Precise crop planting decision-making system based on multi-modal data fusion and artificial intelligence

Abstract

The invention relates to the technical field of agricultural planting management, in particular to a crop precise planting decision system based on multi-modal data fusion and artificial intelligence, which comprises a multi-modal data acquisition module, a data fusion and processing module, an artificial intelligence decision engine, a strategy execution module and a self-learning optimization module, the multi-modal data acquisition module acquires environmental crop farming data; the data fusion and processing module generates a fusion data set with unified space-time labels and establishes a plant electronic file; the artificial intelligence decision engine simulates growth through a crop growth digital twinborn model, intelligently diagnoses deviation and generates a precise decision; the strategy execution module drives the automation equipment to execute a decision; the self-learning optimization module is based on a feedback iterative optimization model and an agronomic rule base. The method improves the decision scientificity and precision, reduces the resource waste, guarantees the crop yield and quality, is suitable for different planting scenes, and meets the precise and large-scale planting demands.

Description

Technical Field

[0001] This invention relates to the field of agricultural planting management technology, specifically to a crop precision planting decision system based on multimodal data fusion and artificial intelligence. Background Technology

[0002] In the current field of crop planting and management, traditional planting models rely heavily on farmers' experience to carry out agricultural operations. They lack systematic integration and analysis of multi-dimensional data on environmental crops and agricultural operations, which can easily lead to resource waste due to improper application of water and fertilizer and delayed detection of pests and diseases. Furthermore, it is difficult to ensure stable crop yield and quality.

[0003] Existing planting management systems generally suffer from a single data collection dimension, focusing only on environmental parameters or crop phenotypic data. They fail to effectively integrate multimodal data, resulting in heterogeneous data that is spatially and temporally misaligned and of varying quality. This makes it difficult to form a dataset that comprehensively reflects the crop growth status and thus cannot provide sufficient support for decision-making.

[0004] Meanwhile, existing systems lack the ability to accurately simulate crop growth based on crop physiological and ecological mechanisms, making it difficult to accurately predict crop growth trends. The diagnosis of growth deviations is mostly limited to the identification of surface phenomena, failing to delve into the causes. Decision generation relies heavily on fixed agronomic rules, lacking flexibility and making it difficult to adapt to the characteristics and differences of different planting areas and crop varieties.

[0005] In addition, most systems have not established a closed-loop optimization mechanism for decision-making, execution, and feedback. The decision-making effect cannot be iteratively optimized through actual execution data, resulting in poor long-term applicability of the system. Furthermore, the connection between automated execution and decision-making in some systems is not smooth, requiring a large amount of manual intervention, which leads to low operational efficiency and makes it difficult to meet the needs of large-scale and precision planting.

[0006] Therefore, a crop precision planting decision system based on multimodal data fusion and artificial intelligence is proposed to address the above problems. Summary of the Invention

[0007] The purpose of this invention is to provide a crop precision planting decision system based on multimodal data fusion and artificial intelligence to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: A precision crop planting decision-making system based on multimodal data fusion and artificial intelligence includes: The multimodal data acquisition module is used to collect crop growth environment data, crop data, and agricultural operation data in real time through sensing devices, image acquisition devices, and user terminals deployed in the planting area. The data fusion and processing module communicates with the multimodal data acquisition module and is used to perform spatiotemporal alignment, cleaning and fusion processing on the heterogeneous data acquired by the multimodal data acquisition module to generate a fusion dataset with unified spatiotemporal labels and to establish a dynamically updated electronic archive for each plant unit. The artificial intelligence decision engine communicates and connects with the data fusion and processing module, including: A crop growth digital twin model, built based on crop physiological and ecological mechanisms and historical fusion data, is used to simulate the crop growth process; The intelligent diagnostic unit is used to input the fused dataset into the crop growth digital twin model, and diagnose crop growth deviations and identify the causes by comparing the model's predicted state with the actual state. The decision generation unit is used to generate precise agricultural operation decisions based on the diagnostic results of the intelligent diagnostic unit, by calling the agronomic rule base and optimization objectives, and through simulation. The strategy execution module communicates with the artificial intelligence decision engine to parse precision agricultural operation decisions into control commands and drive automated agricultural equipment to execute them. The self-learning optimization module communicates with the multimodal data acquisition module and the artificial intelligence decision engine respectively. It is used to collect feedback data after the strategy execution module is executed, and to iteratively optimize the crop growth digital twin model and agronomic rule base through machine learning algorithms based on the feedback data, forming a closed-loop decision.

[0009] As a preferred option, the multimodal data acquisition module includes: The multi-source data sensing unit is used to simultaneously acquire raw multi-source datasets containing environmental physical parameters, crop phenotypic images, and manual operation records through sensor arrays, multispectral image acquisition devices, and farmers' mobile terminals deployed in the planting area. The data preprocessing unit is used to perform preliminary quality control on the original multi-source dataset. This includes calibrating sensor readings and removing outliers for environmental physical parameters, normalizing illumination and segmenting backgrounds for crop phenotypic images, and performing text structured parsing on manual operation records to generate a preprocessed multi-source dataset. The spatiotemporal alignment unit is used to synchronize timestamps and register spatial coordinates of preprocessed multi-source datasets based on geographic information system rasters and crop growth time series, mapping data from different sources to a unified spatiotemporal reference framework to form a spatiotemporal aligned dataset. The data cleaning unit is used to perform integrity checks on the spatiotemporally aligned dataset, employs an adaptive filtering algorithm to suppress noise in high-frequency collected data, and interpolates and reconstructs missing data based on crop phenological characteristics to generate a high-quality cleaned dataset. The multimodal fusion unit is used to fuse environmental physical parameters, crop phenotypic features and agricultural operation records in a high-quality clean dataset at the feature level. By constructing a spatiotemporal correlation matrix with plants as the unit, a fused dataset with unified spatiotemporal labels is generated. The electronic archive building unit is used to create and dynamically update a digital archive of the entire life cycle for each individual plant unit based on a fusion dataset. This archive integrates plant genetic information, real-time growth indicators, environmental response history, and agricultural intervention records.

[0010] As a preferred option, the data fusion and processing module includes: The spatiotemporal alignment unit is used to receive heterogeneous data collected by the multimodal data acquisition module. Based on the geographic information system raster and crop growth time series, it performs timestamp synchronization and spatial coordinate registration on the heterogeneous data to generate spatiotemporal aligned data. The data cleaning unit is used to perform integrity checks on spatiotemporally aligned data, employs an adaptive filtering algorithm to suppress noise in high-frequency acquired data, and interpolates and reconstructs missing data based on crop phenological characteristics to generate high-quality cleaned data. The multimodal fusion unit is used to perform feature-level fusion on high-quality cleaned data. By constructing a spatiotemporal correlation matrix based on plants, it integrates environmental physical parameters, crop phenotypic features and agricultural operation records to generate a fusion dataset with unified spatiotemporal labels. The electronic archive building unit is used to create and dynamically update a digital archive of the entire life cycle for each individual plant unit based on a fusion dataset. The digital archive integrates plant genetic information, real-time growth indicators, environmental response history and agricultural intervention records.

[0011] As a preferred approach, crop growth digital twin models are constructed based on crop physiological and ecological mechanisms and historical fusion data to simulate crop growth processes, including: The model initialization unit is used to construct a theoretical model framework for crop growth processes based on crop physiological and ecological mechanisms. This theoretical model framework includes key physiological processes such as photosynthesis, respiration, and nutrient absorption and distribution. The parameter calibration unit is used to input historical fusion data into the theoretical model framework, calibrate the physiological parameters in the theoretical model framework through parameter optimization algorithms, and generate a parameter-calibrated crop growth model. The state simulation unit is used to input the real-time fused dataset into the calibrated crop growth model, simulate the crop growth process under given environmental conditions through dynamic simulation algorithms, and output the simulated value of the crop growth state. The model validation unit is used to compare the simulated values ​​of crop growth status with the actual observed values, adjust the model parameters or structure based on the error evaluation index, and generate a validated and optimized digital twin model of crop growth.

[0012] As a preferred embodiment, the intelligent diagnostic unit is used to input the fused dataset into the crop growth digital twin model, and by comparing the model's predicted state with the actual state, diagnose crop growth deviations and identify their causes, including: The state prediction subunit is used to input the fused dataset into the crop growth digital twin model, simulate the crop growth process under the current environmental conditions, and output the predicted value of the crop growth state. The state comparison subunit, which is connected to the state prediction subunit, is used to perform difference analysis between the predicted values ​​of crop growth status and the actual observation values ​​in the fused dataset, and generate a crop growth status deviation vector. The deviation diagnosis subunit is connected in communication with the state comparison subunit. It is used to identify potential causes of growth deviation based on the crop growth state deviation vector and through causal reasoning algorithms. The cause output subunit communicates with the deviation diagnosis subunit and is used to output diagnostic results, including the growth deviation type and a list of causative factors.

[0013] As a preferred embodiment, the decision generation unit is used to generate precise agricultural operation decisions based on the diagnostic results of the intelligent diagnostic unit, by invoking the agronomic rule base and optimization objectives, and through simulation, including: The diagnostic result parsing subunit is used to receive the diagnostic results output by the intelligent diagnostic unit, and to parse the growth deviation type and causative factor list in the diagnostic results to generate structured diagnostic information. The rule retrieval subunit is connected to the diagnostic result parsing subunit. It is used to retrieve agronomic rules that match the growth deviation type and causative factors from the agronomic rule base based on structured diagnostic information, and generate a candidate agronomic rule set. The scheme generation subunit communicates with the rule retrieval subunit and is used to generate multiple candidate agricultural operation schemes by combining the candidate agronomic rule set and the preset optimization target. The simulation evaluation subunit is connected to the scheme generation subunit. It is used to input each candidate agricultural operation scheme into the crop growth digital twin model, and predict the crop growth status after each scheme is executed through simulation to generate scheme evaluation results. The decision output subunit communicates with the simulation evaluation subunit and is used to select the optimal agricultural operation plan based on the scheme evaluation results and the optimization objective, and output precise agricultural operation decisions.

[0014] As a preferred option, the strategy execution module includes: The decision parsing subunit is used to receive the precision agricultural operation decisions generated by the artificial intelligence decision engine and parse the precision agricultural operation decisions into an atomic sequence of agricultural operation actions; The instruction arrangement subunit, which is connected to the decision analysis subunit, is used to arrange the atomized agricultural operation action sequence according to the preset operation logic to generate a set of standardized operation instructions that are independent of equipment. The equipment adapter subunit communicates with the instruction orchestration subunit and is used to convert standardized operation instruction sets into equipment control instructions that match the target automated agricultural equipment. The execution drive subunit is connected to the device adapter subunit to send device control commands to the corresponding automated agricultural equipment and drive the automated agricultural equipment to perform precision agricultural operations. The status monitoring subunit is connected to the execution drive subunit and is used to collect execution status data of automated agricultural equipment in real time and feed the execution status data back to the self-learning optimization module.

[0015] As a preferred option, the self-learning optimization module includes: The feedback data acquisition subunit is used to collect feedback data after the strategy execution module has executed the strategy execution module. The feedback data includes crop growth status data, environmental response data, and agricultural operation execution effect data. The effect evaluation subunit is connected to the feedback data acquisition subunit and is used to evaluate the actual implementation effect of precision agricultural operation decisions based on feedback data and generate a decision effect evaluation report. The model optimization subunit communicates with the effect evaluation subunit and is used to adjust the parameters and optimize the structure of the crop growth digital twin model based on the decision effect evaluation report and through machine learning algorithms to generate an optimized crop growth digital twin model. The rule base optimization subunit communicates with the effect evaluation subunit and is used to update the rules and adjust the weights of the agronomic rule base based on the decision effect evaluation report and through reinforcement learning algorithms to generate an optimized agronomic rule base. The closed-loop control subunit communicates with the model optimization subunit and the rule base optimization subunit to deploy the optimized crop growth digital twin model and the optimized agronomic rule base to the artificial intelligence decision engine, thereby completing the closed-loop decision optimization.

[0016] As can be seen from the technical solution provided by the present invention above, the crop precision planting decision system based on multimodal data fusion and artificial intelligence provided by the present invention has the following beneficial effects: To enhance the value of data utilization, multi-modal data collection and fusion processing are used to integrate multi-dimensional data on environment, crops, and agriculture, generating high-quality datasets with unified spatiotemporal labels, and providing comprehensive data support for precise decision-making. To achieve intelligent decision-making, the crop growth digital twin model is used to simulate the growth process. The intelligent diagnostic unit accurately identifies growth deviations and their causes, and the decision generation unit combines agronomic rules and optimization objectives to select the optimal solution, thus avoiding biases in experience-based decision-making. To ensure precise and efficient operation, the strategy execution module transforms decisions into equipment control commands, driving automated equipment to precisely execute agricultural operations, reducing human intervention errors, and improving the efficiency and consistency of agricultural operations. To drive continuous system evolution, the self-learning optimization module iteratively optimizes the model and rule base based on execution feedback data, enabling the system to gradually adapt to changes in the planting environment and crop characteristics, thereby improving decision-making accuracy in the long term. Optimizing planting efficiency and precise water, fertilizer and pesticide management reduce resource waste and planting costs, while ensuring stable crop growth, helping to improve crop yield and quality, and enhancing the economic benefits of planting. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall structure of a crop precision planting decision system based on multimodal data fusion and artificial intelligence according to the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0019] To better understand the above technical solutions, the following will provide a detailed explanation of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0020] like Figure 1 As shown, this embodiment of the invention provides a crop precision planting decision-making system based on multimodal data fusion and artificial intelligence, comprising: The multimodal data acquisition module is used to collect crop growth environment data, crop data, and agricultural operation data in real time through sensing devices, image acquisition devices, and user terminals deployed in the planting area. The data fusion and processing module communicates with the multimodal data acquisition module and is used to perform spatiotemporal alignment, cleaning and fusion processing on the heterogeneous data acquired by the multimodal data acquisition module to generate a fusion dataset with unified spatiotemporal labels and to establish a dynamically updated electronic archive for each plant unit. The artificial intelligence decision engine communicates and connects with the data fusion and processing module, including: A crop growth digital twin model, built based on crop physiological and ecological mechanisms and historical fusion data, is used to simulate the crop growth process; The intelligent diagnostic unit is used to input the fused dataset into the crop growth digital twin model, and diagnose crop growth deviations and identify the causes by comparing the model's predicted state with the actual state. The decision generation unit is used to generate precise agricultural operation decisions based on the diagnostic results of the intelligent diagnostic unit, by calling the agronomic rule base and optimization objectives, and through simulation. The strategy execution module communicates with the artificial intelligence decision engine to parse precision agricultural operation decisions into control commands and drive automated agricultural equipment to execute them. The self-learning optimization module communicates with the multimodal data acquisition module and the artificial intelligence decision engine respectively. It is used to collect feedback data after the strategy execution module is executed, and to iteratively optimize the crop growth digital twin model and agronomic rule base through machine learning algorithms based on the feedback data, forming a closed-loop decision.

[0021] In this embodiment, the multimodal data acquisition module is the core data input terminal of the crop precision planting decision system based on multimodal data fusion and artificial intelligence. By integrating various acquisition devices and processing technologies, it achieves comprehensive capture, purification and integration of crop growth data across the entire process, providing high-quality and standardized data support for subsequent decision analysis. It is the "source of fresh water" to ensure the system's accurate decision-making. The multimodal data acquisition module is primarily responsible for coordinating the deployment of sensor arrays, multispectral image acquisition devices, and farmer mobile terminals in the planting area to simultaneously acquire multi-source heterogeneous data, including physical parameters of the crop growth environment, crop phenotypic images, and records of manual farming operations. Through a series of standardized processing steps, it completes preliminary data quality control, spatiotemporal unification, noise removal, missing value repair, and feature-level fusion, ultimately generating a fused dataset with unified spatiotemporal labels. It also constructs a dynamically updated full-lifecycle electronic archive for each independent plant unit, ensuring data integrity, consistency, and usability. This provides comprehensive and accurate basic data support for the artificial intelligence decision engine. The multimodal data acquisition module includes: Multi-source data sensing unit: The multi-source data sensing unit, acting as the "front-end tentacles" of data acquisition, achieves simultaneous multi-dimensional data capture through the collaborative work of three types of devices. Sensor arrays are distributed at key locations in the planting area, continuously collecting environmental physical parameters such as soil temperature and humidity, air temperature and humidity, light intensity, carbon dioxide concentration, and soil nutrient content. Multispectral image acquisition devices periodically capture images of the crop canopy and individual plants, identifying phenotypic characteristics such as leaf color, texture, morphology, plant height, and leaf area. Farmers' mobile terminals serve as the manual data entry point, recording agricultural operations such as sowing time, fertilizer application, irrigation volume, pesticide application type and time, and field management measures. The simultaneous operation of these three types of devices forms a raw multi-source dataset covering the three dimensions of environment, crops, and agricultural activities, ensuring the comprehensiveness and timeliness of data collection. Data preprocessing unit: The data preprocessing unit is responsible for the initial screening and purification of raw data, implementing targeted quality control based on the characteristics of different data types. For environmental physical parameters, sensor reading calibration technology is used to correct equipment system errors, and outlier identification algorithms are used to remove abnormal data caused by equipment failures or sudden environmental changes. For crop phenotypic images, illumination normalization technology is used to eliminate the impact of light intensity differences at different shooting times on image quality, and background segmentation algorithms are used to remove irrelevant backgrounds such as soil and weeds, while retaining the crop image information. For manual operation records, text structured parsing technology is used to convert non-standardized natural language records into structured data in a unified format, ensuring that raw data from different sources and in different formats can meet the requirements of subsequent processing, generating a well-organized preprocessed multi-source dataset. Spatiotemporal alignment unit: The spatiotemporal alignment unit is responsible for solving the "spatiotemporal misalignment" problem of multi-source data and constructing a unified spatiotemporal benchmark framework. Based on the geographic information system raster technology, it divides the planting area into precise spatial grids and assigns unique spatial coordinates to each grid. Combining crop growth time series, it performs time synchronization calibration on the preprocessed multi-source dataset using a unified timestamp standard. Through spatial coordinate registration, it maps environmental data and image data collected from different locations to the corresponding spatial grids. Through timestamp synchronization, it ensures that environmental parameters, crop phenotypes, and agricultural operation data at the same time node are correlated. Finally, it integrates the originally scattered heterogeneous data into a unified spatiotemporal benchmark to form a spatiotemporally aligned dataset, laying the foundation for subsequent data fusion. Data cleaning unit: The data cleaning unit is dedicated to improving data quality and achieving "refinement and optimization." First, it performs integrity checks on the spatiotemporally aligned dataset to identify missing data. For continuous data generated by high-frequency sensor acquisition, an adaptive filtering algorithm is used to suppress random noise interference, smooth data fluctuations, and retain true data characteristics. For missing data identified during the checks, based on crop phenological characteristics and the correlation between historical data from the same period and adjacent spatial nodes, an interpolation and reconstruction algorithm is used to fill in the data gaps, ensuring data continuity and integrity. After a series of processing steps, a high-quality cleaned dataset is generated, eliminating redundancy, noise, and missing data. Multimodal fusion unit: The multimodal fusion unit is the "core hub" of data integration, enabling deep integration of different types of data. Taking a single plant as the basic unit, a spatiotemporal correlation matrix is ​​constructed to fuse environmental physical parameters, crop phenotypic features, and agricultural operation records in a high-quality clean dataset at the feature level. In the correlation matrix, the horizontal dimension corresponds to the time series, and the vertical dimension covers the three types of data features: environment, crop, and agricultural operations. By mining the inherent correlation of different data types in the same spatiotemporal space, data barriers are broken down, and scattered single-dimensional data is transformed into comprehensive data that can fully reflect the crop growth status and influencing factors. Finally, a fusion dataset with unified spatiotemporal labels is generated, realizing the aggregation and enhancement of data value. Electronic record building unit: The electronic record building unit establishes a "digital identity file" for each individual crop. Based on the fused dataset, it creates a full life cycle digital record for each independent plant unit. The record integrates basic information such as plant genetic information, varietal characteristics, and seed origin; dynamic data such as real-time growth indicators, plant height, leaf area, and flowering and fruiting status; environmental response history, such as records of changes in growth status under different environmental conditions; and agricultural intervention records, such as operation details such as fertilization, irrigation, and pest and disease control. As the fused dataset is dynamically updated, the electronic record iterates synchronously, completely tracing the entire growth cycle of the crop from sowing to harvest, providing individual-level data support for accurate diagnosis and personalized decision-making.

[0022] In this embodiment, the data fusion and processing module is the core data processing link of the crop precision planting decision system based on multimodal data fusion and artificial intelligence. It receives heterogeneous data input from the multimodal data acquisition module and realizes data regularization, purification and deep integration through standardized processing flow. It provides the artificial intelligence decision engine with core data with unified spatiotemporal benchmark, high-quality features and individual correlation attributes. It is a key bridge connecting data acquisition and intelligent decision-making. The data fusion and processing module is primarily responsible for receiving heterogeneous data such as environmental physical parameters, crop phenotypic image data, and agricultural operation records transmitted from the multimodal data acquisition module. It eliminates data misalignment in time and space through spatiotemporal alignment technology, removes noise and repairs missing data to ensure data quality through data cleaning technology, and integrates different types of data at the feature level through multimodal fusion technology, ultimately generating a fused dataset with unified spatiotemporal labels. Simultaneously, based on the fused dataset, it constructs and dynamically updates a full life-cycle electronic profile for each independent plant unit, ensuring that the data accurately reflects the crop's growth status and influencing factors, providing reliable support for subsequent intelligent diagnosis and decision-making. The data fusion and processing module includes: Spatiotemporal alignment unit: The spatiotemporal alignment unit is a core component in addressing the "spatiotemporal heterogeneity" problem of multi-source data. It is primarily responsible for uniformly calibrating the time and spatial dimensions of heterogeneous data input from the multimodal data acquisition module. Using a geographic information system (GIS) raster as the spatial reference, this unit divides the planting area into fixed-size spatial grids with preset precision, assigning unique spatial coordinates to each grid to ensure all data corresponds to a specific physical location. In the time dimension, the unit uses a unified timestamp standard as a benchmark to synchronously calibrate the time information of data input from different acquisition devices (such as sensors, image acquisition instruments, and mobile terminals), eliminating time discrepancies between devices. Through the collaborative processing of timestamp synchronization and spatial coordinate registration, environmental, phenotypic, and agricultural data, originally scattered across different spatiotemporal dimensions, are mapped to a unified spatiotemporal reference framework, generating spatiotemporally aligned data. This lays a spatiotemporal consistency foundation for subsequent data cleaning and fusion. Data cleaning unit: The data cleaning unit is responsible for "quality purification" of the data. Through multi-step processing, it eliminates noise, outliers, and missing data to ensure accuracy and completeness. First, the unit performs a completeness check on the spatiotemporally aligned data, verifying each spatiotemporal node for missing, duplicate, or formatted data. Second, for continuous data generated by high-frequency sensor acquisition (such as soil moisture data collected every 5 seconds), an adaptive filtering algorithm is used to suppress random noise interference, smooth data fluctuations, preserve the true trend of data changes, and avoid interference from high-frequency noise in subsequent analysis. Finally, for any missing data discovered during the check, the unit performs interpolation and reconstruction based on crop phenological characteristics: combining the growth characteristics of the current crop growth stage (such as wheat jointing stage or rice grain-filling stage), referencing data from adjacent time nodes within the same spatial grid, and data from surrounding grids of the same period, data correlation analysis is used to fill in missing values, ultimately generating high-quality cleaned data free of noise and missing data. Multimodal fusion unit: The multimodal fusion unit is key to achieving data "value aggregation." It breaks down the barriers between environmental, crop, and agricultural data through feature-level integration technology, forming comprehensive data that fully reflects crop growth status. This unit constructs a spatiotemporal correlation matrix with a single plant as the core. The horizontal dimension of the matrix corresponds to time series (e.g., multiple daily collection periods), while the vertical dimension covers the core features of three data categories: environmental physical parameters (soil temperature and humidity, light intensity), crop phenotypic characteristics (plant height, leaf area, leaf color), and agricultural operation records (fertilizer application, irrigation time). During construction, the unit analyzes the intrinsic correlations between different data features (e.g., the correlation between light intensity and leaf photosynthetic efficiency, and the correlation between fertilizer application and crop nutrient accumulation), transforming scattered single-dimensional data into a set of interconnected multi-dimensional features, achieving deep data integration. Finally, a fusion dataset with unified spatiotemporal labels is generated. This dataset retains the core information of each data category while enhancing the data's ability to represent crop growth status through correlation and integration. Electronic record building unit: The electronic record building unit establishes a "digital growth profile" for each individual crop. Based on the fusion dataset generated by the multimodal fusion unit, it creates and dynamically updates a full life-cycle electronic record for each independent plant unit. First, this unit assigns a unique identifier to each plant to ensure accurate correspondence between the record and the actual plant. Second, the record content integrates four core types of information: plant genetic information (basic attributes such as varietal characteristics, seed origin, and disease resistance), real-time growth indicators (dynamic changes in plant height, stem diameter, flowering time, and fruit set rate), environmental response history (changes in plant growth status under different environmental conditions, such as the degree of leaf wilting under high-temperature stress), and agricultural intervention records (specific timing and dosage of fertilization, irrigation, and pest and disease control). Finally, as the fusion dataset is updated in real time, this unit synchronously updates the electronic record content, ensuring that the record can completely trace the entire growth process of the crop from sowing to harvest, providing personalized data support for subsequent precise diagnosis and decision-making for individual plants.

[0023] In this embodiment, the artificial intelligence decision engine is the core decision-making hub of the crop precision planting decision system based on multimodal data fusion and artificial intelligence. It receives the high-quality fusion dataset and electronic archives output by the data fusion and processing module, simulates the growth process through the crop growth digital twin model, locates growth deviations by relying on the intelligent diagnostic unit, and generates the optimal agricultural operation decision by combining the agronomic rule base and simulation evaluation. It provides precise instruction basis for the strategy execution module and is the key core to realize the transformation of crop planting from "experience-driven" to "data and model-driven". The AI ​​decision engine is primarily responsible for deeply analyzing the relationship between crop growth patterns and environmental and agricultural operations. Based on a fusion dataset, it reproduces and predicts crop growth status through a digital twin model of crop growth. Using an intelligent diagnostic unit to compare model predictions with actual growth status, it accurately identifies growth deviations and their causes. Combining an agronomic rule base with preset optimization objectives (such as maximizing yield, maximizing water and fertilizer utilization, and reducing pesticide use), it selects the optimal agricultural operation decision through multi-scheme simulation evaluation. Finally, it outputs precise instructions that can directly guide automated equipment execution, achieving intelligent growth simulation, deviation diagnosis, and decision generation throughout the entire process, ensuring the scientific and precise nature of agricultural operations. The AI ​​decision engine includes: Crop growth digital twin model: The crop growth digital twin model is a "virtual mirror" that simulates the crop growth process. By combining mechanism modeling and data calibration, it can accurately reproduce and predict the crop growth status and includes four core sub-units. Model initialization unit: Based on the physiological and ecological mechanisms of crops, a theoretical model framework for crop growth processes is constructed. This framework fully covers the key physiological processes of crop growth, including photosynthesis (the correlation between the rate of light energy conversion into chemical energy and light intensity and CO2 concentration), respiration (the relationship between organic matter consumption and temperature and oxygen), nutrient absorption (the interaction between the absorption efficiency of nitrogen, phosphorus, potassium, and other elements by roots and soil moisture and nutrient concentration), and nutrient distribution (the translocation patterns of photosynthetic products to roots, stems, leaves, and fruits). This lays the theoretical foundation for subsequent model operation and ensures that the model conforms to the natural physiological logic of crop growth. Parameter calibration unit: Historical fusion data (including environmental data from past planting cycles, crop growth indicators, agricultural operation records, and final yield data) are input into the theoretical model framework, and physiological parameters in the model are adjusted through parameter optimization algorithms. For example, for the photosynthesis process, the photosynthetic efficiency coefficient is calibrated by using the correspondence between light intensity and leaf photosynthetic rate in historical data; for the nutrient absorption process, the nutrient absorption rate parameters are optimized based on the changes in crop biomass under different soil nutrient concentrations, and finally, a parameter-calibrated crop growth model is generated, making the model more consistent with the actual growth characteristics of specific planting areas and specific crop varieties. State simulation unit: Input the real-time fusion dataset (current environmental physical parameters, current crop growth indicators, and records of executed agricultural operations) into the parameter-calibrated crop growth model, and simulate the crop growth process under the current environmental and agricultural conditions through dynamic simulation algorithms. During the simulation, the model will output simulated values ​​of crop growth status according to time steps (such as hours or days), including key indicators such as plant height, stem diameter, leaf area index, chlorophyll content, and fruit development progress, fully reproducing the crop growth dynamics from the current moment to a future period (such as a week or a growth cycle), providing a predictive basis for subsequent diagnosis and decision-making. Model Validation Unit: The simulated crop growth status output by the state simulation unit is compared with the actual observed values ​​in the fusion dataset (such as the actual plant height and leaf chlorophyll measured values ​​obtained through image acquisition equipment) index by index. Based on error evaluation indicators (such as root mean square error and mean absolute error), the degree of deviation between the simulated values ​​and the actual values ​​is judged. If the deviation exceeds the preset threshold, the model parameters are adjusted in reverse (such as correcting the photosynthetic efficiency coefficient) or the model structure is optimized (such as supplementing the allocation weight of nutrients at different growth stages). Finally, a validated and optimized digital twin model of crop growth is generated to ensure that the model simulation accuracy continuously meets the decision-making requirements. Intelligent diagnostic unit: The intelligent diagnostic unit is a "precise detector" for locating crop growth problems. By comparing the simulated and actual growth states, it delves into the root causes of deviations and provides problem-oriented guidance for decision generation. It contains four core sub-units. State prediction subunit: Receives the real-time fused dataset transmitted by the data fusion and processing module, inputs it into the validated and optimized crop growth digital twin model, drives the model to simulate the growth state that the crop should exhibit under the current environmental conditions (such as current soil temperature and humidity, light duration) and the agricultural operations performed (such as the previous fertilization amount), and outputs predicted values ​​of various key growth indicators (such as predicting the leaf area index 3 days later and predicting the flowering rate 10 days later), forming the "ideal growth state benchmark". State Comparison Subunit: In conjunction with the State Prediction Subunit, it extracts actual crop growth observations from the same period in the fused dataset (such as actual leaf area index analyzed by multispectral images and actual flowering rate measured by field sampling), and performs a difference analysis on each index and time node with the predicted values ​​output by the State Prediction Subunit. By calculating the difference between the predicted and actual values ​​of each index, a crop growth state deviation vector is generated. Each element in the vector corresponds to the degree of deviation of a growth index (such as a plant height deviation of 2 cm and a chlorophyll content deviation of 0.3 SPAD value), intuitively presenting the deviation of crop growth from the ideal state. Deviation Diagnosis Subunit: Based on the growth state deviation vector generated by the state comparison subunit, the built-in causal reasoning algorithm is invoked to uncover the potential causes behind the deviation. The algorithm combines crop physiological characteristics with historical data correlation patterns to analyze the relationship between the deviation and environmental factors (such as whether insufficient recent light has led to less accumulation of photosynthetic products, thus causing plant height deviation) and agricultural operation factors (such as whether insufficient fertilization in the previous application has led to nutrient deficiency, causing chlorophyll content deviation). Unrelated factors are eliminated (such as when the air CO2 concentration is normal, its influence on the current deviation can be ruled out), and the 1-3 core causes most likely to cause the deviation are identified. Cause Output Subunit: Integrates the analysis results of the deviation diagnosis subunit and outputs a complete diagnostic report in a unified format. The report includes the type of growth deviation (such as "delayed plant height development", "low chlorophyll content", "slow fruit enlargement") and a list of causative factors (such as "Cause 1: The average daily light intensity in the past 7 days is lower than the crop's requirement threshold of 5000 lux; Cause 2: The amount of nitrogen fertilizer applied 15 days ago was 20% less than the model's recommended amount"), providing the decision generation unit with a clear "problem list" to ensure that subsequent decisions can accurately target the causes and formulate solutions. Decision generation unit: The decision generation unit is a "planner" for formulating optimal agricultural plans. It combines diagnostic results, agronomic rules and simulation evaluations to select agricultural operation decisions that take into account both effectiveness and efficiency. It includes five core sub-units. The diagnostic result parsing subunit receives the diagnostic report output by the intelligent diagnostic unit and performs structured parsing on the list of growth deviation types and causative factors. For example, it parses "delayed plant height development, caused by insufficient light and insufficient nitrogen fertilizer" as "deviation type: plant height deviation; cause 1: light intensity < 5000 lux (requirement threshold); cause 2: nitrogen fertilizer application amount < 20% of the model recommendation amount", generating standardized structured diagnostic information, eliminating the ambiguity of natural language description, and providing accurate matching basis for subsequent rule retrieval. Rule retrieval subunit: Linked with the diagnostic result parsing subunit, it retrieves matching agronomic rules from the built-in agronomic rule base based on structured diagnostic information. The agronomic rule base contains a large number of standardized rules based on agronomic research and practical experience (such as "If plant height deviation is due to insufficient light and the crop is in the jointing stage, supplemental lighting equipment can be used to extend the light exposure by 2 hours / day", "If insufficient nitrogen fertilizer leads to low chlorophyll, urea 5 kg / mu can be applied, and it should be applied after rain or irrigation"). The retrieval process locates relevant rules through keyword matching (such as "plant height deviation + insufficient light", "low chlorophyll + insufficient nitrogen fertilizer") and generates a candidate agronomic rule set. The scheme generation sub-unit combines the candidate agronomic rule set with preset optimization goals (set by the user or system according to planting needs, such as "yield priority", "water and fertilizer conservation priority", "green and environmental protection priority") to generate multiple candidate agricultural operation schemes. For example, for the diagnosis results of "plant height deviation + insufficient light + insufficient nitrogen fertilizer", if the optimization goal is "yield priority", Scheme 1 may be "supplementary light for 2 hours / day + topdressing with urea 5kg / mu"; if the optimization goal is "water and fertilizer conservation priority", Scheme 2 may be "supplementary light for 3 hours / day + topdressing with urea 3kg / mu". Each scheme clearly defines the operation type, operation parameters (such as supplementary light duration and fertilizer application amount) and execution time window (such as "execute within the next 3 days"). Simulation evaluation subunit: Each candidate agricultural operation plan is input into the crop growth digital twin model. The model is used to simulate and predict the changes in crop growth status after each plan is implemented. For example, after implementing plan 1, it is estimated whether the plant height can recover to the ideal state and whether the chlorophyll content meets the standard in the next 10 days. After implementing plan 2, it is estimated whether the growth indicators recover within the same time period. At the same time, the resource consumption (such as the electricity and fertilizer costs of plan 1 and the electricity cost of plan 2) and potential risks (such as whether excessive supplemental lighting will cause leaf burn) of each plan are evaluated, and the plan evaluation results including "growth recovery effect", "resource consumption" and "risk level" are generated. Decision output subunit: Based on simulation evaluation results, it sorts the data in multiple dimensions according to preset optimization objectives. If the optimization objective is "yield priority", the scheme with the best growth recovery effect is selected first. If the optimization objective is "water and fertilizer conservation priority", the scheme with the lowest resource consumption is selected under the premise that the growth recovery effect meets the standard. Finally, one optimal agricultural operation scheme is selected and standardized precision agricultural operation decision is output. The decision content includes operation type (supplementary lighting, fertilization, irrigation, etc.), operation parameters (specific duration, dosage, pressure, etc.), execution time, and target area (spatial grid of the corresponding planting area), providing clear instructions for the strategy execution module.

[0024] In this embodiment, the strategy execution module is the core link connecting the artificial intelligence decision engine and automated agricultural equipment in the crop precision planting decision system based on multimodal data fusion and artificial intelligence. It undertakes the key responsibility of transforming abstract precision agricultural operation decisions into concrete and executable equipment control instructions and driving the equipment to implement them. At the same time, it monitors the execution process and status in real time and provides feedback data for the self-learning optimization module. It is an important link to ensure the integrity of the "decision-execution-feedback" closed loop. The strategy execution module is primarily responsible for receiving precise agricultural operation decisions from the AI ​​decision engine. Through a series of standardized processing steps, it breaks down these decisions into atomized sequences of agricultural operation actions, then arranges them into a set of standardized, device-independent operation instructions according to operational logic. This instruction set is then adapted to different types and brands of automated agricultural equipment (such as intelligent irrigation equipment, precision fertilizers, unmanned plant protection drones, and automatic supplemental lighting equipment), generating control commands that the equipment can recognize and issuing them for execution. Simultaneously, the module collects real-time data on equipment execution status and operational effects, feeding this data back to the self-learning optimization module. This ensures the precise transformation, efficient implementation, and traceability of agricultural operations from decision to execution, guaranteeing that every agricultural instruction is completed accurately as expected. The strategy execution module includes: Decision Analysis Subunit: The decision analysis subunit is the primary link in the "decision-to-action" process. It is responsible for breaking down the overall precision agricultural operation decision output by the AI ​​decision engine into the smallest executable atomized sequence of agricultural operation actions. Precision agricultural operation decisions typically include information such as operation type (e.g., irrigation, fertilization, supplemental lighting, pesticide application), target execution area (e.g., planting area A1 grid, tomato greenhouse row 3), core execution parameters (e.g., irrigation volume 15 m³ / mu, fertilizer application volume 5 kg / mu urea, supplemental lighting duration 2 hours / day), and execution time window (e.g., 8:00-10:00 AM). The decision analysis subunit, through pre-defined action decomposition rules, transforms this information into clearly defined and logically clear atomized actions. For example, "Precision fertilization (5 kg / mu urea) for wheat in grid A1..." The decision to "execute at 8:00 AM" is broken down into a sequence of actions: "Atomized Action 1: Drive the fertilizer applicator to the starting coordinates of grid A1; Atomized Action 2: Check if the urea remaining in the fertilizer applicator's hopper meets the requirement of 5 kg / acre; Atomized Action 3: Set the fertilizer applicator's discharge flow rate to 0.5 kg / minute; Atomized Action 4: Open the fertilizer applicator's discharge port; Atomized Action 5: Drive the fertilizer applicator to move at a constant speed (0.5 m / s) along the preset path of grid A1; Atomized Action 6: After moving to the end coordinates of grid A1, close the fertilizer applicator's discharge port; Atomized Action 7: Record the fertilization completion time and actual fertilization amount of grid A1." Each atomized action contains a clear executing entity, operation content, parameter standards, and timing requirements, laying the foundation for subsequent instruction programming. Instruction arrangement subunit: The instruction orchestration subunit is the core of the "action-to-instruction" process. It is responsible for reorganizing and orchestrating the atomized agricultural operation sequence according to actual operational logic and safety regulations, generating a standardized set of equipment-independent operation instructions. This subunit has a built-in operational logic rule library, covering the sequential requirements of different agricultural operations (e.g., checking soil moisture before irrigation to ensure it's below a threshold, and cleaning fertilizer pipelines after fertilization to prevent blockages), equipment coordination rules (e.g., fertilizer applicators and irrigation equipment cannot operate simultaneously in the same area to prevent fertilizer loss), and safety operation regulations (e.g., confirming no personnel are inside the greenhouse before turning on supplemental lighting, and checking pesticide residue before operating a plant protection machine). During the orchestration process, the instruction orchestration subunit adjusts the timing and performs logical checks on the atomized action sequence based on the rule library, for example, moving the "check soil moisture" action to the "start" action. Before the irrigation valve is activated, ensure that the irrigation operation conforms to the "on-demand irrigation" logic. Simultaneously, generate a standardized set of operating instructions using a unified format of "operation type - execution parameters - timing requirements - verification standards." This instruction set is independent of specific equipment models and only describes "what to do, how to do it, and what standards to meet." For example, "Standardized Instruction 1: Soil Moisture Detection - Execution Area A2 Grid - Detection Parameter: Soil Volumetric Moisture Content - Verification Standard: If ≤20%, execute subsequent irrigation instructions; Standardized Instruction 2: Irrigation Valve Opening - Execution Parameter: Flow Rate 10 m³ / hour - Timing Requirements: Execute immediately after moisture detection meets the standard; Standardized Instruction 3: Irrigation Duration Control - Execution Parameter: Continuous Irrigation for 18 minutes - Verification Standard: Close the valve after cumulative irrigation volume reaches 15 m³ / acre," providing a universal interface for subsequent equipment adaptation. Equipment adapter subunit: The equipment adaptation subunit is crucial for converting general instructions into equipment-specific instructions. It is responsible for transforming standardized operational instruction sets into equipment control instructions that are perfectly matched to the target automated agricultural equipment. This subunit has a built-in equipment parameter library that stores information such as communication protocols (e.g., Modbus, TCP / IP, LoRaWAN), instruction formats (e.g., ASCII, hexadecimal codes), parameter ranges (e.g., a certain brand of fertilizer applicator with a maximum discharge flow rate of 1 kg / min, a certain model of irrigation valve supporting a pressure range of 0.2-0.8 MPa), and status feedback fields (e.g., "Running" corresponding to code 0x01, "Fault" corresponding to code 0x02). During the adaptation process, the equipment adaptation subunit first reads the target equipment model specified in the precision agricultural operation decision (e.g., "Intelligent Irrigation System - Brand A - Model IR-2023") and retrieves the equipment's specific configuration from the equipment parameter library. Then, it converts the standardized operational instructions one by one according to the configuration. For example, it converts the standardized instruction "Irrigation valve open - flow rate 10 m³ / hour" into the specific instruction "IR_CMD" for the Brand A IR-2023 model irrigation valve. OPEN VALVE 1 FLOW 10MODBUS_ADDR 0x03” ensures that the converted instructions conform to the device’s communication requirements and syntax specifications, and can be directly recognized and responded to by the device; if the target device is a new model, staff can update the device parameter library through the module interface to expand the module’s device compatibility range; Execution driver subunit: The execution drive subunit is the actuator that "converts commands into actions," responsible for sending equipment control commands to the target automated agricultural equipment and driving the equipment to complete specific agricultural operations. This subunit has multi-channel communication capabilities, establishing stable communication links with different equipment via wired (RS485, Ethernet) or wireless (4G, 5G, LoRa) methods. Before issuing commands, it performs a final verification of the completeness and legality of the commands (e.g., checking if the flow parameters in the command are within the maximum range supported by the equipment and if the communication address is correct) to avoid invalid or erroneous commands causing equipment failure. After successful verification, the equipment control commands are issued sequentially, and the response signals of the equipment are monitored in real time. If the equipment does not respond with "command received successfully" within a preset time (e.g., 5 seconds), it automatically retryes issuing the command (up to 3 times). If the retry fails, an alarm is triggered and fault information is recorded. During equipment execution, the execution drive subunit continuously maintains communication with the equipment to ensure stable operation according to the command parameters. For example, during the execution of a fertilizer applicator, if the equipment reports "discharge flow rate deviates from the set value by 5%", the drive subunit will issue a parameter adjustment command "ADJUST". "FLOW +0.025kg / minute" corrects execution deviations in real time, ensuring that the operation results meet the decision requirements; Status monitoring subunit: The status monitoring subunit acts as the "eyes" of the "execution process tracking and feedback," responsible for collecting real-time execution status data and agricultural operation effect data of automated agricultural equipment and feeding it back to the self-learning optimization module. This subunit acquires data in two ways: first, by directly reading the operating status fed back by the equipment through the communication link (e.g., "Fertilizer applicator current position X=120m, Y=80m," "Irrigation valve current opening 80%," "Supplemental lighting equipment power 2000W," "Equipment operating status: normal / fault"); second, by collecting operational effect data through sensors deployed in the field (e.g., changes in soil nitrogen content after fertilization, changes in soil moisture after irrigation, changes in crop leaf temperature after supplemental lighting). The status monitoring subunit processes the collected data in real time, generating an execution status report in the format of "Equipment number - Operation command ID - Execution time - Status parameter - Effect parameter," for example, "Equipment number: Fertilizer applicator F-01; Operation command ID: FER-20240520-001; Execution time: 2024-05-20." 08:00-08:20; Status parameters: moving speed 0.5m / s, discharge flow rate 0.5kg / min, no faults; Effect parameters: soil nitrogen content in grid A1 increased from 15mg / kg to 25mg / kg, meeting the target requirements; The execution status report is then transmitted to the self-learning optimization module in real time, providing real execution effect data support for subsequent iterative optimization of the model and rule base.

[0025] In this embodiment, the self-learning optimization module is a key support for the "decision-execution-feedback-optimization" closed loop of the crop precision planting decision system based on multimodal data fusion and artificial intelligence. It receives the actual operating data of the strategy execution module, evaluates the decision effect through data analysis, and then uses machine learning algorithms to iteratively optimize the crop growth digital twin model and agronomic rule base. Finally, the optimization results are fed back to the artificial intelligence decision engine, enabling the system to continuously adapt to changes in the planting environment and differences in crop growth characteristics, and achieve long-term improvement in decision accuracy. It is the core driving force for the system to transform from "fixed function" to "intelligent evolution". The self-learning optimization module is primarily responsible for collecting three types of core feedback data after the strategy execution module executes the operation: crop growth status data (such as changes in plant height, chlorophyll content, and fruit development progress after the operation), environmental response data (such as the duration of soil moisture maintenance after irrigation and the conversion efficiency of soil nutrients after fertilization), and agricultural operation execution effect data (such as the deviation between actual irrigation amount and the decided irrigation amount, and the difference between the actual running time of supplemental lighting equipment and the set value). Based on this data, the module evaluates the actual value of precise agricultural operation decisions through multi-dimensional effect assessment, and then uses machine learning algorithms to optimize the parameters and structure of the crop growth digital twin model, updates the rule weights and content of the agronomic rule base through reinforcement learning algorithms, and finally deploys the optimized model and rule base to the artificial intelligence decision engine to complete a closed-loop optimization. Through this continuous cycle, the module enables the system's decision-making ability to continuously improve as the planting cycle progresses, gradually adapting to the microenvironment and crop variety characteristics of specific planting areas. The self-learning optimization module includes: Feedback data acquisition subunit: The feedback data acquisition subunit serves as the entry point for "optimizing data sources," responsible for comprehensively and in real-time collecting various feedback data after strategy execution to ensure data coverage of core dimensions of decision-making effectiveness. This subunit acquires data through two core methods: First, it connects to the sensing and image acquisition devices of the multimodal data acquisition module to obtain dynamic changes in crop growth status after agricultural operations, including key growth indicators such as plant height, stem diameter, leaf area index, leaf chlorophyll content, and fruit enlargement rate, as well as environmental response data such as soil temperature and humidity, nutrient content, and air light intensity. Second, it connects to the status monitoring subunit of the strategy execution module. Acquire actual execution data for agricultural operations, including actual operating parameters of automated agricultural equipment (such as actual fertilizer output flow rate of fertilizer applicators and actual opening duration of irrigation valves), deviation data during execution (such as the difference between actual fertilizer application and the decided fertilizer application, and the difference between actual supplemental lighting duration and the set value), and equipment fault records (such as the duration of operation interruption caused by fertilizer applicator blockage during execution). The collected data will be structured and stored in the format of "Operation Instruction ID-Collection Time-Data Type-Value-Unit", for example, "Operation Instruction ID: FER-20240520-001; Collection Time: 2024-05-21 08:00; Data Type: Crop Growth Status - Plant Height; Value: 78; Unit: cm", providing complete and standardized raw data for subsequent effect evaluation. Effect evaluation sub-unit: The effect evaluation subunit is the core of "optimization direction judgment," responsible for evaluating the actual implementation effect of precision agricultural operation decisions based on feedback data and generating clear optimization criteria. This subunit first establishes an effect evaluation index system covering three core indicators: 1) the effect achievement rate (e.g., the decision expected "plant height increase of 2 cm," but the actual increase was 1.8 cm, achieving a 90% success rate); 2) resource utilization efficiency (e.g., the decision to apply 5 kg / mu of fertilizer resulted in actual absorption of 4 kg / mu, achieving a utilization rate of 80%); and 3) the deviation risk rate (e.g., the percentage of times irrigation deviation exceeded 10% during implementation). Subsequently, the subunit compares the feedback data with the expected decision effect output by the artificial intelligence decision engine (e.g., the expectation in the decision that "chlorophyll content increases by 0.3 SPAD value after 2 hours / day of supplemental lighting") dimension by dimension, analyzing the reasons for deviations. For example, the deviation of "chlorophyll content only increased by 0.1 SPAD" may stem from inaccurate simulation parameters of the crop growth digital twin model for supplemental lighting response, or the recommended rules for supplemental lighting duration in the agronomic rule base not being suitable for the current crop growth stage. Finally, the sub-unit generates a decision-making effect evaluation report in the format of "Operation Instruction ID - Evaluation Index - Actual Value - Expected Value - Deviation Rate - Preliminary Analysis of Deviation Causes", for example, "Operation Instruction ID: LIT-20240522-001; Evaluation Index: Chlorophyll Content Increase; Actual Value: 0.1 SPAD; Expected Value: 0.3 SPAD; Deviation Rate: -66.7%; Preliminary Analysis of Deviation Causes: The supplemental lighting efficiency coefficient during the jointing stage in the crop growth digital twin model is too high, resulting in the expected effect being higher than the actual effect," providing a clear direction for subsequent model and rule base optimization. Model optimization subunit: The model optimization subunit is the core of the "crop growth digital twin model iteration." It is responsible for adjusting model parameters and structure based on the decision-making effectiveness evaluation report, using machine learning algorithms to improve the accuracy of model simulation and prediction. The optimization process of this subunit consists of two steps: The first step is parameter adjustment. For simulation deviations pointed out in the evaluation report (such as "simulated plant height is 5 cm higher than the actual value"), parameter optimization algorithms (such as gradient descent and random forest regression) are used to adjust key physiological parameters in the crop growth digital twin model. For example, if the deviation stems from inaccurate photosynthesis simulation, the photosynthetic efficiency coefficient is adjusted; if the deviation stems from inaccurate nutrient allocation simulation, the distribution of photosynthetic products to the stem is optimized. The first step is to adjust the allocation ratio parameters to gradually reduce the error between the simulated values ​​and the actual observed values ​​output by the model. The second step is structural optimization. If the deviation still exceeds the preset threshold after parameter adjustment (e.g., the error is greater than 10%), it is analyzed whether there are defects in the model structure. For example, if the model does not consider the inhibitory effect of high temperature stress on root nutrient absorption, resulting in excessive deviation in the nutrient absorption simulation at a certain time period, the sub-unit will supplement the nutrient absorption logic module under high temperature stress to improve the model structure. After optimization, the sub-unit generates an optimized crop growth digital twin model and verifies its effectiveness (comparing the simulated values ​​of the optimized model with historical feedback data to confirm that the error has been reduced) to ensure that the optimization effect meets the standards. Rule base optimization subunit: The rule base optimization subunit is the core of the "agronomic rule base iteration." It is responsible for updating the weights and content of rules based on decision-making effectiveness evaluation reports using reinforcement learning algorithms, thereby improving the applicability and effectiveness of the rules. The optimization of this subunit includes three aspects: First, rule weight adjustment. For existing rules in the agronomic rule base, weights are assigned based on their execution effects. For example, the rule "nitrogen fertilizer application rate of 5 kg / mu during the jointing stage" increases yield by 10% (good effect) in sandy soil but only by 3% (medium effect) in clay soil. Therefore, the subunit will increase the weight of this rule in the sandy soil scenario and decrease its weight in the clay soil scenario, ensuring that subsequent decisions prioritize rules suitable for the current scenario. Second, rule content modification. If a rule is executed... If deviations continue to occur (e.g., the rule "3 hours / day of supplemental lighting during the tomato fruiting period" resulted in uneven fruit coloring after actual implementation), the sub-unit will adjust the rule parameters based on feedback data. For example, the supplemental lighting duration may be adjusted to 2.5 hours / day, and the supplemental lighting time may be changed from morning to afternoon to optimize the applicability of the rule. Thirdly, new rules may be added. If it is found that the existing rules cannot cover new scenarios (e.g., the existing fertilization rules are not applicable when a new crop variety is introduced to the planting area), the sub-unit will generate new agronomic rules (e.g., "3 kg / acre of potassium fertilizer application during the fruiting period of the new tomato variety") based on feedback data under the new scenario (e.g., the crop growth effect corresponding to different fertilization amounts) and add them to the agronomic rule library. After optimization, the sub-unit generates an optimized agronomic rule library and marks the applicable scenarios and weights of each rule. Closed-loop control subunit: The closed-loop control subunit is crucial for the "implementation of optimization results." It is responsible for the safe and efficient deployment of the optimized crop growth digital twin model and agronomic rule base to the artificial intelligence decision engine, completing the final step of closed-loop optimization. This subunit first performs compatibility testing on the optimized model and rule base, confirming that their interface format and data interaction protocol are consistent with the artificial intelligence decision engine to avoid operational failures after deployment. Then, a "gray-scale deployment" strategy is adopted, initially testing the optimized model and rule base in a small planting area (such as a specific grid) of the artificial intelligence decision engine, monitoring the decision output and execution effects in real time. Once no anomalies are confirmed, the deployment scope is gradually expanded until the entire planting area is covered. Finally, the subunit records key information about this optimization (such as optimization time, optimization content, and error comparison before and after optimization) and stores it in the system log, providing a basis for subsequent optimization effect tracking and analysis. Simultaneously, this subunit establishes a rollback mechanism for the optimization results; if a decrease in decision accuracy is found after deployment, the crop growth digital twin model and agronomic rule base can be quickly rolled back to the pre-optimization version, ensuring stable system operation.

[0026] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A crop precision planting decision-making system based on multimodal data fusion and artificial intelligence, characterized in that: include: The multimodal data acquisition module is used to collect crop growth environment data, crop data, and agricultural operation data in real time through sensing devices, image acquisition devices, and user terminals deployed in the planting area. The data fusion and processing module communicates with the multimodal data acquisition module and is used to perform spatiotemporal alignment, cleaning and fusion processing on the heterogeneous data acquired by the multimodal data acquisition module to generate a fusion dataset with unified spatiotemporal labels and to establish a dynamically updated electronic archive for each plant unit. The artificial intelligence decision engine communicates and connects with the data fusion and processing module, including: A crop growth digital twin model, built based on crop physiological and ecological mechanisms and historical fusion data, is used to simulate the crop growth process; The intelligent diagnostic unit is used to input the fused dataset into the crop growth digital twin model, and diagnose crop growth deviations and identify the causes by comparing the model's predicted state with the actual state. The decision generation unit is used to generate precise agricultural operation decisions based on the diagnostic results of the intelligent diagnostic unit, by calling the agronomic rule base and optimization objectives, and through simulation. The strategy execution module communicates with the artificial intelligence decision engine to parse precision agricultural operation decisions into control commands and drive automated agricultural equipment to execute them. The self-learning optimization module communicates with the multimodal data acquisition module and the artificial intelligence decision engine respectively. It is used to collect feedback data after the strategy execution module is executed, and to iteratively optimize the crop growth digital twin model and agronomic rule base through machine learning algorithms based on the feedback data, forming a closed-loop decision.

2. The crop precision planting decision-making system based on multimodal data fusion and artificial intelligence according to claim 1, characterized in that: The multimodal data acquisition module includes: The multi-source data sensing unit is used to simultaneously acquire raw multi-source datasets containing environmental physical parameters, crop phenotypic images, and manual operation records through sensor arrays, multispectral image acquisition devices, and farmers' mobile terminals deployed in the planting area. The data preprocessing unit is used to perform preliminary quality control on the original multi-source dataset. This includes calibrating sensor readings and removing outliers for environmental physical parameters, normalizing illumination and segmenting backgrounds for crop phenotypic images, and performing text structured parsing on manual operation records to generate a preprocessed multi-source dataset. The spatiotemporal alignment unit is used to synchronize timestamps and register spatial coordinates of preprocessed multi-source datasets based on geographic information system rasters and crop growth time series, mapping data from different sources to a unified spatiotemporal reference framework to form a spatiotemporal aligned dataset. The data cleaning unit is used to perform integrity checks on the spatiotemporally aligned dataset, employs an adaptive filtering algorithm to suppress noise in high-frequency collected data, and interpolates and reconstructs missing data based on crop phenological characteristics to generate a high-quality cleaned dataset. The multimodal fusion unit is used to fuse environmental physical parameters, crop phenotypic features and agricultural operation records in a high-quality clean dataset at the feature level. By constructing a spatiotemporal correlation matrix with plants as the unit, a fused dataset with unified spatiotemporal labels is generated. The electronic archive building unit is used to create and dynamically update a digital archive of the entire life cycle for each individual plant unit based on a fusion dataset. This archive integrates plant genetic information, real-time growth indicators, environmental response history, and agricultural intervention records.

3. The crop precision planting decision-making system based on multimodal data fusion and artificial intelligence according to claim 1, characterized in that: The data fusion and processing module includes: The spatiotemporal alignment unit is used to receive heterogeneous data collected by the multimodal data acquisition module. Based on the geographic information system raster and crop growth time series, it performs timestamp synchronization and spatial coordinate registration on the heterogeneous data to generate spatiotemporal aligned data. The data cleaning unit is used to perform integrity checks on spatiotemporally aligned data, employs an adaptive filtering algorithm to suppress noise in high-frequency acquired data, and interpolates and reconstructs missing data based on crop phenological characteristics to generate high-quality cleaned data. The multimodal fusion unit is used to perform feature-level fusion on high-quality cleaned data. By constructing a spatiotemporal correlation matrix based on plants, it integrates environmental physical parameters, crop phenotypic features and agricultural operation records to generate a fusion dataset with unified spatiotemporal labels. The electronic archive building unit is used to create and dynamically update a digital archive of the entire life cycle for each individual plant unit based on a fusion dataset. The digital archive integrates plant genetic information, real-time growth indicators, environmental response history and agricultural intervention records.

4. The crop precision planting decision-making system based on multimodal data fusion and artificial intelligence according to claim 1, characterized in that: The crop growth digital twin model is constructed based on crop physiological and ecological mechanisms and historical fusion data, and is used to simulate the crop growth process, including: The model initialization unit is used to construct a theoretical model framework for crop growth processes based on crop physiological and ecological mechanisms. This theoretical model framework includes key physiological processes such as photosynthesis, respiration, and nutrient absorption and distribution. The parameter calibration unit is used to input historical fusion data into the theoretical model framework, calibrate the physiological parameters in the theoretical model framework through parameter optimization algorithms, and generate a parameter-calibrated crop growth model. The state simulation unit is used to input the real-time fused dataset into the calibrated crop growth model, simulate the crop growth process under given environmental conditions through dynamic simulation algorithms, and output the simulated value of the crop growth state. The model validation unit is used to compare the simulated values ​​of crop growth status with the actual observed values, adjust the model parameters or structure based on the error evaluation index, and generate a validated and optimized digital twin model of crop growth.

5. The crop precision planting decision-making system based on multimodal data fusion and artificial intelligence according to claim 1, characterized in that: The intelligent diagnostic unit is used to input the fused dataset into the crop growth digital twin model, and diagnose crop growth deviations and identify their causes by comparing the model's predicted state with the actual state, including: The state prediction subunit is used to input the fused dataset into the crop growth digital twin model, simulate the crop growth process under the current environmental conditions, and output the predicted value of the crop growth state. The state comparison subunit, which is connected to the state prediction subunit, is used to perform difference analysis between the predicted values ​​of crop growth status and the actual observation values ​​in the fused dataset, and generate a crop growth status deviation vector. The deviation diagnosis subunit is connected in communication with the state comparison subunit. It is used to identify potential causes of growth deviation based on the crop growth state deviation vector and through causal reasoning algorithms. The cause output subunit communicates with the deviation diagnosis subunit and is used to output diagnostic results, including the growth deviation type and a list of causative factors.

6. The crop precision planting decision-making system based on multimodal data fusion and artificial intelligence according to claim 1, characterized in that: The decision generation unit is used to generate precise agricultural operation decisions based on the diagnostic results of the intelligent diagnostic unit, by calling the agronomic rule base and optimization objectives, and through simulation, including: The diagnostic result parsing subunit is used to receive the diagnostic results output by the intelligent diagnostic unit, and to parse the growth deviation type and causative factor list in the diagnostic results to generate structured diagnostic information. The rule retrieval subunit is connected to the diagnostic result parsing subunit. It is used to retrieve agronomic rules that match the growth deviation type and causative factors from the agronomic rule base based on structured diagnostic information, and generate a candidate agronomic rule set. The scheme generation subunit communicates with the rule retrieval subunit and is used to generate multiple candidate agricultural operation schemes by combining the candidate agronomic rule set and the preset optimization target. The simulation evaluation subunit is connected to the scheme generation subunit. It is used to input each candidate agricultural operation scheme into the crop growth digital twin model, and predict the crop growth status after each scheme is executed through simulation to generate scheme evaluation results. The decision output subunit communicates with the simulation evaluation subunit and is used to select the optimal agricultural operation plan based on the scheme evaluation results and the optimization objective, and output precise agricultural operation decisions.

7. The crop precision planting decision-making system based on multimodal data fusion and artificial intelligence according to claim 1, characterized in that: The strategy execution module includes: The decision parsing subunit is used to receive the precision agricultural operation decisions generated by the artificial intelligence decision engine and parse the precision agricultural operation decisions into an atomic sequence of agricultural operation actions; The instruction arrangement subunit, which is connected to the decision analysis subunit, is used to arrange the atomized agricultural operation action sequence according to the preset operation logic to generate a set of standardized operation instructions that are independent of equipment. The equipment adapter subunit communicates with the instruction orchestration subunit and is used to convert standardized operation instruction sets into equipment control instructions that match the target automated agricultural equipment. The execution drive subunit is connected to the device adapter subunit to send device control commands to the corresponding automated agricultural equipment and drive the automated agricultural equipment to perform precision agricultural operations. The status monitoring subunit is connected to the execution drive subunit and is used to collect execution status data of automated agricultural equipment in real time and feed the execution status data back to the self-learning optimization module.

8. The crop precision planting decision-making system based on multimodal data fusion and artificial intelligence according to claim 1, characterized in that: The self-learning optimization module includes: The feedback data acquisition subunit is used to collect feedback data after the strategy execution module has executed the strategy execution module. The feedback data includes crop growth status data, environmental response data, and agricultural operation execution effect data. The effect evaluation subunit is connected to the feedback data acquisition subunit and is used to evaluate the actual implementation effect of precision agricultural operation decisions based on feedback data and generate a decision effect evaluation report. The model optimization subunit communicates with the effect evaluation subunit and is used to adjust the parameters and optimize the structure of the crop growth digital twin model based on the decision effect evaluation report and through machine learning algorithms to generate an optimized crop growth digital twin model. The rule base optimization subunit communicates with the effect evaluation subunit and is used to update the rules and adjust the weights of the agronomic rule base based on the decision effect evaluation report and through reinforcement learning algorithms to generate an optimized agronomic rule base. The closed-loop control subunit communicates with the model optimization subunit and the rule base optimization subunit to deploy the optimized crop growth digital twin model and the optimized agronomic rule base to the artificial intelligence decision engine, thereby completing the closed-loop decision optimization.

Images