedding model, key feature word groups are converted into vector representations, and a set of individual word group vectors is obtained. Vector average pooling is used to aggregate the vector sets of individual word groups to obtain a semantic vector describing the whole; The similarity between the overall semantic vector and the node vectors in the intrinsic relationship graph is calculated by cosine similarity. If the similarity is higher than a preset threshold, the corresponding node is retained, and a semantic feature set is obtained.

5. The intelligent agricultural management system and method based on digital technology according to claim 1, characterized in that, The process involves matching similar symptom descriptions in historical case records using the semantic feature set. If the matching degree is higher than a preset threshold, the corresponding agricultural input usage records and irrigation operation logs are extracted to obtain historical operation data related to the current pest and disease symptoms, including: By comparing semantic features with symptom descriptions, records similar to the current pest and disease symptoms are selected from historical cases to obtain preliminary matching results; If the matching degree of the preliminary matching result is higher than the preset threshold, the relevant operation records of agricultural input use and irrigation operation are extracted to obtain a set of historical data associated with the current symptoms; Based on the agricultural input usage records in the historical data set, analyze the correspondence between different types of agricultural inputs and the symptoms of pests and diseases, and determine potential agricultural input application solutions for the current symptoms; By using irrigation operation logs from historical datasets and considering the severity of current symptoms, we can determine the appropriate irrigation frequency and water volume configuration to obtain a preliminary operation adjustment plan. By comprehensively comparing potential agricultural input application schemes and operational adjustment schemes, the historical operational combinations that best match the current symptoms are selected to obtain optimized operational guidance data; Based on the optimized operational guidance data and combined with the real-time changes in the current symptoms, the priority of agricultural input use and irrigation operations is dynamically adjusted to determine the final implementation plan; Based on the final execution plan, automated operation instructions targeting the symptoms of pests and diseases are generated and output to the relevant agricultural equipment systems to complete the operation deployment.

6. The intelligent agricultural management system and method based on digital technology according to claim 1, characterized in that, The process involves analyzing the correlation between historical operational data and yield change trends, using time series analysis to trace the impact path of operations on yield, and identifying key operational factors leading to pest and disease symptoms, including: A combined time series curve was plotted based on historical operating data and production change trends. The lag correlation sequence between operation and output is obtained by performing lag correlation calculation on the joint time series curve using time series analysis methods. Determine the set of operational change points within a significant lag window based on the lag correlation sequence; Extract subsequences of operational data within the corresponding time period from the set of operational change points; Granger causality test is used to determine the causal direction of the operation data subsequence on the change in output, and the causal operation subset is obtained; The sequence of symptom-related operations is determined by aligning subsets of causal operations with the timing of the occurrence of pest and disease symptoms. Key operational factor groups were obtained by clustering symptom-related operational sequences.

7. The intelligent agricultural management system and method based on digital technology according to claim 1, characterized in that, The process involves generating index tags based on the key operational factors, integrating the key operational factors with their degree of influence, and binding the index tags with preset reference information to obtain comprehensive treatment suggestions data for the current pest and disease symptoms, including: Obtain symptom description data based on the current symptoms of pests and diseases; Extracting key operational factors from symptom description data; Determine the numerical extent of impact for key operational factors; Index labels are generated based on the numerical values ​​of key operational factors and their degree of influence. Bind and match index tags with preset reference information; Obtain comprehensive processing suggestion data from the binding and matching results; The random forest algorithm is used to classify and sort the comprehensive processing suggestion data to obtain priority processing suggestions.

8. The intelligent agricultural management system and method based on digital technology according to claim 1, characterized in that, The process involves generating structured search results using the comprehensive processing suggestion data. This integrates the natural descriptions of pest and disease symptoms, historical case records, and the comprehensive processing suggestion data to obtain a complete information view presented to the terminal, including: Key symptom keywords and corresponding case identifiers are extracted from natural descriptions of pest and disease symptoms and historical case records. Based on the extracted key symptom keywords, similar case records are matched in a pre-established pest and disease case database to obtain related historical case records; By matching historical case records and associating them with corresponding comprehensive processing suggestion data, a preliminary set of processing suggestions is determined. Based on the preliminary set of treatment suggestions and the natural descriptions of pest and disease symptoms, cosine similarity is used to calculate the text matching degree between the symptom descriptions and the suggestion data, and the treatment suggestion sequence after matching degree ranking is determined. If the similarity of the first suggestion in the sorted processing suggestion sequence is higher than the preset threshold, then the suggestion is determined to be the priority processing suggestion; By prioritizing suggestions and associated historical case records, a structured data object is constructed that includes natural descriptions of symptoms, case records, and prioritizing suggestions; Based on the constructed structured data objects, a complete information view in a unified format is generated and presented to the terminal.

Images