Agricultural planting optimization system and method based on big data analysis

Abstract

The invention relates to the technical field of electric digital data processing, in particular to an agricultural planting optimization system and method based on big data analysis. The method comprises the following steps: acquiring agricultural planting environment parameter data, and performing preprocessing and standardization to obtain a standardized environment feature vector; acquiring image, sound and smell data of crops; performing deep feature extraction on the image, sound and smell data to generate a multi-modal feature representation matrix; and performing multi-modal data fusion based on the standardized environment feature vector and the multi-modal feature representation matrix to obtain a comprehensive feature space. According to the invention, through multi-modal feature fusion and digital twinborn modeling, the precision and real-time performance of crop growth risk and pest and disease prediction are improved, and optimal regulation and control and efficient resource utilization of precision agriculture are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of digital data processing, and in particular, to an agricultural planting optimization system and method based on big data analysis. Background Art

[0002] The agricultural planting method based on big data analysis is an important direction of modern precision agriculture, and its technological development has experienced a transformation from traditional empirical planting to data-driven decision-making. With the development of the Internet of Things (IoT), remote sensing technology, and intelligent sensors, the collection of agricultural data has been automated and made real-time. In recent years, the progress of computer vision, deep learning, and multi-modal data analysis technologies has made the identification of crop growth status, pests and diseases, and yield prediction more accurate. At the same time, the combination of cloud computing and edge computing has improved the storage and processing capabilities of agricultural data. However, there are still many defects in the existing technologies. The sources of agricultural data are extensive but the standardization degree is low, resulting in difficulty in fusing different data types. The digital twin technology of the agricultural ecosystem is still in the development stage, and the simulation accuracy and dynamic adjustment ability need to be improved. The existing agricultural intelligent management systems still have problems of insufficient accuracy and poor real-time performance in growth risk assessment and pest and disease prediction, and it is difficult to achieve precise intervention. Summary of the Invention

[0003] Based on this, it is necessary for the present invention to provide an agricultural planting optimization system and method based on big data analysis to solve at least one of the above technical problems.

[0004] To achieve the above object, an agricultural planting optimization method based on big data analysis includes the following steps:

[0005] Step S1: Obtain agricultural planting environment parameter data, perform preprocessing and standardization, and obtain a standardized environmental feature vector;

[0006] Step S2: Obtain image, sound, and odor data of crops; extract crop growth characteristics from the image, sound, and odor data to generate a multi-modal feature representation matrix;

[0007] Step S3: Perform multi-modal data fusion based on the standardized environmental feature vector and the multi-modal feature representation matrix to obtain a comprehensive feature space;

[0008] Step S4: Model a digital twin ecosystem for the comprehensive feature space to generate an agricultural ecosystem prediction model;

[0009] Step S5: Based on the agricultural ecosystem prediction model, conduct a growth risk assessment of the crops to obtain a crop growth risk early warning report; based on the agricultural ecosystem prediction model, conduct a pest and disease prediction of the crops to obtain a crop pest and disease early warning report; record the crop growth risk early warning report and the crop pest and disease early warning report as the crop early warning report.

[0010] Step S6: According to the crop early warning report, conduct optimized regulation of agricultural planting to obtain an optimized planting management plan.

[0011] Through the fusion of standardized agricultural environment data and multi-modal features, the present invention improves the unity and comparability of agricultural data, solves the problem of low standardization degree of agricultural data sources in the prior art, and enhances the fusion ability of different types of data. By extracting the deep features of image, sound and odor data, the accuracy of crop growth status and pest and disease monitoring is improved, the crop growth analysis is extended from a single dimension to multi-modal information, and the reliability of pest and disease identification and growth status monitoring is improved. The construction of the comprehensive feature space realizes the fusion of multi-source data, lays a foundation for the construction of the digital twin ecosystem, makes the modeling of the agricultural ecosystem more comprehensive, and overcomes the problem of insufficient simulation accuracy of the digital twin technology of the agricultural ecosystem. The establishment of the agricultural ecosystem prediction model improves the scientific nature of agricultural intelligent management, makes the growth risk assessment and pest and disease prediction more accurate, and makes up for the deficiencies of the existing system in terms of risk assessment and prediction accuracy. The combination of crop growth risk early warning and pest and disease early warning realizes the dynamic monitoring of the entire growth cycle of crops, improves the real-time performance and accuracy of agricultural management, and makes up for the defects of the prior art in precise intervention. The agricultural planting optimization regulation plan based on the crop early warning report makes the allocation of agricultural production factors more scientific, improves the resource utilization efficiency and environmental adaptability, realizes the refined management of agricultural production, and promotes the development of precision agriculture.

[0012] Preferably, the present invention also provides an agricultural planting optimization system based on big data analysis for implementing the above-mentioned agricultural planting optimization method based on big data analysis. The agricultural planting optimization system based on big data analysis includes:

[0013] An agricultural environment perception and data preprocessing module, which is used to obtain agricultural planting environment parameter data, and conduct preprocessing and standardization to obtain a standardized environmental feature vector.

[0014] A multi-modal agricultural perception module, which is used to obtain image, sound and odor data of the crops; extract crop growth characteristics from the image, sound and odor data to generate a multi-modal feature representation matrix.

[0015] A multimodal data fusion module, which is used to perform multimodal data fusion based on the standardized environmental feature vector and the multimodal feature representation matrix to obtain a comprehensive feature space;

[0016] A digital twin agricultural ecological modeling module, which is used to model a digital twin ecosystem for the comprehensive feature space to generate an agricultural ecosystem prediction model;

[0017] A crop intelligent early warning module, which is used to evaluate the growth risk of crops based on the agricultural ecosystem prediction model to obtain a crop growth risk early warning report; predict crop diseases and pests based on the agricultural ecosystem prediction model to obtain a crop diseases and pests early warning report; record the crop growth risk early warning report and the crop diseases and pests early warning report as the crop early warning report;

[0018] An intelligent agricultural optimization and regulation module, which is used to perform agricultural planting optimization and regulation according to the crop early warning report to obtain an optimized planting management plan.

[0019] The present invention can convert environmental data from different sources and with different units into a unified standard format by efficiently acquiring parameter data of an agricultural planting environment, performing preprocessing and standardization, ensuring the comparability and accuracy of the data, and providing a reliable basis for subsequent analysis. By collecting image, sound, and odor data of crops and using deep feature extraction technology to convert these data into a highly information-based feature matrix, multi-dimensional monitoring data for the growth state of crops and environmental changes can be provided, making the analysis results more comprehensive and accurate. By combining the standardized environmental feature vector with the multimodal feature representation matrix, a comprehensive feature space is created, which can fuse data from different sources into an overall analysis framework, improving the comprehensiveness and accuracy of data processing and providing a more comprehensive perspective for the modeling of agricultural ecosystems. Using the comprehensive feature space to generate an agricultural ecosystem prediction model can accurately simulate the change process of the agricultural ecological environment and provide a theoretical basis for the risk assessment of crop growth and the prediction of crop diseases and pests. Based on the agricultural ecosystem prediction model, the risks in the growth process of crops can be identified in a timely manner, and growth risk and diseases and pests early warning reports can be generated, thereby providing preventive intervention measures for farmers and reducing crop losses. Using the crop early warning report to perform agricultural planting optimization and regulation, through a scientific data-driven method, an optimized planting management plan is proposed to ensure the high efficiency and stability of agricultural production and minimize the negative impact of environmental factors on crop growth. Description of the Drawings

[0020] By reading the detailed description of the non-restrictive embodiments with reference to the following drawings, other features, purposes, and advantages of the present invention will become more obvious:

[0021] Figure 1Schematic diagram of the step process of an agricultural planting optimization method based on big data analysis according to the present invention;

[0022] Figure 2 is Figure 1 detailed step process diagram of step S1 in;

[0023] Figure 3 is Figure 1 detailed step process diagram of step S2 in. Specific implementation manner

[0024] The technical method of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative efforts belong to the scope of protection of the present invention.

[0025] In addition, the accompanying drawings are only schematic diagrams of the present invention and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, so repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. The functional entities can be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor methods and / or microcontroller methods.

[0026] It should be understood that although the terms "first", "second", etc. may be used here to describe each unit, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, the first unit can be called the second unit, and similarly the second unit can be called the first unit. The term "and / or" used here includes any and all combinations of one or more of the listed related items.

[0027] To achieve the above object, please refer to Figures 1 to 3 , the present invention provides an agricultural planting optimization method based on big data analysis, and the method includes the following steps:

[0028] Step S1: Obtain agricultural planting environment parameter data, perform preprocessing and standardization to obtain a standardized environmental feature vector;

[0029] Step S2: Obtain image, sound and odor data of crops; extract crop growth characteristics from the image, sound and odor data to generate a multi-modal feature representation matrix;

[0030] Step S3: Perform multimodal data fusion based on the standardized environmental feature vector and the multimodal feature representation matrix to obtain a comprehensive feature space;

[0031] Step S4: Model the digital twin ecosystem for the comprehensive feature space to generate an agricultural ecosystem prediction model;

[0032] Step S5: Perform growth risk assessment on the crops based on the agricultural ecosystem prediction model to obtain a crop growth risk warning report; perform pest and disease prediction on the crops based on the agricultural ecosystem prediction model to obtain a crop pest and disease warning report; record the crop growth risk warning report and the crop pest and disease warning report as the crop warning report;

[0033] Step S6: Perform optimized regulation of agricultural planting according to the crop warning report to obtain an optimized planting management plan.

[0034] In the embodiment of the present invention, refer to Figure 1 As shown, it is a schematic diagram of the step flow of an agricultural planting optimization method based on big data analysis according to the present invention. In this example, the agricultural planting optimization method based on big data analysis includes the following steps:

[0035] Step S1: Obtain agricultural planting environment parameter data, and perform preprocessing and standardization to obtain a standardized environmental feature vector;

[0036] In the embodiment of the present invention, the acquisition of agricultural planting environment parameter data is realized by deploying a sensor network, including a soil humidity sensor, a temperature sensor, a light sensor, an atmospheric humidity sensor, and a soil pH sensor. The sensors collect data every 15 minutes and transmit the data to the central data processing platform in real time. For data preprocessing, outlier detection is first performed, and data points exceeding 1.5 times the interquartile range are removed using the interquartile range method. Missing values are filled using the linear interpolation method. Data standardization uses the Z-score standardization method, and the calculation formula is standardized value = (original value - mean) / standard deviation. The obtained environmental feature vector after standardization contains the normalized temperature, humidity, light intensity, and soil pH values, the vector dimension is 5, and the value range of each dimension is between [-1, 1].

[0037] Step S2: Obtain the image, sound, and odor data of the crops; extract the crop growth characteristics from the image, sound, and odor data to generate a multimodal feature representation matrix;

[0038] In the embodiments of the present invention, when collecting multi-modal data on the crop growth environment, high-definition industrial cameras, professional acoustic sensors, and odor analysis instruments are first deployed, and data acquisition is carried out according to a standardized process. The ground camera is a 5-megapixel RGB color camera with a resolution of 2592×1944 pixels, a frame rate set at 30 frames per second, an aperture range of F1.4 - F8, and a color depth of 24 bits. Image noise is eliminated through a preprocessing algorithm, and histogram equalization and adaptive contrast enhancement are performed. Convolution feature extraction is performed on the collected standardized image data. A convolution kernel with a window size of 3×3, a stride of 1, and a padding method of same is used to extract the texture, edge, and color features of the image. The soil hydrophone selects a professional acoustic sensor with a frequency response range of 20 Hz - 20 kHz, a sampling rate of 44.1 kHz, and a quantization bit number of 16 bits. Noise reduction and spectrum enhancement are performed through fast Fourier transform to extract the time-frequency domain features of the sound. The MOS odor sensor selects a metal oxide semiconductor sensor array, including multiple gas detection units such as CO2, methane, and ammonia, with a sampling frequency of 1 Hz and a resolution of 0.1 ppm. Through baseline drift correction and peak extraction algorithms, the drift error of the sensor itself is removed, and the chemical feature vector of the odor is extracted. Finally, multi-modal feature fusion of weight normalization and information entropy calculation is performed on the image feature sub-matrix, sound feature sub-matrix, and odor feature sub-matrix to generate a multi-modal feature representation matrix with a dimension of 128.

[0039] Step S3: Perform multi-modal data fusion based on the standardized environmental feature vector and the multi-modal feature representation matrix to obtain a comprehensive feature space;

[0040] In the embodiments of the present invention, for the standardized environmental feature vector, dimension mapping and feature deconstruction are first performed. The original 50-200 dimensional feature space is linearly transformed into a 128-dimensional representation space, and an orthogonal projection matrix is used for feature reconstruction to ensure the integrity of feature information. For the multi-modal feature representation matrix, the entropy weight method is used to calculate the weights of each feature dimension. The information entropy is calculated through the Shannon entropy formula, and the weight normalization range is limited to [0,1] to eliminate the differences in different feature dimensions. When constructing the feature association mapping model, the correlation coefficient matrix and the mutual information matrix are used to establish the association strength mapping between the environmental features and the multi-modal features. The association degree threshold is set to 0.6, and the feature association connections below the threshold are removed. Cross-modal feature interaction is performed on the feature association mapping matrix. Through orthogonal transformation and the information gain algorithm, the information integration of the heterogeneous feature space is realized, and the intermediate result of feature interaction is generated. Based on the intermediate result of feature interaction, the principal component analysis method is used to reduce and reconstruct the multi-dimensional heterogeneous features, and the principal components with a cumulative variance contribution rate exceeding 95% are retained, and the feature dimensions are compressed to the optimal subspace. Finally, the reliability evaluation and consistency test of the comprehensive feature space are carried out, and the Cronbach's α coefficient and the structural equation model are used to ensure the reliability and validity of the feature space, and a stable and reliable comprehensive feature space of the agricultural ecosystem is obtained.

[0041] Step S4: Perform digital twin ecosystem modeling on the comprehensive feature space to generate an agricultural ecosystem prediction model;

[0042] In the embodiments of the present invention, first, topological structure analysis is performed on the comprehensive feature space. Using the complex network analysis method in graph theory, a feature node network is constructed. The number of nodes is limited to 50 - 200, and the edge connection weight threshold is set to 0.5. Key elements are calculated through degree centrality, closeness centrality, and eigenvector centrality. Based on the topological mapping of the ecosystem, mathematical modeling of the key elements of the agricultural ecosystem is carried out, and a system of difference equations including four core elements of temperature, humidity, light, and soil nutrients is established. The equation constraint conditions include: the temperature change rate ≤ 3 °C / day, the humidity change rate ≤ 10% / day, the light intensity change rate ≤ 15% / day, and the soil nutrient concentration change rate ≤ 5% / day. Dynamic parameter calibration is performed on the initial mathematical model of the ecosystem. Using the Monte Carlo iteration method, the number of iterations is set to 1000 times, and the parameter error is controlled within the range of ±5%. A multi-dimensional state space simulation framework is constructed. Using the discrete-time dynamic system modeling method, the state space dimension is 128, the time step is set to 1 hour, and the simulation period is one growing season. Iterative verification is performed on the digital twin ecosystem simulation model. Using the bootstrap resampling method, the model prediction error is calculated, and the error convergence threshold is set to 0.01. The model is corrected through the error backpropagation algorithm. Finally, based on the corrected digital twin ecosystem simulation model, scenario deduction of the agricultural ecosystem is performed to generate an agricultural ecosystem prediction model including four dimensions of temperature, humidity, nutrients, and light.

[0043] Step S5: Based on the agricultural ecosystem prediction model, perform a growth risk assessment on the crops to obtain a crop growth risk warning report; based on the agricultural ecosystem prediction model, perform a pest and disease prediction on the crops to obtain a crop pest and disease warning report; record the crop growth risk warning report and the crop pest and disease warning report as the crop warning report;

[0044] In the embodiments of the present invention, first, by decomposing the weights of risk characteristic factors of the agricultural ecosystem prediction model, the risk characteristics are divided into three dimensions: environmental adaptability, physiological response, and resource constraint. The entropy weight method is used to calculate the weight coefficients of each characteristic factor, and they are arranged in descending order according to importance to construct a risk characteristic weight vector, where the weight of environmental adaptability is 30%, the weight of physiological response is 40%, and the weight of resource constraint is 30%. Subsequently, the fuzzy comprehensive evaluation method is used to conduct multi-dimensional quantitative evaluation of the risk factors in the crop growth process, and a risk factor evaluation matrix is constructed. The number of rows in the matrix corresponds to the number of risk characteristic factors, and the number of columns is the risk level dimension, and the sum of each row is always equal to 100. Then, a probability distribution analysis is performed on the risk factor evaluation matrix, and the Kolmogorov-Smirnov test method is used to calibrate the critical value, and the risk levels are divided into low risk (0-25 points), medium risk (26-50 points), high risk (51-75 points), and extremely high risk (76-100 points). Furthermore, based on the risk level determination standard, the comprehensive risk index of the crop growth process is calculated by the linear weighted summation method, comprehensively considering the risk factors in the three dimensions of environmental adaptability, physiological response, and resource constraint. Finally, threshold mapping and risk grading are performed on the crop growth risk index, the hierarchical risk assessment results are generated by the clustering analysis method, and a systematic diagnosis report and early warning information are generated according to the risk assessment results to form a crop growth risk early warning report.

[0045] Step S6: Perform agricultural planting optimization and regulation according to the crop early warning report to obtain an optimized planting management plan.

[0046] In the embodiments of the present invention, first, the entropy weight method is used to systematically analyze the risk factors in the crop early warning report. By calculating the weight entropy values of each risk factor, a set of key regulation indicators is constructed, and they are arranged in descending order according to the information entropy to determine the 3-5 most critical regulation indicators. Subsequently, based on the set of key regulation indicators, the grey relational analysis method is used to accurately allocate the agricultural planting environmental resources, and a resource optimization model is constructed, clarifying four core parameters including soil humidity, nutrient concentration, light intensity, and temperature. Upper and lower limit thresholds are set for each parameter. The soil humidity is controlled at 15%-35%, the nutrient concentration is controlled at 0.1-0.5 g / kg, and the light intensity is controlled at 500-1500 μmol / m 2 / s, control the temperature at 18 - 28 °C; then, conduct multi-scenario verification on the environmental parameter regulation plan through the Monte Carlo simulation method, set 100 random simulation experiments, generate an alternative set of planting management plans, and use the analysis of variance method to evaluate the effectiveness of each plan, screening out the plan with the smallest variance and the narrowest parameter fluctuation range; furthermore, based on the alternative set of planting management plans, use the entropy weight-TOPSIS comprehensive evaluation method to dynamically allocate the weights of crop production factors, determine the optimal allocation ratio of the four production factors of seeds, fertilizers, water resources, and pesticides, and construct an optimized resource allocation model; finally, use the fuzzy comprehensive evaluation method to conduct a feasibility analysis and risk hedging assessment on the optimized resource allocation model, establish a risk matrix, calculate the membership degree of each allocation plan, and finally generate a risk reduction planting management plan, and through the precise regulation of the agricultural planting environment, form the final optimized planting management plan.

[0047] Through the integration of standardized agricultural environment data and multi-modal features, the present invention improves the unity and comparability of agricultural data, solves the problem of low standardization of agricultural data sources in the prior art, and enhances the fusion ability of different types of data. By extracting the deep features of image, sound, and odor data, the accuracy of crop growth status and pest and disease monitoring is improved, the crop growth analysis is extended from a single dimension to multi-modal information, and the reliability of pest and disease identification and growth status monitoring is enhanced. The construction of the comprehensive feature space realizes the fusion of multi-source data, lays a foundation for the construction of the digital twin ecosystem, makes the modeling of the agricultural ecosystem more comprehensive, and overcomes the problem of insufficient simulation accuracy of the digital twin technology of the agricultural ecosystem. The establishment of the agricultural ecosystem prediction model improves the scientific nature of agricultural intelligent management, makes the growth risk assessment and pest and disease prediction more accurate, and makes up for the deficiencies of the existing system in terms of risk assessment and prediction accuracy. The combination of crop growth risk warning and pest and disease warning realizes the dynamic monitoring of the entire growth cycle of crops, improves the real-time performance and accuracy of agricultural management, and makes up for the defects of the prior art in precise intervention. The agricultural planting optimization regulation plan based on the crop warning report makes the allocation of agricultural production factors more scientific, improves the resource utilization efficiency and environmental adaptability, realizes the refined management of agricultural production, and promotes the development of precision agriculture.

[0048] Preferably, step S1 includes the following steps:

[0049] Step S11: Deploy a modular sensor array in the agricultural planting area, and use the modular sensor array to collect temperature, humidity, and light intensity to obtain agricultural plant physical parameter data;

[0050] Step S12: Based on the agricultural plant physical parameter data, use a multi-layer soil sensor probe to collect the soil moisture content at 0 - 100 cm to obtain agricultural planting soil layer data, where the layer sampling interval is 10 cm;

[0051] Step S13: Detect the concentrations of nitrogen, phosphorus, and potassium elements in the soil based on the agricultural planting soil stratification data to obtain agricultural planting environment parameter data;

[0052] Step S14: Perform outlier detection and normalization on the agricultural planting environment parameter data to obtain a standardized interval mapping, where the high-precision calibration reference sources for the normalization process are a temperature standard blackbody furnace, a humidity generator, and a spectral radiation standard source;

[0053] Step S15: Perform feature selection and dimensionality reduction on the standardized interval mapping to obtain a key environmental feature subset;

[0054] Step S16: Conduct probability distribution statistics based on the key environmental feature subset and perform vector transformation on the probability distribution statistics results to obtain a standardized environmental feature vector.

[0055] As an embodiment of the present invention, refer to Figure 2 shown in Figure 1 for the detailed step flow diagram of step S1 in

[0056] Step S11: Deploy a modular sensor array in the agricultural planting area and use the modular sensor array to collect temperature, humidity, and light intensity to obtain agricultural planting physical parameter data;

[0057] Step S12: Based on the agricultural planting physical parameter data, use a multi-layer soil sensor probe to collect the soil water content at 0 - 100 cm to obtain agricultural planting soil stratification data, where the stratification sampling interval is 10 cm;

[0058] Step S13: Detect the concentrations of nitrogen, phosphorus, and potassium elements in the soil based on the agricultural planting soil stratification data to obtain agricultural planting environment parameter data;

[0059] Step S14: Perform outlier detection and normalization on the agricultural planting environment parameter data to obtain a standardized interval mapping, where the high-precision calibration reference sources for the normalization process are a temperature standard blackbody furnace, a humidity generator, and a spectral radiation standard source;

[0060] Step S15: Perform feature selection and dimensionality reduction on the standardized interval mapping to obtain a key environmental feature subset;

[0061] Step S16: Conduct probability distribution statistics based on the key environmental feature subset and perform vector transformation on the probability distribution statistics results to obtain a standardized environmental feature vector.

[0062] In the embodiment of the present invention, in the agricultural planting area, firstly, accurate environmental data collection is realized by modular sensor array deployment. The sensor array includes a temperature sensor (model DS18B20), a humidity sensor (model SHT20), and a light intensity sensor (model BH1750). These sensors are evenly arranged at a density of 4-6 per square meter to ensure the spatial representativeness of the data. The temperature measurement range is -55 to 125°C, with an accuracy of ±0.5°C; the humidity measurement range is 0-100% RH, with an accuracy of ±3%; the light intensity measurement range is 0-65535lx, with an accuracy of ±5lx. The physical parameter data is recorded in real time through the data acquisition module (model NI9172), and the sampling frequency is set to 1 time / 5 minutes. The multi-layer soil sensor probe (model EC5) detects soil moisture content at six depth levels of 0-10cm, 10-20cm, 20-40cm, 40-60cm, 60-80cm, and 80-100cm, and the probe detection accuracy is ±3%. Soil nutrient detection uses a soil nutrient rapid tester (model DJYSD-1A) to measure the concentrations of nitrogen, phosphorus, and potassium, respectively, with a detection accuracy of ±5%. The 3σ principle is used for data outlier detection, and the Z-score method is used to identify and remove data points that are beyond the normal range. The temperature standard blackbody furnace (model DTI-2000), humidity generator (model HG-3800) and spectral radiation standard source (model OL-751) are introduced for data normalization processing, and the original data are mapped to the [0,1] interval by linear interpolation method. The information gain method is used for feature selection to retain the 10-15 environmental features that have the most significant impact on agricultural planting. The principal component analysis (PCA) technique is used for dimensionality reduction processing to extract the principal components with a contribution rate of more than 85% of the explained variance. Probability distribution statistics is based on the kernel density estimation method to construct the probability density function of the feature, and finally the feature vector is standardized by multidimensional Gaussian distribution.

[0063] Through the deployment of a modular sensor array, the present invention enhances the flexibility and coverage of agricultural crop physical parameter data collection, achieving precise measurement of temperature, humidity, and light intensity. Using multi-layer soil sensor probes for stratified sampling within the range of 0 - 100 cm with a 10 cm interval, the spatial resolution of soil moisture data is improved, making the monitoring of agricultural soil moisture conditions more accurate. Based on the soil stratified data, the concentrations of nitrogen, phosphorus, and potassium elements are further detected to ensure the integrity of agricultural planting environment parameter data, providing basic data support for subsequent environmental regulation. Through outlier detection and normalization processing, interference data during the collection process is eliminated, and high-precision calibration is performed using a temperature standard blackbody furnace, a humidity generator, and a spectral radiation standard source, thereby improving the accuracy and comparability of the normalized data. Feature selection and dimensionality reduction are performed on the normalized data to effectively reduce data redundancy, improve computational efficiency, and ensure that key information is not lost. Finally, through probability distribution statistical methods, the key environmental feature subset is vectorized and transformed to form a standardized environmental feature vector, providing high-quality input data for agricultural planting data analysis and intelligent decision-making.

[0064] Preferably, step S2 includes the following steps:

[0065] Step S21: Collect crop leaf characteristics using a ground camera based on agricultural planting environment parameter data; perform leaf layer distribution detection on the crop leaf characteristics to obtain leaf spatial distribution data;

[0066] Step S22: Statistically analyze the distribution law of crop leaf characteristics, and use the statistical results of the leaf characteristic distribution law to determine the crop plant morphology; quantify the mechanical relationship between the stem bending stiffness and the branching angle using the crop plant morphology to obtain stem branching data;

[0067] Step S23: Calculate the Pearson correlation coefficients of plant height, leaf area, and fruit size based on the crop plant morphology, and use the calculation results of the Pearson correlation coefficients to determine the crop fruit characteristics; perform volatile organic compound detection based on the crop fruit characteristics to obtain fruit physiological state data;

[0068] Step S24: Calculate the fruit sugar-acid ratio based on the crop fruit characteristics, and use the calculation result of the fruit sugar-acid ratio to determine the crop root characteristics; calculate the root growth range using the crop root characteristics to obtain root spatial distribution data;

[0069] Step S25: Based on the leaf spatial distribution data, stem branching data, fruit physiological state data, and root spatial distribution data, use a meteorological sensor to collect images of the crop growth environment, perform preprocessing and enhancement to obtain standardized image data; perform convolutional feature extraction on the standardized image data to obtain an image feature submatrix;

[0070] Step S26: Deploy a multi-node acoustic sensor array and an acoustic omnidirectional pickup array through soil hydrophones, collect sound signals from the crop growth environment, perform noise reduction and spectral enhancement to obtain standardized sound data; extract time-frequency domain features from the standardized sound data to obtain a sound feature submatrix;

[0071] Step S27: Construct an odor sensor network through MOS sensors, collect odor concentrations from the crop growth environment, perform baseline drift correction and peak extraction to obtain standardized odor data; extract chemical features from the standardized odor data to obtain an odor feature submatrix;

[0072] Step S28: Perform dynamic feature selection on the image feature submatrix, sound feature submatrix and odor feature submatrix through an edge computing gateway, and perform multi-modal feature fusion based on the dynamic feature selection results to obtain a multi-modal feature representation matrix.

[0073] As an embodiment of the present invention, refer to Figure 3 shown, for Figure 1 the detailed step flow diagram of step S2 in

[0074] Step S21: Collect crop leaf features using a ground camera according to agricultural planting environment parameter data; perform leaf layer distribution detection on the crop leaf features to obtain leaf spatial distribution data;

[0075] Step S22: Statistically analyze the distribution law of leaf features of the crop, and use the statistical results of the leaf feature distribution law to determine the crop plant morphology; quantify the mechanical relationship between the stem bending stiffness and the branch angle using the crop plant morphology to obtain stem branch data;

[0076] Step S23: Calculate the Pearson correlation coefficients of plant height, leaf area and fruit size based on the crop plant morphology, and use the calculation results of the Pearson correlation coefficients to determine the crop fruit features; perform volatile organic compound detection based on the crop fruit features to obtain fruit physiological state data;

[0077] Step S24: Calculate the fruit sugar-acid ratio based on the crop fruit features, and use the calculation results of the fruit sugar-acid ratio to determine the crop root features; calculate the root growth range using the crop root features to obtain root spatial distribution data;

[0078] Step S25: Based on the blade spatial distribution data, stem branching data, fruit physiological state data, and root spatial distribution data, use a meteorological sensor to collect images of the crop growth environment, perform preprocessing and enhancement to obtain standardized image data; perform convolutional feature extraction on the standardized image data to obtain an image feature sub-matrix;

[0079] Step S26: Deploy a multi-node acoustic sensor array and an acoustic omnidirectional pickup array through a soil hydrophone to collect sound signals of the crop growth environment, perform noise reduction and spectral enhancement to obtain standardized sound data; perform time-frequency domain feature extraction on the standardized sound data to obtain a sound feature sub-matrix;

[0080] Step S27: Construct an odor sensor network through a MOS sensor to collect odor concentrations of the crop growth environment, perform baseline drift correction and peak extraction to obtain standardized odor data; perform chemical feature extraction on the standardized odor data to obtain an odor feature sub-matrix;

[0081] Step S28: Perform dynamic feature selection on the image feature sub-matrix, sound feature sub-matrix, and odor feature sub-matrix through an edge computing gateway, and perform multi-modal feature fusion based on the dynamic feature selection results to obtain a multi-modal feature representation matrix.

[0082] In the embodiments of the present invention, a high-resolution ground camera (model HIKIVISION-DS-2CD3T56WD-I5, 4K resolution) captures the characteristics of crop leaves based on agricultural planting environment parameter data at a deployment density of 4-6 per hectare, and detects the stacked distribution of leaves through an image segmentation algorithm (based on region growing and threshold segmentation), calculates the leaf overlap rate, vertical projection area ratio, and occlusion index to obtain leaf spatial distribution data. Using morphological analysis methods, calculates the leaf area ratio, leaf deformity rate, and leaf margin integrity index, establishes a statistical model of the leaf feature distribution law, and determines the plant morphology of the crops. Through stress-strain analysis, quantifies the bending stiffness of the stem K = EI / L3 (E is the elastic modulus, I is the moment of inertia of the cross-section, L is the length of the rod), measures the branching angle θ, establishes a stem mechanics model, and obtains stem branching data. Uses the Pearson correlation coefficient r(x,y) = Cov(X,Y) / [σ(X)×σ(Y)] to calculate the correlation between plant height, leaf area, and fruit size, with the R value ranging from 0.6 to 0.92, and performs volatile organic compound detection using a gas chromatography-mass spectrometry instrument (model Agilent 7890A / 5975C) to obtain fruit physiological state data. Uses a spectral analyzer (model SC-B) to measure the sugar concentration and acidity of the fruits, calculates the sugar-acid ratio in the range of 1.2-2.5, and combines with a root scanner (model WinRHIZO) to perform root morphology analysis, measures the root growth range, branch number, and length distribution to obtain root spatial distribution data. A meteorological sensor (model DAVIS-6162) collects images of the crop growth environment, obtains a standardized image through image preprocessing techniques (gray normalization, histogram equalization, denoising), and uses a convolutional neural network (removing the fully connected layer) to extract the image feature submatrix. A soil hydrophone (model HY-SW01) constructs an 8-12 node acoustic sensor array with a frequency range of 20Hz-20kHz and a signal-to-noise ratio > 60dB, applies short-time Fourier transform and wavelet transform for noise reduction and spectrum enhancement, and extracts the sound feature submatrix. A MOS sensor network (model TGS2600) detects the concentrations of gases such as CO, CH4, and H2, controls the baseline drift correction amplitude within ±5%, sets the peak extraction threshold to 20%-80% of the sensor detection range, performs chemical feature extraction, and obtains the odor feature submatrix. An edge computing gateway (model NVIDIA-Jetson) performs dynamic feature selection through the mutual information criterion and the maximum correlation minimum redundancy criterion, and uses a weighted linear fusion algorithm with the weight range of 0.4-0.5 for images, 0.2-0.3 for sounds, and 0.2-0.3 for odors to generate a multi-modal feature representation matrix.

[0083] The present invention collects the characteristics of crop leaves through ground cameras, improves the accuracy of plant growth monitoring, and lays a foundation for subsequent plant morphology analysis. Based on the statistical distribution law of leaf characteristics, the morphological characteristics of plants can be effectively determined, making the growth trend assessment of crops more scientific. Calculating the Pearson correlation coefficients of plant height, leaf area, and fruit size helps to quantitatively analyze the relationship between plant morphology and fruit growth, and improves the accuracy of fruit characteristic recognition. Further, calculating the root characteristics of crops using the fruit sugar-acid ratio makes the assessment of the overall growth status of crops more comprehensive. Integrating leaf, plant morphology, fruit, and root characteristics to form crop status image data provides complete visual information support for agricultural intelligent analysis. Combining with meteorological sensors to collect crop growth environment images, and performing preprocessing and enhancement, makes the standardized image data more consistent. At the same time, convolutional feature extraction improves the depth information expression ability of image data. Deploying a multi-node acoustic sensor array and an omnidirectional sound pickup array makes the sound signal collection more spatially covered. After noise reduction and spectrum enhancement, more stable standardized sound data is obtained, and the analytical ability of the sound feature submatrix is improved through time-frequency domain feature extraction. Using the MOS sensor network to collect the odor concentration of the crop growth environment, and performing baseline drift correction and peak extraction, improves the reliability of the odor signal, and then high-quality chemical feature information is obtained. Finally, with the help of an edge computing gateway, multi-modal fusion is performed on the image, sound, and odor feature submatrices, so that the finally generated multi-modal feature representation matrix can comprehensively reflect the growth status of crops, improve the accuracy and stability of crop intelligent perception, and ensure the scientificity and reliability of data fusion by reasonably allocating the weights of each modality.

[0084] Especially importantly, the utilization of the mechanical relationship between the stem bending stiffness and the branching angle by quantifying the plant morphology of crops includes:

[0085] Measuring the geometric characteristics of the stem cross-section based on the plant morphology of crops to obtain the stem diameter and wall thickness data;

[0086] Calculating the elastic modulus of the stem diameter and wall thickness data to obtain the stem material property parameters;

[0087] Performing a three-point bending test based on the stem material property parameters to obtain the relationship curve between the stem bending strength and the deflection;

[0088] Performing stress-strain analysis on the relationship curve between the stem bending strength and the deflection to obtain the stem bending stiffness coefficient;

[0089] Measuring the branching angle between the main stem and the branches based on the stem bending stiffness coefficient to obtain the branching angle distribution data;

[0090] Perform a mechanical equilibrium analysis on the branch angle distribution data, calculate the mechanical equilibrium relationship between the branch load and the support structure, and obtain the branch mechanical equilibrium index;

[0091] Based on the branch mechanical equilibrium index, conduct statistics on the branch quantity and spatial distribution to obtain the stem branch data.

[0092] In the embodiments of the present invention, first, use a high-precision laser scanner to perform three-dimensional modeling on the crop plants, obtain the cross-sectional morphology of the stem using the slicing measurement method, record the numerical value of the stem diameter with an accuracy of 0.1 mm, and measure the stem wall thickness using a non-contact ultrasonic thickness gauge with the measurement error controlled within 0.05 mm to ensure data accuracy; then, calculate the elastic modulus of the stem diameter and wall thickness data. Apply a tensile force of 0.1 N to 10 N using a tensile testing machine, and calculate the elastic modulus E = σ / ε according to the stress σ = F / A and the strain ε = ΔL / L0, with the result reserved to three decimal places to characterize the mechanical properties of the stem material; in the three-point bending test session, fix the sample stem on a loading platform with a support point spacing of 100 mm at both ends, use a high-precision pressure sensor to record the force F applied at the loading point, and use a displacement sensor to measure the deflection δ. Draw the relationship curve between the bending strength and deflection of the stem according to the bending strength calculation formula σ = (F×L) / (4×I) and the deflection formula δ = (F×L 3 ) / (48×E×I), where L is the support spacing and I is the moment of inertia of the cross-section; subsequently, conduct stress-strain analysis on the relationship curve, and calculate the bending stiffness coefficient K = (ΔM / Δθ) of the stem using the numerical differentiation method, where M is the bending moment and θ is the change in the bending angle, with a calculation accuracy of 0.001; in the branch angle measurement session, use a high-resolution optical goniometer to record the angle between the main stem and the branch, with the resolution set to 0.1°, and the measurement results form the branch angle distribution data; then, perform a mechanical equilibrium analysis based on the branch angle distribution data, use the fulcrum moment balance equation ΣM = 0 to calculate the branch load P = m×g×sin(θ), where m is the branch mass and g is taken as 9.81 m / s 2 , and combine with the support reaction force R to calculate the mechanical equilibrium relationship between the branch load and the support structure, and obtain the branch mechanical equilibrium index; finally, based on the branch mechanical equilibrium index, conduct statistics on the branch quantity and spatial distribution, use an image processing algorithm to extract the skeleton from the three-dimensional point cloud data, and classify the spatial distribution through the K-means clustering method, and calculate the average branch spacing d = Σd i / N, where d i is the adjacent branch spacing and N is the total number of branches, and finally form the stem branch data.

[0093] Through the precise measurement of the morphological characteristics of crop plants, high-resolution data on the stem diameter and wall thickness can be obtained, thereby improving the accuracy of describing the geometric characteristics of the stem cross-section. Calculating the elastic modulus can effectively quantify the mechanical properties of the stem material, providing a scientific basis for evaluating the lodging resistance of different crop varieties. Through the three-point bending test, the relationship curve between the bending strength and deflection obtained can intuitively reflect the stress and deformation law of the stem, which helps to optimize the crop breeding strategy. Combining stress-strain analysis, the bending stiffness coefficient of the stem can be calculated, providing reliable data support for the study of the mechanical properties of plants at different growth stages or under different environmental conditions. By measuring the branching angle and analyzing the distribution data, the structural relationship between the main stem and branches can be revealed, which is of great significance for optimizing the plant architecture and improving the wind resistance and lodging resistance. Through mechanical balance analysis, the influence of branches on the main stem can be calculated, thereby optimizing the crop planting density and management strategy and improving the overall growth stability of the crop. Statistical analysis of the number and spatial distribution of branches can provide a scientific basis for crop variety improvement and precision cultivation, and improve the yield and quality of crops.

[0094] Preferably, step S3 includes the following steps:

[0095] Step S31: Perform dimensionality mapping and feature deconstruction on the standardized environmental feature vector to obtain an environmental feature representation space;

[0096] Step S32: Perform feature weight normalization on the multi-modal feature representation matrix to obtain a normalized multi-modal feature weight matrix;

[0097] Step S33: Based on the environmental feature representation space and the normalized multi-modal feature weight matrix, construct a feature correlation mapping model to obtain a feature correlation mapping matrix;

[0098] Step S34: Perform cross-modal feature interaction and information integration on the feature correlation mapping matrix to obtain an intermediate result of feature interaction;

[0099] Step S35: Based on the intermediate result of feature interaction, reconstruct the multi-dimensional heterogeneous features to obtain an initial comprehensive feature space;

[0100] Step S36: Perform Kappa coefficient consistency test on the initial comprehensive feature space to obtain a comprehensive feature space.

[0101] In the embodiments of the present invention, dimension mapping and feature deconstruction are performed on the standardized environmental feature vectors. The principal component analysis (PCA) method is adopted, and the cumulative contribution rate threshold is set to 95%. The dimensionality of the environmental features is reduced, and the high-dimensional environmental features are projected onto the low-dimensional environmental feature representation space. The key feature patterns are extracted through singular value decomposition (SVD) to form the environmental feature representation space. The feature weights of the multi-modal feature representation matrix are normalized. The min-max normalization method is used to map the weights of each modal feature to the interval [0, 1], obtaining the normalized multi-modal feature weight matrix. Based on the environmental feature representation space and the normalized multi-modal feature weight matrix, the feature correlation is calculated. The Pearson correlation coefficient is used to calculate the linear relationship between the environmental features and the multi-modal features. The calculation formula is where X and Y represent the environmental features and the multi-modal features respectively, and finally the feature association mapping matrix is formed. Cross-modal feature interaction and information integration are performed on the feature association mapping matrix. The weighted average fusion method is used to linearly combine different modal features according to the weight ratio, and the information contribution degree of each feature is calculated through information entropy. The calculation formula is H(X) = -∑p(x i )logp(x i ), where p(x i ) is the probability distribution of the feature x i . The intermediate result of feature interaction is calculated. Based on the intermediate result of feature interaction, the linear discriminant analysis (LDA) method is used to reduce the multi-dimensional heterogeneous features. The objective function of maximizing the class discrimination is set to improve the feature discrimination ability. At the same time, the cosine similarity is used to calculate the similarity between the feature vectors, and the feature set with a similarity lower than 0.85 is screened to reduce redundant features, completing the feature reconstruction and forming the initial comprehensive feature space. The credibility evaluation and consistency test are performed on the comprehensive feature space. The Mahalanobis distance is used to calculate the deviation degree between the feature data and the mean distribution, and the threshold 3.0 is set to eliminate the abnormal feature points. At the same time, the Kappa consistency coefficient is used to calculate the consistency between the feature space and the original feature distribution, and the consistency coefficient threshold is set to 0.75. If it is lower than this threshold, the feature weight allocation is readjusted, and finally the comprehensive feature space is formed.

[0102] The construction of the environmental feature representation space in the present invention enhances the analytical ability of agricultural planting environment parameters, makes the relationship of multi-dimensional environmental variables clearer, and contributes to the accuracy of subsequent data fusion. The feature weight normalization process balances the influence of different modality data in the information fusion process, avoids the over-dominance of a certain modality data on the final result, and improves the scientificity and rationality of the fusion. The construction of the feature correlation mapping matrix makes the connection between environmental features and multi-modal features of crops closer, providing theoretical support for cross-modal data fusion. Cross-modal feature interaction and information integration enhance the synergy between different modality data, fully explore the hidden feature correlations, and improve the effectiveness of feature fusion. The reduction and reconstruction of multi-dimensional heterogeneous features improve the data expression ability, reduce redundant information at the same time, optimize the computing efficiency, and make data processing more efficient. The credibility evaluation and consistency test ensure the stability and reliability of the comprehensive feature space, reduce the influence of noise interference on agricultural data analysis, improve the credibility of the comprehensive features, and provide a high-quality data basis for subsequent agricultural ecosystem modeling.

[0103] Preferably, step S4 includes the following steps:

[0104] Step S41: Conduct a topological structure analysis on the comprehensive feature space, construct a multi-dimensional feature correlation network, and obtain an ecological system topological mapping;

[0105] Step S42: Based on the ecological system topological mapping, conduct mathematical modeling and parameter definition of the key elements of the agricultural ecosystem to obtain an initial ecological system mathematical model;

[0106] Step S43: Conduct dynamic parameter calibration and constraint condition optimization on the initial ecological system mathematical model to obtain a parameter-optimized ecological system mathematical model, where the mean square error (MSE) of the dynamic parameter calibration is less than 0.1, and the R 2 goodness of fit is greater than 0.8, the parameter change range of the constraint condition optimization is ±20%, and the model stability is that the coefficient of variation is less than 0.15;

[0107] Step S44: Based on the parameter-optimized ecological system mathematical model, construct a multi-dimensional state space simulation framework to obtain a digital twin ecological system simulation model, where the sampling frequency of the state variables of the multi-dimensional state space simulation framework is 1 - 4 times per time step;

[0108] Step S45: Conduct iterative verification and error correction on the digital twin ecological system simulation model to obtain a corrected digital twin ecological system simulation model;

[0109] Step S46: Based on the corrected digital twin ecological system simulation model, conduct scenario deduction and prediction of the agricultural ecosystem to obtain an agricultural ecosystem prediction model.

[0110] In the embodiments of the present invention, topological structure analysis is performed on the comprehensive feature space. A network construction method based on the minimum spanning tree (MST) is adopted to calculate the Euclidean distance between each feature node, and the connection threshold is set to 0.7. Strongly correlated features are screened to construct a multi-dimensional feature association network. The adjacency matrix is used to represent the connection relationship between feature nodes, and finally an ecosystem topological mapping is formed. Based on the ecosystem topological mapping, mathematical modeling is carried out for the key elements of the agricultural ecosystem. Mathematical expressions for crop growth rate, dynamic changes in soil nutrients, water evapotranspiration balance, and photosynthetic efficiency are set. Among them, the crop growth rate is modeled using the Logistic growth equation, and the calculation formula where W is the crop biomass, r is the growth rate coefficient, and K is the environmental carrying capacity. At the same time, diffusion equations for soil nitrogen, phosphorus, and potassium nutrients and a water balance equation are set to obtain an initial ecosystem mathematical model. Dynamic parameter calibration and constraint condition optimization are performed on the initial ecosystem mathematical model. The least squares method (OLS) is used for parameter estimation, and the mean square error (MSE) between the model output and the measured data is calculated to ensure that it is less than 0.1. At the same time, the coefficient of determination (R 2 ) is calculated to verify the model fitting degree and ensure that it is greater than 0.8. The parameter change range for constraint condition optimization is set to ±20%. The genetic algorithm (GA) is used to optimize the model parameters to make the coefficient of variation of model stability less than 0.15, and a parameter-optimized ecosystem mathematical model is obtained. Based on the parameter-optimized ecosystem mathematical model, a multi-dimensional state space simulation framework is constructed. The state variables include temperature, humidity, light intensity, soil moisture, and crop growth stage. A dynamic update mechanism with a time step of 1 hour is adopted, and the state variable sampling frequency is set to 1 - 4 times per time step. Based on the finite difference method, the dynamic change process is solved to form a digital twin ecosystem simulation model. Iterative verification and error correction are performed on the digital twin ecosystem simulation model. The Kalman filter algorithm is used to fuse the simulation results and the measured data, and the mean square error (MSE) is calculated. If the error exceeds the set threshold of 0.05, the state variable weights are adjusted and recalculated until the error meets the set standard, and a corrected digital twin ecosystem simulation model is obtained. Based on the corrected digital twin ecosystem simulation model, different environmental variable change scenarios are set, including a 5% increase in air temperature, a 10% decrease in precipitation, and a 20% increase in fertilization amount. The state transition matrix is used to calculate the dynamic change process of the ecosystem under each scenario, and the future crop growth trend, pest and disease risk, and soil nutrient evolution are deduced. Finally, an agricultural ecosystem prediction model is obtained.

[0111] The construction of the multi-dimensional feature association network in the present invention enhances the correlation analysis ability among the key elements of the agricultural ecosystem, makes the topological mapping of the ecosystem more accurate, and provides a solid data basis for subsequent mathematical modeling. The mathematical modeling and parameter definition of the key elements of the agricultural ecosystem enable clear mathematical descriptions of the core variables and influencing factors of the ecosystem, improving the interpretability and modeling accuracy of the system. The dynamic parameter calibration and constraint condition optimization reduce the model error, improve the fitting degree, ensure the rationality of the parameters and the stability of the calculation results, and enable the model to more accurately simulate the actual state of the agricultural ecosystem. The construction of the multi-dimensional state space simulation framework endows the ecosystem simulation with higher dynamics and scalability, and improves the fineness and adaptability of the digital twin ecosystem simulation model. The process of iterative verification and error correction ensures the accuracy and stability of the simulation model after multiple rounds of calculations, and effectively improves the prediction ability of the model in different scenarios. The scenario deduction and prediction of the agricultural ecosystem realize the simulation analysis of different planting schemes, improve the scientificity and reliability of the prediction, and provide accurate data support and optimized decision-making basis for agricultural production.

[0112] Preferably, step S41 includes the following steps:

[0113] Step S411: Decompose the dimension of the comprehensive feature space and analyze the feature relationship to obtain the initial feature association mapping;

[0114] Step S412: Evaluate the connectivity of the multi-dimensional feature node network based on the initial feature association mapping to obtain the feature association strength matrix;

[0115] Step S413: Perform weight normalization and hierarchical clustering on the feature association strength matrix to obtain the topological structure of the feature association network;

[0116] Step S414: Conduct correlation measurement analysis of the key elements of the agricultural ecosystem based on the topological structure of the feature association network to obtain the dynamic correlation data of the elements;

[0117] Step S415: Conduct interdependence analysis based on the dynamic correlation data of the elements to obtain the key node correlation index;

[0118] Step S416: Calculate the network centrality and evaluate the influence of the key node correlation index to obtain the feature network centrality mapping;

[0119] Step S417: Construct the agricultural ecosystem network based on the feature network centrality mapping to obtain the topological mapping of the ecosystem.

[0120] In the embodiments of the present invention, the dimensional deconstruction and feature relationship analysis of the comprehensive feature space are carried out. The principal component analysis (PCA) is used to reduce the dimension of the features, calculate the variance contribution rate of each feature, and screen the features with the cumulative contribution rate greater than 85% as the main feature subset. The Pearson correlation coefficient is used to calculate the correlation between each feature, the significance level is set to 0.05, and the initial mapping of feature association is constructed. Based on the initial mapping of feature association, the network connectivity of multi-dimensional feature nodes is calculated, the weighted degree centrality method is used to calculate the association strength between features, the association weight threshold between features is set to 0.6, and the feature connections below the threshold are screened out to obtain the feature association strength matrix. The weights of the feature association strength matrix are normalized, the Z-score normalization method is used to map the weights to the interval [-1, 1], and the features are clustered based on hierarchical clustering analysis (HCA). The Euclidean distance is set as the similarity measurement standard, and the bottom-up agglomerative hierarchical clustering method is used until all features converge to five main clusters, and the topological structure of the feature association network is constructed. Based on the topological structure of the feature association network, the association measurement of the key elements of the agricultural ecosystem is analyzed, the dynamic association between crop growth, pest and disease risk, soil nutrient flow and meteorological environment is calculated, and the sliding window cross mutual information method is used to calculate the degree of change of the association over time. The size of the sliding window is set to 7 days and the step size is 1 day to obtain the element dynamic association data. Based on the element dynamic association data, the mutual dependence between key elements is calculated, the Granger causality analysis method is used to verify the influence direction and influence time lag of different features on the crop growth state, determine the causal relationship strength of each feature, and the features with the causal influence coefficient greater than 0.3 are used as strong dependence relationships to obtain the key node association index. The network centrality calculation and influence evaluation of the key node association index are carried out. The betweenness centrality, degree centrality and eigenvector centrality are used to comprehensively calculate the importance of the key features. The calculation formula is where C B (n) is the betweenness centrality of node n, σ st is the number of shortest paths from node s to node t, σ st (n) is the number of shortest paths passing through node n. The features ranked in the top 10% in the calculation results are defined as core nodes to obtain the feature network centrality mapping. Based on the feature network centrality mapping, the agricultural ecosystem network is constructed, the connection relationship between features is stored by the adjacency matrix, the force-directed layout algorithm is used to visualize the feature network, and the edge weight value of the connection relationship is set to the calculated normalized weight value, and finally the topological mapping of the ecosystem is formed.

[0121] Through dimensional deconstruction and feature relationship analysis, the present invention can transform complex agricultural ecosystem data into an initial mapping of feature associations that is easier to understand, providing a clear data framework for subsequent in-depth analysis. The assessment of the connectivity of the multi-dimensional feature node network enhances the ability to identify the mutual relationships between features, enabling the quantification of the association strength between different features and providing support for the further feature association strength matrix. Weight normalization and hierarchical clustering of the feature association strength matrix help to more accurately identify the influence degree of each feature and reasonably group related features, thus optimizing the construction of the feature association network topology and improving the association degree between features and the operability of the network. The analysis of the association metrics of the key elements of the agricultural ecosystem quantifies the interaction relationships between the elements in the ecosystem, providing a strong basis for dynamic association data and further providing a more accurate reference for system decision-making. The mutual dependence analysis deeply explores the mutual relationships between key elements and generates specific key node association indicators, which provides an efficient node analysis tool for agricultural management. Through network centrality calculation and influence assessment, it is possible to clearly understand which nodes have the greatest impact on the system and help optimize the management strategy of the agricultural ecosystem. Finally, the agricultural ecosystem network constructed based on the feature network centrality mapping provides a clear view of the topology of the entire ecosystem, enabling the system to be more accurate and efficient in subsequent decision-making and adjustment.

[0122] Particularly importantly, the dynamic parameter calibration and constraint condition optimization of the initial ecosystem mathematical model include:

[0123] Construct a parameter sensitivity analysis framework based on the initial ecosystem mathematical model, conduct a global sensitivity analysis of the model parameters, and obtain a parameter sensitivity ranking matrix;

[0124] Select a key parameter set according to the parameter sensitivity ranking matrix and design a multi-level parameter calibration experiment to obtain a parameter calibration experimental plan;

[0125] Execute an iterative parameter optimization process based on the parameter calibration experimental plan, use the Bayesian optimization algorithm to dynamically adjust the parameters, and obtain an intermediate result of parameter optimization;

[0126] Calculate the mean square error (MSE) of the intermediate result of parameter optimization. When the MSE is less than 0.1, perform an R 2 goodness-of-fit assessment to obtain a preliminary calibrated parameter set;

[0127] Define the boundary of the constraint conditions based on the preliminary calibrated parameter set, set the parameter change range to ±20%, construct an objective function for constraint condition optimization, and obtain a definition of the constraint optimization problem;

[0128] The Lagrange multiplier method and the gradient descent algorithm are used to solve the definition of the constrained optimization problem, and an optimized parameter set that satisfies the constraint conditions is obtained;

[0129] Based on the optimized parameter set, the initial mathematical model of the ecosystem is reconstructed and verified to obtain a mathematically modeled ecosystem with optimized parameters.

[0130] In the embodiment of the present invention, first, a parameter sensitivity analysis framework is constructed, and the Sobol method is used to perform global sensitivity analysis on the initial mathematical model of the ecosystem, calculate the first-order sensitivity index S1 and the total sensitivity index S1, and screen out the key parameters with S t greater than 0.05, and arrange all parameters in descending order of S t values to form a parameter sensitivity ranking matrix; secondly, a key parameter set is determined based on the parameter sensitivity ranking matrix, and a full-factor experimental scheme is designed. The value range of the key parameters is set according to historical observation data, and the sampling step size is 0.01. 10,000 groups of experimental data are generated and uniformly sampled using the Latin hypercube sampling method to form a parameter calibration experimental scheme; subsequently, an iterative parameter optimization process is executed, the Bayesian optimization algorithm is used, and the Gaussian process regression is used to establish the mapping relationship between the parameters and the objective function value. The objective function is selected as the mean square error of the ecosystem prediction error where yi is the true value, is the model prediction value, n is the number of samples, and dynamic parameter adjustment is performed. When the MSE is less than 0.1, calculate R 2 goodness of fit, and the calculation formula is where is the mean value of the observed values. When R 2 is greater than 0.8, a preliminary calibrated parameter set is obtained; then, based on the preliminary calibrated parameter set, the constraint conditions are optimized, the parameter change range is set to ±20%, and the constraint optimization objective function is defined and the Lagrange multiplier method constrained optimization problem L(x,λ) = f(x) + λ(Σgi(x)) is constructed, where gi(x) is the constraint condition, and the gradient descent algorithm is used to calculate the optimal solution of, and an optimized parameter set that satisfies the constraint conditions is obtained; finally, based on the optimized parameter set, the initial mathematical model of the ecosystem is reconstructed, the Bootstrap method is used to evaluate the model error, the K-fold cross-validation is used to ensure the generalization ability of the model, and the historical data backtest is used to verify the model stability, and finally a mathematically modeled ecosystem with optimized parameters is obtained.

[0131] The present invention identifies key parameters through global sensitivity analysis, ensuring that the focus of the calibration work is on the variables that have the greatest impact on the model, thereby improving the efficiency and accuracy of parameter optimization. By adopting a multi-level parameter calibration experiment, the parameter adjustment process becomes more systematic, which helps to capture the dynamic changes of complex ecosystems and improve the adaptability of the model to different environmental conditions. Using the Bayesian optimization algorithm for dynamic parameter adjustment can reduce the computational cost, accelerate the optimization convergence speed, and make the parameter optimization more accurate and reliable. Through the calculation of the mean square error (MSE) and the R 2 goodness-of-fit evaluation, the error in the parameter adjustment process is guaranteed to be controlled within a reasonable range, ensuring the rationality and stability of the calibrated parameters. By setting reasonable parameter change ranges and optimization objective functions, the constraint conditions are made more in line with the actual operation laws of the ecosystem, improving the practical applicability of the model prediction. Combining the Lagrange multiplier method with the gradient descent algorithm improves the numerical stability of the solution process, ensuring that the optimized parameter set can meet the actual constraint conditions of the ecosystem. Through the reconstruction and verification of the optimized parameter set, it is ensured that the finally obtained mathematical model of the ecosystem can more accurately simulate the real ecological process, providing accurate data support for environmental management and ecological regulation.

[0132] Preferably, the growth risk assessment of crops based on the agricultural ecosystem prediction model in step S5 includes the following steps:

[0133] Perform weight decomposition and importance ranking of risk characteristic factors for the agricultural ecosystem prediction model to obtain a risk characteristic weight vector;

[0134] Based on the risk characteristic weight vector, conduct a multi-dimensional quantitative assessment of risk factors in the crop growth process to obtain a risk factor assessment matrix;

[0135] The number of rows of the matrix of the risk factor assessment matrix is the corresponding number of risk characteristic factors, the number of columns of the matrix is the risk level dimension, and the sum of each row is always equal to 100;

[0136] Conduct probability distribution analysis and critical value calibration on the risk factor assessment matrix to obtain a risk level determination criterion, where the risk levels are divided into low risk 0 - 25 points, medium risk 26 - 50 points, high risk 51 - 75 points, and extremely high risk 76 - 100 points;

[0137] Based on the risk level determination criterion, calculate the comprehensive risk of the environmental adaptability, physiological response, and resource constraints in the crop growth process to obtain a crop growth risk index;

[0138] In the crop growth risk index, environmental adaptability accounts for 30%, physiological response accounts for 40%, and resource constraints account for 30%;

[0139] Perform threshold mapping and risk grading on the crop growth risk index to obtain a hierarchical risk assessment result;

[0140] Based on the hierarchical risk assessment result, conduct a systematic diagnosis of crop growth risk and generate early warning information to obtain a crop growth risk early warning report.

[0141] In the embodiments of the present invention, for the assessment of crop growth risk based on the agricultural ecosystem prediction model, first, decompose the weights of risk characteristic factors of the agricultural ecosystem prediction model, calculate the weights of each risk characteristic factor using the entropy weight method, and set the risk characteristic factors to include air temperature, humidity, precipitation, light intensity, soil moisture, soil nutrients, pest and disease index, and crop physiological state variables. Calculate the weight values based on information entropy, and the calculation formula is Where represents the information entropy of the jth feature, and p ij represents the normalized value of the ith sample on the jth feature, and finally obtain the risk characteristic weight vector. Based on the risk characteristic weight vector, conduct a multi-dimensional quantitative assessment of risk elements in the crop growth process, establish a risk factor assessment matrix using the fuzzy comprehensive evaluation method, where the number of rows of the matrix corresponds to the number of risk characteristic factors, the number of columns of the matrix is the risk grade dimension, the sum of the row vectors of the matrix is always equal to 100, and the matrix elements represent the normalized assessment value of a certain risk factor at a specific risk grade. Conduct a probability distribution analysis and critical value calibration on the risk factor assessment matrix, set the risk grade boundaries using the percentile method, and divide the risk grades into low risk (0 - 25 points), medium risk (26 - 50 points), high risk (51 - 75 points), and extremely high risk (76 - 100 points). Based on the risk grade determination standard, calculate the comprehensive risk of environmental adaptability, physiological response, and resource constraints in the crop growth process. The environmental adaptability risk involves temperature, humidity, light, and precipitation factors, accounting for 30%, the physiological response risk involves crop growth rate, chlorophyll content, and pest and disease impact, accounting for 40%, and the resource constraint risk involves soil moisture and nutrient supply, accounting for 30%. The calculation formula is R = 0.3R env + 0.4R phy + 0.3R res , where R is the crop growth risk index, R env 、R phy and R resThey are the environmental adaptability, physiological response, and resource constraint risk scores respectively. Threshold mapping and risk grading are performed on the crop growth risk index. The grading standard is set to be consistent with the risk level determination standard, and linear normalization mapping is carried out to normalize the original risk index to the interval [0, 100], obtaining the hierarchical risk assessment result. Based on the hierarchical risk assessment result, a systematic diagnosis of the crop growth risk is carried out. Logistic regression analysis is used to calculate the influence probability of different risk factors on the crop growth state, generating early warning information. According to the change trend of the risk index, a risk time series curve is drawn, and risk trend prediction is carried out in combination with historical data. Finally, a crop growth risk early warning report is formed.

[0142] Through the weight decomposition and importance ranking of risk characteristic factors of the agricultural ecosystem prediction model, the present invention can accurately identify which factors have the greatest impact on crop growth risks, providing a clear basis for subsequent risk assessment. Based on the multi-dimensional quantitative assessment of the risk characteristic weight vector, quantitative data support is provided for each risk element in the crop growth process, and then a risk factor assessment matrix is generated to help comprehensively understand the distribution and influence of each risk factor. Probability distribution analysis and critical value calibration are carried out on the risk factor assessment matrix, making the division of risk levels more accurate, ensuring the scientificity and rationality of the risk judgment standard. The clear division of risk levels (low, medium, high, extremely high) provides clear guidance for crop risk control. By calculating the comprehensive risks of environmental adaptability, physiological response, and resource constraint in the crop growth process, the generated growth risk index provides an intuitive and quantitative risk indicator for the potential risks in the crop growth process, and according to the specific weight distribution, the influence of each factor is reasonably integrated. Threshold mapping and risk grading of the crop growth risk index help to clearly identify the risk levels of crops at different growth stages, providing an important reference for subsequent decision-making. Finally, through the systematic diagnosis of the hierarchical risk assessment result and the generation of early warning information, a timely and scientific risk early warning report is provided for crop risk prevention and control and precise intervention.

[0143] Preferably, the step of predicting crop diseases and pests based on the agricultural ecosystem prediction model in step S5 includes the following steps:

[0144] Classify and code the risk characteristics of crop diseases and pests of the agricultural ecosystem prediction model to obtain the disease and pest characteristic vectors;

[0145] Based on the disease and pest characteristic vectors, perform multi-dimensional correlation analysis on the occurrence probability of crop diseases and pests to obtain the disease and pest potential risk matrix;

[0146] Perform spatio-temporal sequence prediction modeling on the disease and pest potential risk matrix to obtain the disease and pest evolution trend prediction model;

[0147] Predict the occurrence areas, severity levels, and transmission paths of crop pests and diseases based on the pest and disease evolution trend prediction model to obtain the pest and disease risk assessment results;

[0148] Classify and perform threshold mapping on the pest and disease risk assessment results to obtain the pest and disease early warning level list;

[0149] Extract pest and disease risk information and generate early warning information based on the pest and disease early warning level list to obtain the crop pest and disease early warning report.

[0150] In the embodiments of the present invention, for crop pest and disease prediction based on the agricultural ecosystem prediction model, first, classify the pest and disease risk characteristics of the agricultural ecosystem prediction model. Adopt the feature classification method based on decision trees and set the pest and disease risk characteristics, including environmental temperature and humidity, precipitation, light intensity, wind speed, soil moisture content, soil nutrients, historical incidence of pests and diseases, and crop growth state variables, and convert them into pest and disease feature vectors using the One-Hot encoding or label encoding method. Based on the pest and disease feature vectors, conduct multi-dimensional correlation analysis of the occurrence probability of crop pests and diseases. Calculate the correlation between different environmental and crop growth factors and the occurrence probability of pests and diseases using the Pearson correlation coefficient. The calculation formula is where X i represents the value of a certain environmental factor, and Y i represents the occurrence probability of pests and diseases, and finally form the pest and disease potential risk matrix. Perform spatio-temporal sequence prediction modeling on the pest and disease potential risk matrix. Adopt the moving window method to construct the time series data set, and set the sliding step length to 7 days. Calculate the pest and disease occurrence trend based on the ARIMA (Autoregressive Integrated Moving Average) method, and set the model parameters to p = 2, d = 1, q = 2. Finally, obtain the pest and disease evolution trend prediction result. Based on the pest and disease evolution trend prediction result, predict the occurrence areas, severity levels, and transmission paths of pests and diseases. Set the division criteria for the occurrence areas of pests and diseases, and perform spatial interpolation calculation based on the GIS geographic information system combined with historical pest and disease data. Calculate the spatial distribution of the occurrence probability of pests and diseases using the Kriging interpolation method. The calculation formula is where Z(x) represents the occurrence probability of pests and diseases at the interpolation point, and w iIt represents the spatial weight coefficient and finally generates the pest and disease risk assessment results. The pest and disease risk assessment results are classified and threshold mapped. According to the occurrence probability of pests and diseases, the threshold range is set, and the pest and disease early warning levels are divided, including low risk (0 - 25%), medium risk (26 - 50%), high risk (51 - 75%), and extremely high risk (76 - 100%). The piecewise linear mapping method is used for numerical conversion to obtain the pest and disease early warning level list. Based on the pest and disease early warning level list, the pest and disease risk information is extracted. The TF-IDF (Term Frequency - Inverse Document Frequency) method is used to analyze the historical pest and disease records of the agricultural ecosystem, extract the high-frequency pest and disease risk characteristics, and combine with the changing trend of the pest and disease occurrence probability to set the early warning trigger conditions, forming a pest and disease early warning rule set. Finally, based on the rule set, a crop pest and disease early warning report is automatically generated, including the pest and disease risk level, affected area, potential transmission path, and recommended control measures.

[0151] Through the classification and coding of pest and disease risk characteristics for the agricultural ecosystem prediction model, the present invention can systematically organize various characteristics related to pests and diseases, form a pest and disease characteristic vector, and provide accurate input data for subsequent pest and disease prediction. Based on the multi-dimensional correlation analysis of the pest and disease characteristic vector, the potential relationships between different factors can be identified and quantified, and then a pest and disease potential risk matrix can be constructed to provide a clear quantitative assessment of the potential risks of crop pests and diseases. Through spatio-temporal sequence prediction modeling and in-depth analysis of the pest and disease potential risk matrix, the evolution trend of pests and diseases can be predicted, revealing the spatio-temporal variation law of the occurrence of pests and diseases, and providing an accurate trend prediction model for the occurrence prediction of crop pests and diseases. Based on the pest and disease evolution trend prediction model, the occurrence area, severity, and transmission path of crop pests and diseases can be predicted, providing accurate risk assessment results for crop production and helping to take targeted prevention and control measures. By classifying and threshold mapping the pest and disease risk assessment results, the risk levels of pests and diseases can be clearly divided, providing a clear early warning level list, and providing specific response guidance for decision-makers. Finally, based on the pest and disease early warning level list, the pest and disease risk information is extracted and early warning information is generated to form a scientific crop pest and disease early warning report, helping agricultural managers to take corresponding prevention and control measures in a timely manner to ensure the healthy growth of crops.

[0152] Preferably, step S6 includes the following steps:

[0153] Step S61: Systematically analyze and prioritize the risk factors of the crop early warning report to obtain a set of key control indicators;

[0154] Step S62: Optimize the allocation of agricultural planting environmental resources based on the set of key control indicators to obtain an environmental parameter control plan;

[0155] Step S63: Conduct scenario simulation verification on the environmental parameter regulation plan, and perform effect performance evaluation based on the scenario simulation verification results to obtain an alternative set of planting management plans;

[0156] Step S64: Perform dynamic weight allocation of crop production factors based on the alternative set of planting management plans to obtain an optimized resource allocation model;

[0157] Step S65: Conduct a feasibility analysis of planting benefits for the optimized resource allocation model to obtain a feasibility report of the allocation model;

[0158] Step S66: Conduct a risk hedging assessment based on the feasibility report of the allocation model to obtain a risk hedging factor;

[0159] Step S67: Determine a risk-reduced planting management plan according to the risk hedging factor;

[0160] Step S68: Conduct agricultural planting optimization regulation based on the risk-reduced planting management plan to obtain an optimized planting management plan.

[0161] In the embodiment of the present invention, for agricultural planting optimization regulation based on the crop warning report, first, a systematic analysis of risk factors in the crop warning report is carried out, risk factors of environmental adaptability, physiological response, and resource constraints are extracted, and the weights of each risk factor are calculated by the analytic hierarchy process (AHP). The calculation formula is where A i represents the risk factor score, and a set of key regulation indicators is obtained. Based on the set of key regulation indicators, the linear programming method is used to optimize the allocation of agricultural planting environmental resources, an environmental parameter regulation objective function is constructed, and the regulation range of environmental factors is set, including temperature and humidity (15 - 30 °C, 40% - 80%), light intensity (10000 - 50000 lux), soil pH (5.5 - 7.5), nutrient concentration (nitrogen 10 - 50 mg / kg, phosphorus 5 - 30 mg / kg, potassium 10 - 60 mg / kg), and the optimal environmental parameter regulation plan is solved. Conduct multi-scenario simulation verification on the environmental parameter regulation plan, use the Monte Carlo simulation method to conduct 10000 times of random perturbation tests, calculate the impact of environmental parameter fluctuations on crop growth, evaluate the yield, the incidence of pests and diseases, and the resource utilization rate, and form an alternative set of planting management plans. Based on the alternative set of planting management plans, adopt a dynamic weight allocation strategy for production factors to construct an optimized resource allocation model based on the Lagrange multiplier method. The objective function is maxY = f(X1, X2,..., X n) The constraint conditions include water resource utilization rate (60%-90%), chemical fertilizer usage (reduced by 10%-30%), and labor input (reduced by 5%-20%). Solve the optimized resource allocation model. Conduct a feasibility analysis of the optimized resource allocation model, use the sensitivity analysis method to evaluate the impact of changes in each parameter on planting benefits, and conduct a risk hedging assessment. Set the revenue-risk ratio to be greater than 1.5 and control the production cost fluctuation range within ±15% to form a risk reduction planting management plan. Based on the risk reduction planting management plan, adopt a closed-loop control strategy to optimize and regulate agricultural planting, set up an environmental monitoring feedback mechanism, collect environmental data every 6 hours, and adjust the temperature, humidity, light, water, and fertilizer supply in real time to ensure a stable growth environment for crops. Finally, form an optimized planting management plan to achieve optimized regulation of the entire process of agricultural planting.

[0162] The present invention systematically analyzes risk factors and ranks their priorities for the crop warning report, can effectively identify the key factors that have the greatest impact on crop growth risks, rank their priorities, form a set of key regulation indicators, and provide a clear direction for subsequent management decisions. Optimize the allocation of agricultural planting environmental resources based on the set of key regulation indicators. Through a scientific resource allocation plan, the resource utilization efficiency can be maximized, the growth environment of crops can be ensured to reach the best state, and the production benefits can be improved. Verify the environmental parameter regulation plan through multi-scenario simulation verification and performance evaluation, can evaluate the actual effects of different regulation plans, ensure that the selected plan can effectively improve the planting management efficiency in multiple scenarios, and finally obtain a set of alternative planting management plans to provide multiple feasible options for decision-making. Dynamically configure the dynamic weights of crop production factors based on the set of alternative planting management plans. Through fine-grained dynamic adjustment, the resources can be optimally allocated, ensure that the allocation of production factors is both reasonable and efficient, optimize the resource allocation model, and improve the production potential of crops. Conduct a feasibility analysis and risk hedging assessment of the optimized resource allocation model, can identify potential risks and prevent them, formulate effective risk response measures, obtain a risk reduction planting management plan, and ensure the sustainable development of agricultural planting while reducing risks. Finally, based on the risk reduction planting management plan, conduct optimized regulation of agricultural planting to obtain an optimized planting management plan, provide a comprehensive and scientific management plan for agricultural production, and can improve the stability and economic benefits of crop production.

[0163] Preferably, the present invention also provides an agricultural planting optimization system based on big data analysis for implementing the above-mentioned agricultural planting optimization method based on big data analysis. The agricultural planting optimization system based on big data analysis includes:

[0164] An agricultural environment perception and data preprocessing module for obtaining agricultural planting environment parameter data, preprocessing and standardizing it to obtain a standardized environmental feature vector;

[0165] A multimodal agricultural perception module, which is used to acquire image, sound and odor data of crops; extract crop growth characteristics from the image, sound and odor data to generate a multimodal feature representation matrix;

[0166] A multimodal data fusion module, which is used to perform multimodal data fusion based on the standardized environmental feature vector and the multimodal feature representation matrix to obtain a comprehensive feature space;

[0167] A digital twin agricultural ecological modeling module, which is used to model a digital twin ecosystem for the comprehensive feature space to generate an agricultural ecosystem prediction model;

[0168] A crop intelligent early warning module, which is used to evaluate the growth risk of crops based on the agricultural ecosystem prediction model to obtain a crop growth risk early warning report; predict crop diseases and pests based on the agricultural ecosystem prediction model to obtain a crop diseases and pests early warning report; record the crop growth risk early warning report and the crop diseases and pests early warning report as the crop early warning report;

[0169] An intelligent agricultural optimization and regulation module, which is used to perform agricultural planting optimization and regulation according to the crop early warning report to obtain an optimized planting management plan.

[0170] The present invention can convert environmental data from different sources and with different units into a unified standard format by efficiently acquiring parameter data of an agricultural planting environment, performing preprocessing and standardization, ensuring the comparability and accuracy of the data, and providing a reliable basis for subsequent analysis. By collecting image, sound and odor data of crops and using deep feature extraction technology to convert these data into a highly information-based feature matrix, multi-dimensional monitoring data for the growth state of crops and environmental changes can be provided, making the analysis results more comprehensive and accurate. By combining the standardized environmental feature vector with the multimodal feature representation matrix, a comprehensive feature space is created, which can fuse data from different sources into an overall analysis framework, improving the comprehensiveness and accuracy of data processing and providing a more comprehensive perspective for the modeling of agricultural ecosystems. Using the comprehensive feature space to generate an agricultural ecosystem prediction model can accurately simulate the change process of the agricultural ecological environment and provide a theoretical basis for the risk assessment of crop growth and the prediction of crop diseases and pests. Based on the agricultural ecosystem prediction model, the risks in the growth process of crops can be identified in a timely manner, and growth risk and diseases and pests early warning reports can be generated, so as to provide preventive intervention measures for farmers and reduce crop losses. Using the crop early warning report to perform agricultural planting optimization and regulation, through a scientific data-driven method, an optimized planting management plan is proposed to ensure the high efficiency and stability of agricultural production and minimize the negative impact of environmental factors on crop growth.

[0171] Therefore, in all aspects, the embodiments should be regarded as exemplary and non-limiting. The scope of the present invention is not defined by the above description. Therefore, all changes falling within the meaning and scope of the equivalent elements of the application documents are intended to be encompassed within the present invention.

[0172] The above are only specific embodiments of the present invention, enabling those skilled in the art to understand or implement the present invention. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to these embodiments shown herein, but rather to the broadest scope consistent with the principles and novel features invented herein.

Claims

1. An agricultural planting optimization method based on big data analysis, characterized in that, It includes the following steps: Step S1: Obtain agricultural planting environment parameter data, perform preprocessing and standardization to obtain a standardized environmental feature vector; Step S2: Obtain image, sound, and odor data of crops; extract crop growth characteristics from the image, sound, and odor data to generate a multi-modal feature representation matrix; Step S3: Perform multi-modal data fusion based on the standardized environmental feature vector and the multi-modal feature representation matrix to obtain a comprehensive feature space; Step S4: Model a digital twin ecosystem for the comprehensive feature space to generate an agricultural ecosystem prediction model; Step S5: Based on the agricultural ecosystem prediction model, evaluate the growth risk of crops to obtain a crop growth risk warning report; Based on the agricultural ecosystem prediction model, predict crop diseases and pests to obtain a crop diseases and pests warning report; Record the crop growth risk warning report and the crop diseases and pests warning report as the crop warning report; Step S6: According to the crop warning report, perform optimized regulation of agricultural planting to obtain an optimized planting management plan.

2. The agricultural planting optimization method based on big data analysis according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Deploy a modular sensor array in the agricultural planting area and use the modular sensor array to collect temperature, humidity, and light intensity to obtain agricultural planting physical parameter data; Step S12: Based on the agricultural planting physical parameter data, use a multi-layer soil sensor probe to collect the soil moisture content at 0-100 cm to obtain agricultural planting soil layer data, where the layer sampling interval is 10 cm; Step S13: Based on the agricultural planting soil layer data, detect the concentrations of nitrogen, phosphorus, and potassium elements in the soil to obtain agricultural planting environment parameter data; Step S14: Perform outlier detection and normalization processing on the agricultural planting environment parameter data to obtain a standardized interval mapping, where the high-precision calibration reference source for the normalization processing is a temperature standard blackbody furnace, a humidity generator, and a spectral radiation standard source; Step S15: Perform feature selection and dimensionality reduction on the standardized interval mapping to obtain a key environmental feature subset; Step S16: Based on the key environmental feature subset, perform probability distribution statistics and vector transformation on the probability distribution statistics results to obtain a standardized environmental feature vector.

3. The agricultural planting optimization method based on big data analysis according to claim 2, wherein Step S2 includes the following steps: Step S21: According to the agricultural planting environment parameter data, use a ground camera to collect crop leaf characteristics; detect the leaf layer distribution of the crop leaf characteristics to obtain leaf spatial distribution data; Step S22: Perform statistics on the distribution law of crop leaf characteristics, and use the statistical results of the leaf characteristic distribution law to determine the crop plant morphology; quantify the mechanical relationship between the stem bending stiffness and the branching angle using the crop plant morphology to obtain stem branching data; Step S23: Calculate the Pearson correlation coefficient of plant height, leaf area, and fruit size based on the crop plant morphology, and use the calculation results of the Pearson correlation coefficient to determine the crop fruit characteristics; based on the crop fruit characteristics, perform volatile organic compound detection to obtain fruit physiological state data; Step S24: Calculate the sugar-acid ratio of the crop fruits based on the fruit characteristics, and determine the root characteristics of the crops using the calculation result of the fruit sugar-acid ratio; calculate the root growth range using the root characteristics of the crops to obtain the root spatial distribution data; Step S25: Based on the leaf spatial distribution data, stem branching data, fruit physiological state data, and root spatial distribution data, use a meteorological sensor to collect images of the crop growth environment, and perform preprocessing and enhancement to obtain standardized image data; perform convolutional feature extraction on the standardized image data to obtain an image feature sub-matrix; Step S26: Deploy a multi-node acoustic sensor array and an acoustic omnidirectional pickup array through a soil hydrophone to collect sound signals from the crop growth environment, and perform noise reduction and spectrum enhancement to obtain standardized sound data; perform time-frequency domain feature extraction on the standardized sound data to obtain a sound feature sub-matrix; Step S27: Construct an odor sensor network through a MOS sensor to collect odor concentrations from the crop growth environment, and perform baseline drift correction and peak extraction to obtain standardized odor data; perform chemical feature extraction on the standardized odor data to obtain an odor feature sub-matrix; Step S28: Perform dynamic feature selection on the image feature sub-matrix, sound feature sub-matrix, and odor feature sub-matrix through an edge computing gateway, and perform multi-modal feature fusion based on the dynamic feature selection result to obtain a multi-modal feature representation matrix.

4. The agricultural planting optimization method based on big data analysis according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Perform dimensionality mapping and feature deconstruction on the standardized environmental feature vector to obtain an environmental feature representation space; Step S32: Perform feature weight normalization on the multi-modal feature representation matrix to obtain a normalized multi-modal feature weight matrix; Step S33: Based on the environmental feature representation space and the normalized multi-modal feature weight matrix, construct a feature association mapping model to obtain a feature association mapping matrix; Step S34: Perform cross-modal feature interaction and information integration on the feature association mapping matrix to obtain an intermediate result of feature interaction; Step S35: Reconstruct the multi-dimensional heterogeneous features based on the intermediate result of feature interaction to obtain an initial comprehensive feature space; Step S36: Perform a Kappa coefficient consistency test on the initial comprehensive feature space to obtain a comprehensive feature space.

5. The agricultural planting optimization method based on big data analysis according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Perform topological structure analysis on the comprehensive feature space, construct a multi-dimensional feature association network, and obtain an ecological system topological mapping; Step S42: Based on the ecological system topological mapping, perform mathematical modeling and parameter definition of the key elements of the agricultural ecosystem to obtain an initial ecological system mathematical model; Step S43: Perform dynamic parameter calibration and constraint condition optimization on the initial ecosystem mathematical model to obtain a parameter-optimized ecosystem mathematical model, where the mean square error (MSE) of the dynamic parameter calibration is less than 0.1, and the R 2 goodness of fit is greater than 0.8, the parameter change range of the constraint condition optimization is ±20%, and the model stability is that the coefficient of variation is less than 0.15; Step S44: Based on the parameter-optimized ecological system mathematical model, construct a multi-dimensional state space simulation framework to obtain a digital twin ecological system simulation model, where the sampling frequency of the state variables of the multi-dimensional state space simulation framework is 1-4 times per time step; Step S45: Perform iterative verification and error correction on the digital twin ecological system simulation model to obtain a corrected digital twin ecological system simulation model; Step S46: Perform scenario deduction and prediction on the agricultural ecosystem based on the corrected digital twin ecosystem simulation model to obtain an agricultural ecosystem prediction model.

6. The agricultural planting optimization method based on big data analysis according to claim 5, wherein, Step S41 includes the following steps: Step S411: Decompose the dimensions of the comprehensive feature space and analyze the feature relationships to obtain an initial mapping of feature associations; Step S412: Evaluate the connectivity of the multi-dimensional feature node network based on the initial mapping of feature associations to obtain a feature association strength matrix; Step S413: Perform weight normalization and hierarchical clustering on the feature association strength matrix to obtain a topological structure of the feature association network; Step S414: Conduct an analysis of the association metrics of the key elements of the agricultural ecosystem based on the topological structure of the feature association network to obtain element dynamic association data; Step S415: Conduct an interdependence analysis based on the element dynamic association data to obtain key node association indicators; Step S416: Calculate the network centrality and evaluate the influence of the key node association indicators to obtain a feature network centrality mapping; Step S417: Construct an agricultural ecosystem network based on the feature network centrality mapping to obtain an ecosystem topological mapping.

7. The agricultural planting optimization method based on big data analysis according to claim 1, characterized in that The growth risk assessment of crops based on the agricultural ecosystem prediction model in Step S5 includes the following steps: Decompose the weights of the risk characteristic factors of the agricultural ecosystem prediction model and rank their importance to obtain a risk characteristic weight vector; Conduct a multi-dimensional quantitative assessment of the risk factors in the crop growth process based on the risk characteristic weight vector to obtain a risk factor assessment matrix; The number of rows of the matrix of the risk factor assessment matrix is the corresponding number of risk characteristic factors, the number of columns of the matrix is the risk level dimension, and the sum of each row is always equal to 100; Conduct a probability distribution analysis and critical value calibration on the risk factor assessment matrix to obtain a risk level determination standard, where the risk levels are divided into low risk 0 - 25 points, medium risk 26 - 50 points, high risk 51 - 75 points, and extremely high risk 76 - 100 points; Calculate the comprehensive risk of the environmental adaptability, physiological response, and resource constraints in the crop growth process based on the risk level determination standard to obtain a crop growth risk index; In the crop growth risk index, environmental adaptability accounts for 30%, physiological response accounts for 40%, and resource constraints account for 30%; Conduct a threshold mapping and risk classification on the crop growth risk index to obtain a hierarchical risk assessment result; Conduct a systematic diagnosis of the crop growth risk and generate early warning information based on the hierarchical risk assessment result to obtain a crop growth risk early warning report.

8. The agricultural planting optimization method based on big data analysis according to claim 1, characterized in that The pest and disease prediction of crops based on the agricultural ecosystem prediction model in Step S5 includes the following steps: Classify and code the pest and disease risk characteristics of the agricultural ecosystem prediction model to obtain a pest and disease characteristic vector; Conduct a multi-dimensional correlation analysis of the occurrence probabilities of crop pests and diseases based on the pest and disease characteristic vector to obtain a pest and disease potential risk matrix; Build a time - space sequence prediction model for the pest and disease potential risk matrix to obtain a pest and disease evolution trend prediction model; Predict the occurrence area, severity, and transmission path of crop pests and diseases based on the pest and disease evolution trend prediction model to obtain a pest and disease risk assessment result; Classify the results of the pest and disease risk assessment and perform threshold mapping to obtain a list of pest and disease warning levels; Extract pest and disease risk information and generate warning information based on the list of pest and disease warning levels to obtain a crop pest and disease warning report.

9. The agricultural planting optimization method based on big data analysis according to claim 1, characterized in that Step S6 includes the following steps: Step S61: Systematically analyze the risk factors of the crop warning report and rank them by priority to obtain a set of key control indicators; Step S62: Optimize the allocation of agricultural planting environmental resources based on the set of key control indicators to obtain an environmental parameter control plan; Step S63: Conduct scenario simulation verification on the environmental parameter control plan and perform effect performance evaluation based on the results of the scenario simulation verification to obtain a set of alternative planting management plans; Step S64: Dynamically configure the weights of crop production factors based on the set of alternative planting management plans to obtain an optimized resource allocation model; Step S65: Conduct a feasibility analysis of the planting benefits of the optimized resource allocation model to obtain a feasibility report of the allocation model; Step S66: Conduct a risk hedging assessment based on the feasibility report of the allocation model to obtain a risk hedging factor; Step S67: Determine a risk reduction planting management plan according to the risk hedging factor; Step S68: Conduct optimized control of agricultural planting based on the risk reduction planting management plan to obtain an optimized planting management plan.

10. An agricultural planting optimization system based on big data analysis, characterized in that, For implementing the agricultural planting optimization method based on big data analysis as described in claim 1, the agricultural planting optimization system based on big data analysis includes: An agricultural environment perception and data preprocessing module, configured to obtain agricultural planting environment parameter data, and perform preprocessing and standardization to obtain a standardized environmental feature vector; A multi-modal agricultural perception module, configured to obtain image, sound, and odor data of crops; extract crop growth characteristics from the image, sound, and odor data to generate a multi-modal feature representation matrix; A multi-modal data fusion module, configured to perform multi-modal data fusion based on the standardized environmental feature vector and the multi-modal feature representation matrix to obtain a comprehensive feature space; A digital twin agricultural ecological modeling module, configured to perform digital twin ecosystem modeling on the comprehensive feature space to generate an agricultural ecosystem prediction model; A crop intelligent warning module, configured to perform growth risk assessment on crops based on the agricultural ecosystem prediction model to obtain a crop growth risk warning report; perform pest and disease prediction on crops based on the agricultural ecosystem prediction model to obtain a crop pest and disease warning report; record the crop growth risk warning report and the crop pest and disease warning report as a crop warning report; An intelligent agricultural optimization control module, configured to perform optimized control of agricultural planting according to the crop warning report to obtain an optimized planting management plan.

Images