Crop growth state evaluation method based on agricultural Internet of Things

Abstract

The invention provides a crop growth state evaluation method based on the agricultural Internet of Things, and the method is characterized in that the method specifically comprises the following steps: S1, collecting the multi-dimensional data of a crop growth environment in real time, and outputting an original data set; s2, aligning the multi-dimensional data according to geographic coordinates, extracting features, and constructing a multi-modal fusion feature vector; s3, automatically identifying the current growth stage of the crop and the confidence of the current growth stage according to the fused feature vector; s4, calculating a crop health degree score and grading according to the growth stage information and the feature data, and identifying a stress state detection result at the same time; s5, according to a quantitative evaluation result, predicting a growth trend and generating a suggested decision scheme; and S6, the decision is monitored, the decision execution effect is fed back to the system, and the model parameters and the decision rules are continuously optimized. A closed-loop mechanism of evaluation, decision, feedback and optimization is formed, the applicability and evaluation precision of the system are continuously improved, and agricultural production is promoted to be upgraded to precision and intelligence.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of crop management, and particularly relates to a crop growth state evaluation method based on an agricultural Internet of Things. BACKGROUND

[0002] Crop management technology refers to a comprehensive technical system for systematically managing the whole life cycle of crops from planting to harvesting by using scientific methods and modern agricultural technical means. The core is to realize precise monitoring and intervention of the growth process of crops by combining modern intelligent equipment and data analysis technology, through soil improvement, water and fertilizer regulation, pest control and growth environment optimization measures, aiming to improve yield, quality and resource utilization efficiency, and reduce environmental pollution. The technology integrates agronomic knowledge, biotechnology and information technology, covering variety selection, seeding and seedling, field management to harvesting and storage links, and is a key support for the transformation of modern agriculture to digitalization and precision.

[0003] The traditional crop growth state evaluation scheme mainly relies on manual observation and single instrument detection, and the core process is to manually measure the crop height, leaf color and shape indicators on site, combine handheld detection equipment to obtain local data, and then judge the growth state according to experience. This method has significant defects, and the data collection efficiency is low. Manual observation cannot realize real-time monitoring of large-area farmland, and handheld instruments need to be measured point by point, resulting in serious lack of data coverage and timeliness. The evaluation standard is highly subjective, and depends on the experience of agricultural technicians to judge the growth stage and health state, lacks a quantitative index system, and the evaluation results of different personnel differ greatly. The decision feedback is disconnected, and the management scheme formulated manually is difficult to accurately match the actual needs of crops, and cannot track the execution effect to form a closed-loop optimization. These defects make it difficult for the traditional method to ensure evaluation accuracy and realize dynamic regulation and control required by precision agriculture in large-scale agricultural production. SUMMARY

[0004] Therefore, the present application aims to provide a crop growth state evaluation method based on an agricultural Internet of Things, which collects multi-dimensional data of the growth environment of crops in real time, outputs an original data set, aligns the multi-dimensional data according to geographical coordinates and extracts features, constructs a multi-modal fusion feature vector, automatically identifies the current growth stage of crops and its confidence according to the fusion feature vector, calculates the health score and grades of crops according to the growth stage information and feature data, identifies the stress state detection result, predicts the growth trend and generates a recommended decision scheme according to the quantitative evaluation result, monitors the decision, feeds back the decision execution effect to the system, and continuously optimizes the model parameters and decision rules, effectively solving the problems mentioned in the background technology.

[0005] The purpose of the application can be realized by the following technical solutions: a crop growth state evaluation method based on an agricultural Internet of Things, characterized by comprising a data acquisition end, a feature extraction end, a growth recognition end, a score calculation end, a prediction decision end, and a feedback optimization end, and specifically comprising the following steps:

[0006] S1, real-time collection of multi-dimensional data of a crop growth environment, output of an original data set;

[0007] S2, alignment of the multi-dimensional data according to geographical coordinates and extraction of related features, construction of a multi-modal fusion feature vector;

[0008] S3, automatic recognition of the current growth stage of the crop and its confidence according to the fusion feature vector;

[0009] S4, calculation of a crop health score and grading according to the growth stage information and feature data, and identification of a stress state detection result;

[0010] S5, prediction of a growth trend and generation of a recommended decision scheme according to the quantitative evaluation result;

[0011] S6, monitoring of the decision, feedback of the decision execution effect to the system, and continuous optimization of the model parameters and decision rules.

[0012] The multi-dimensional data of the crop growth environment includes: environmental data including soil temperature and humidity, pH value, conductivity, nitrogen, phosphorus and potassium content, air temperature and humidity, light intensity, CO2 concentration, wind speed and direction; physiological data including crop canopy temperature, leaf humidity, stem diameter change; image data including hyperspectral image data and visible light image data.

[0013] The way of constructing the multi-modal fusion feature vector is: extracting a standardized feature vector from the original data set; adopting a three-layer progressive fusion strategy, the first layer performing intra-modal weighted combination, the second layer performing inter-modal interaction to cross-fuse the features of the modes, and the third layer performing global feature integration to generate a unified representation; dynamically adjusting the weights of each mode according to the growth stage of the crop, collecting the temporal context information, and constructing a comprehensive multi-dimensional feature vector containing environmental weight features, physiological weight features, image weight features, interactive fusion features, temporal context features and stage adaptation features.

[0014] The way of dynamically adjusting the weights of each mode is: establishing a dynamic weight adjustment mechanism based on the growth stage, increasing the weight of the environmental features in the seedling stage; increasing the weight of the physiological features in the jointing stage; increasing the weight of the image features in the heading stage, and the environmental features and physiological features are coordinated; increasing the weight of the image features in the mature stage, and the environmental features serve as an auxiliary.

[0015] The method for automatically identifying the current growth stage of a crop is as follows: based on historical data, the crop growth process is divided into several key stages; the current fused feature vector is compared with the feature thresholds of each stage in the rule base, and if the threshold conditions are met, it is determined to be the corresponding stage; the network parameters and judgment thresholds are dynamically adjusted in combination with crop variety-specific parameters; and a confidence percentage is generated based on the feature matching degree.

[0016] The method for calculating crop health scores is as follows: calculate the individual score for each influencing factor; if the actual value is within a preset benchmark range, then use the expression:

[0017]

[0018] Where X is the actual score, a is the actual value, b is the preset lower limit, and c is the preset upper limit;

[0019] If the actual value is lower than the lower limit, the score for a single item is 0; if the actual value is higher than the upper limit, the score for a single item is 100.

[0020] Based on the different growth stages, dynamic weights are assigned to each influencing factor for each growth stage.

[0021] The overall health score can be calculated using the following expression:

[0022] K = ∑(X) i ×W i )

[0023] Where K is the health score, Xi is the individual score of each impact factor, and Wi is the dynamic weight of each impact factor.

[0024] The preferred method for classifying health status is as follows: a health status score of ed is considered excellent, and current management measures should be maintained, with a focus on monitoring critical indicators for high yield; a health status score of fe is considered good, and targeted adjustments should be made to influencing factors with low individual scores; a health status score of gf is considered moderate, and intervention should be implemented to address stress conditions; a health status score of hg is considered poor, and emergency measures should be taken to address influencing factors, and data sources should be reviewed; a health status score below h is considered poor, and disaster management should be implemented, with optimization of the planting plan for the next quarter.

[0025] The stress state detection results include: abiotic stresses include drought, floods, salinity, high or low temperature, and nutrient deficiency; biotic stresses include pest and disease infection.

[0026] The method for predicting growth trends is as follows: analyze historical health score sequences and environmental data time series to construct a multi-dimensional time series matrix; use a model to output future health change curves and key node prediction results; quantify risk probabilities and mark risks for daily average health decline exceeding the standard or abnormal weather conditions.

[0027] The method for continuously optimizing model parameters and decision rules is as follows: a closed-loop feedback mechanism is constructed, in which monitoring data is input into the effect evaluation model. The accuracy of equipment execution is verified from a technical perspective, the physiological response of crops is evaluated from an agronomic perspective, and the input-output ratio is calculated from an economic perspective. The evaluation results are pushed to the model optimization module and the rule adjustment module through a preset feedback path. In terms of model parameter optimization, newly generated effective data is used as training samples to dynamically update the weight parameters of the prediction model. In terms of decision rule optimization, a rule utility evaluation matrix is ​​established based on the feedback data, and the existing rule base is iteratively updated.

[0028] Combining all the above technical solutions, the positive effects of this invention are as follows: 1. By collecting multi-dimensional environmental data in real time, it overcomes the limitations of single and delayed data collection in traditional agriculture, providing more comprehensive and timely raw data support for crop growth status assessment, ensuring the objectivity and timeliness of the assessment basis. Aligning environmental data and crop ontological data according to geographical coordinates and fusing features solves the problem of data dispersion from different sources and modalities, strengthens the spatial correlation and internal logic between data, and the constructed multi-modal fusion feature vector can more realistically reflect the comprehensive state of crop growth, laying a high-quality feature foundation for subsequent assessment.

[0029] 2. This invention enables automatic identification and confidence quantification of crop growth stages, replacing the problems of strong subjectivity, low efficiency, and large errors in traditional manual identification. It improves the accuracy and automation level of growth stage judgment, providing precise stage basis for targeted management.

[0030] 3. This invention upgrades crop health status from a qualitative description to a quantitative assessment by calculating and grading health scores and combining them with stress state detection. This not only intuitively reflects the overall health level of the crop but also accurately identifies stress problems, facilitating rapid risk identification and targeted measures to reduce losses. Based on the quantitative assessment results, it predicts growth trends and generates suggested decision-making plans, realizing a shift from passive response to proactive management. This provides farmers with scientific and personalized management guidance, helping to optimize resource input and improve crop yield and quality.

[0031] 4. This invention forms a closed-loop mechanism of evaluation, decision-making, feedback and optimization through decision monitoring and effect feedback, so that model parameters and decision rules can dynamically adapt to changes in different crops, regions and environments, continuously improve the applicability and evaluation accuracy of the system, and promote the upgrading of agricultural production towards precision and intelligence. Attached Figure Description

[0032] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0033] Figure 1 This is a flowchart illustrating the implementation steps of the method of the present invention. Detailed Implementation

[0034] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0035] See Figure 1 As shown, this invention proposes a crop growth status assessment method based on agricultural Internet of Things, including a data acquisition end, a feature extraction end, a growth identification end, a scoring calculation end, a prediction and decision-making end, and a feedback optimization end.

[0036] In a more specific application of this invention, multi-dimensional data on crop growth environment are collected, and raw datasets are output. This can be achieved by: Collecting environmental data by deploying monitoring equipment in both soil and air scenarios. Among soil-related indicators, soil temperature, humidity, and electrical conductivity can be collected using integrated soil sensors with a probe design, buried in the soil layer where crop roots are densely distributed, avoiding direct contact with fertilizers or stones to reduce interference; soil pH requires a dedicated soil pH sensor, ensuring full contact between the probe and the soil, and regular calibration to maintain measurement accuracy; soil nitrogen, phosphorus, and potassium content can be collected using portable soil nutrient rapid analyzers or online monitoring terminals. Online equipment needs to be pre-buried in representative plots, and data is output in real time using ion-selective electrodes or spectral analysis technology. Regarding air environment data, integrated environmental sensors can be used for air temperature, humidity, and CO2 concentration, installed above the crop canopy to avoid direct sunlight and dead air zones. Light intensity should be measured using photosynthetically active radiation sensors, horizontally installed above unobstructed crops to capture actual light energy. Wind speed and direction can be collected using anemometers and wind vanes from small weather stations, installed in open areas around the plot, at a height no less than the highest point of the crops to minimize the impact of crop canopy obstructing airflow. All environmental sensors are connected via a data acquisition module and transmit raw data to the terminal system in real time.

[0037] Physiological data collection directly targets the crop itself using non-invasive or minimally invasive equipment. Crop canopy temperature can be collected non-contactly using infrared thermometers or thermal imagers. The equipment is mounted on an adjustable-height stand, with the lens aimed at the center of the canopy, avoiding images of soil or bare areas. The collection frequency is consistent with environmental data. Leaf humidity sensors are clamped to the back of functional leaves, and the sensor surface must be kept clean to prevent data distortion caused by dew or raindrops. Stem diameter changes are monitored using micro-change sensors, which are fixed to the middle of the crop stem using elastic clamps to ensure close contact without affecting growth. High-precision displacement sensors capture minute changes in stem expansion and contraction throughout the day and night, capturing short-term physiological fluctuations. Data from these physiological sensors is uploaded in real-time via a wireless transmission module, creating a temporal match with environmental data.

[0038] Image data acquisition requires a multi-device collaborative approach, considering both spectral information and morphological characteristics. Hyperspectral image data is acquired using a hyperspectral camera mounted on a drone or fixed track. Visible light image data is acquired using a high-definition industrial camera, with multiple fixed shooting points set up around the perimeter and center of the plot. The lens is aimed at the crop population, individual plants, and leaf details, with the shooting frequency adjusted according to the growth stage. Simultaneously, the lighting conditions and camera parameters at the time of shooting are recorded. Both types of image data are stored in their raw format and transmitted to a data server via wired or wireless means, preserving complete spectral curves and pixel information to provide raw material for subsequent analysis.

[0039] Aligning multi-dimensional data with geographic coordinates and extracting relevant features to construct a multimodal fusion feature vector can be achieved by: defining the time boundaries of the seedling, jointing, heading, and maturity stages through prior data annotation and crop growth models, and setting initial weight benchmarks for each stage. During the seedling stage, crop growth depends on soil nutrients and environmental conditions such as temperature and humidity; therefore, environmental features have the highest weight, followed by physiological features, and image features have the lowest weight. During the jointing stage, the crop stem elongates rapidly, and changes in stem diameter and canopy temperature become core indicators of growth potential; therefore, the weight of physiological features increases, the weight of environmental features decreases, and the weight of image features remains constant. During the heading stage, panicle development is closely related to photosynthetic efficiency, and image features dominate; simultaneously, environmental and physiological features synergistically influence the seed setting rate. During the maturity stage, crop quality and yield are determined; the weight of image features further increases, the weight of environmental features as an auxiliary basis for quality judgment decreases, and the weight of physiological features is the lowest.

[0040] A dynamic correction mechanism based on real-time data feedback is established to avoid weight solidification. Within each stage, the correlation between each modal feature and crop growth indicators is calculated in real-time using a feature importance assessment algorithm. For stage transition periods, a weight smoothing transition strategy is implemented to avoid abrupt changes affecting feature consistency. The dynamically adjusted weights are embedded into a three-layer progressive fusion framework: in the first layer, during intra-modal weighted combination, the standardized feature vector is weighted using the current weights of each modality; in the second layer, during inter-modal interaction, the weights determine the strength of cross-fusion; in the third layer, during global integration, the weights serve as adjustment coefficients for global features, ultimately generating a multi-dimensional feature vector containing environmental weight features, physiological weight features, image weight features, interactive fusion features, temporal context features, and stage adaptation features. This approach enables the multi-modal fusion feature vector to accurately match the growth needs of crops at each stage and dynamically respond to real-time data changes, providing a reliable basis for subsequent growth status assessment and management decisions.

[0041] Based on the fused feature vectors, the system automatically identifies the current growth stage of the crop and its confidence level. This is achieved by: calling a preset rule base and comparing key indicators in the current feature vector with the feature thresholds for each stage; when an indicator meets the threshold range for a certain stage, it is preliminarily determined to be in that growth stage. The system also dynamically adjusts the network parameters and judgment thresholds of the identification model by incorporating crop variety-specific parameters to avoid misjudgments due to variety differences. A confidence percentage is generated by calculating the matching degree between the feature vector and the standard templates for each stage; a higher matching degree indicates higher confidence. If the confidence level is below n%, the system will trigger a secondary data collection and verification to ensure the reliability of the growth stage identification results.

[0042] Based on growth stage information and characteristic data, crop health scores are calculated and graded, while stress state detection results are identified. The crop health score is calculated as follows: Each influencing factor's individual score is calculated. If the actual value is within a preset benchmark range, it is then expressed using the following expression:

[0043]

[0044] Where X is the actual score, a is the actual value, b is the preset lower limit, and c is the preset upper limit;

[0045] If the actual value is lower than the lower limit, the score for a single item is 0; if the actual value is higher than the upper limit, the score for a single item is 100.

[0046] Based on the different growth stages, dynamic weights are assigned to each influencing factor for each growth stage.

[0047] The overall health score can be calculated using the following expression:

[0048] K = ∑(X) i ×W i )

[0049] Where K is the health score, Xi is the individual score of each impact factor, and Wi is the dynamic weight of each impact factor.

[0050] The health rating is graded as follows: a health rating of ed is considered excellent, and current management measures should be maintained, with a focus on monitoring critical indicators for high yield; a health rating of fe is considered good, and targeted adjustments should be made to factors with low individual scores; a health rating of gf is considered moderate, and intervention should be implemented to address stress conditions; a health rating of hg is considered poor, and emergency measures should be taken to address influencing factors, and data sources should be reviewed; a health rating below h is considered poor, and disaster control measures should be implemented, with optimization of the planting plan for the next quarter.

[0051] The method for identifying stress state detection results is as follows: For abiotic stress, specific characteristic indicators need to be associated with different stress types. Drought stress is judged by a combination of factors, including soil moisture consistently below the threshold of the corresponding growth stage, a sudden drop in leaf moisture, a reduction in the daily variation of stem diameter, and an increase in reflectance of water-sensitive bands in hyperspectral images; flood stress is identified by soil moisture being less than field capacity, abnormal fluctuations in electrical conductivity in the root region, and abnormally low canopy temperature; saline-alkali stress is determined by soil pH, electrical conductivity, and abnormally increased reflectance in leaf spectra; high and low temperature stress is determined by the difference between air temperature and canopy temperature; nutrient deficiency is identified by soil nitrogen, phosphorus, and potassium content being below the stage threshold, combined with the proportion of yellowed areas in visible light images and hyperspectral red-edge shift characteristics.

[0052] Biological stress identification emphasizes the synergistic verification of image features and physiological anomalies. Disease infection is determined by the shape, color, and distribution pattern of lesions in visible light images, combined with changes in reflectance in specific bands of the hyperspectral image. Insect pests are identified based on the shape of insect holes and traces of insect frass in images, as well as abnormally high leaf humidity and stagnation of stem diameter growth in physiological data. Simultaneously, the judgment weights are adjusted according to the growth stage; for example, seedlings are susceptible to root rot, requiring a focus on correlating root area images with soil moisture data; during the heading stage, the detection of disease and pest spots in ear images is strengthened.

[0053] By calculating and classifying health scores, and combining them with stress status detection, crop health status is upgraded from a qualitative description to a quantitative assessment. This not only reflects the overall health level of crops intuitively, but also accurately identifies stress problems, facilitating rapid risk identification and targeted measures to reduce losses.

[0054] Based on the quantitative assessment results, growth trends can be predicted and suggested decision-making schemes can be generated by: constructing a multi-dimensional time series matrix to integrate historical information. This involves collecting historical data from the first three growth cycles of the crop, aligning the historical health score sequence with the environmental data time series by timestamp, and forming a matrix with the dimension of time × feature.

[0055] A fusion model is employed to predict growth trends. A hybrid model combining LSTM and attention mechanisms can be used. The LSTM layer captures long-term dependencies, processes the multi-dimensional time series matrix, and outputs a preliminary health change curve. The attention layer assigns higher weights to key time points, strengthening the influence of key stage features. The model is trained using historical health scores as labels, with 80% training data and 20% validation data. Optimization is achieved by adjusting the learning rate and iteration count to reduce prediction error. After training, the model is input as a time series matrix of the most recent 30 days, outputting future health change curves and extracting key nodes, including the time of health peaks or troughs, growth stage transition times, and potential stress-sensitive points.

[0056] Based on predicted curves and historical data statistics, the probability of abnormal decline in crop health and the probability of abnormal weather are calculated, and these probabilities are correlated with corresponding stress types. Combining predicted trends and risk markers, recommended decision-making schemes are generated. For predicted health change curves, if an upward trend is observed and no risk marker is present, routine management recommendations are generated; if a decline in health is predicted and associated with risk markers, targeted intervention schemes are output. These schemes must specify the operation time, implementation indicators, and expected effects, and adjust the recommendation intensity according to the growth stage. This technology automates the entire process from historical data to trend prediction, risk warning, and decision-making recommendations, providing dynamic and actionable support for precision crop management.

[0057] Monitoring decision-making and feeding back the results of its implementation to the system allows for continuous optimization of model parameters and decision rules. This can be achieved through: constructing a closed-loop feedback mechanism, inputting monitoring data into the effect evaluation model, verifying equipment execution accuracy from a technical perspective, assessing crop physiological responses from an agronomic perspective, and calculating the input-output ratio from an economic perspective. The evaluation results are then pushed to the model optimization module and rule adjustment module via pre-defined feedback paths. For model parameter optimization, newly generated effective data is used as training samples to dynamically update the weight parameters of the prediction model. For decision rule optimization, a rule utility evaluation matrix is ​​established based on feedback data, and the existing rule base is iteratively updated. In this way, model parameters and decision rules can dynamically adapt to changes in different crops, regions, and environments, continuously improving the system's applicability and evaluation accuracy, and promoting the upgrading of agricultural production towards precision and intelligence.

[0058] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other.

[0059] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assessing crop growth status based on agricultural Internet of Things (IoT), characterized in that, It includes a data acquisition end, a feature extraction end, a growth recognition end, a scoring calculation end, a prediction and decision-making end, and a feedback and optimization end, specifically including the following steps: S1. Real-time collection of multi-dimensional data on crop growth environment, outputting raw dataset; S2. Align the multi-dimensional data according to geographic coordinates and extract features to construct a multimodal fusion feature vector; S3. Based on the fused feature vector, automatically identify the current growth stage of the crop and its confidence level; S4. Calculate and classify crop health scores based on growth stage information and characteristic data, and identify stress state detection results at the same time; S5. Based on the quantitative assessment results, predict the growth trend and generate suggested decision-making solutions; S6. Monitor the decision-making process, feed back the results of the decision-making process to the system, and continuously optimize the model parameters and decision-making rules.

2. The crop growth status assessment method based on agricultural Internet of Things as described in claim 1, characterized in that: The multi-dimensional data of the crop growth environment includes: environmental data such as soil temperature and humidity, pH value, electrical conductivity, nitrogen, phosphorus and potassium content, air temperature and humidity, light intensity, CO2 concentration, wind speed and direction; physiological data such as crop canopy temperature, leaf humidity, and stem diameter changes; and image data such as hyperspectral image data and visible light image data.

3. The crop growth status assessment method based on agricultural Internet of Things as described in claim 1, characterized in that: The method for constructing a multimodal fusion feature vector is as follows: a standardized feature vector is extracted from the original dataset; a three-layer progressive fusion strategy is adopted, in which the first layer performs intramodal weighted combination, the second layer performs intermodal interaction to cross-fuse the features of the modes, and the third layer performs global feature integration to generate a unified representation; the weights of each mode are dynamically adjusted according to the crop growth stage, temporal context information is collected, and a comprehensive multidimensional feature vector containing environmental weight features, physiological weight features, image weight features, interaction fusion features, temporal context features, and stage adaptation features is constructed.

4. The crop growth status assessment method based on agricultural Internet of Things as described in claim 3, characterized in that: The method for dynamically adjusting the weights of each modality is as follows: a dynamic weight adjustment mechanism based on growth stages is established. During the seedling stage, the weight of environmental features is increased; during the jointing stage, the weight of physiological features is increased; and during the heading stage, the weight of image features is increased, with environmental features and physiological features working in synergy. During the mature stage, the weight of image features is increased, while environmental features are used as auxiliary features.

5. The crop growth status assessment method based on agricultural Internet of Things as described in claim 1, characterized in that: The method for automatically identifying the current growth stage of a crop is as follows: based on historical data, the crop growth process is divided into several key stages; the current fused feature vector is compared with the feature thresholds of each stage in the rule base, and if the threshold conditions are met, it is determined to be the corresponding stage; the network parameters and judgment thresholds are dynamically adjusted in combination with crop variety-specific parameters; and a confidence percentage is generated based on the feature matching degree.

6. The crop growth status assessment method based on agricultural Internet of Things as described in claim 1, characterized in that: The method for calculating crop health scores is as follows: calculate the individual score for each influencing factor; if the actual value is within a preset benchmark range, then use the expression: Where X is the actual score, a is the actual value, b is the preset lower limit, and c is the preset upper limit; If the actual value is lower than the lower limit, the score for a single item is 0; if the actual value is higher than the upper limit, the score for a single item is 100. Based on the different growth stages, dynamic weights are assigned to each influencing factor for each growth stage. The overall health score can be calculated using the following expression: K=∑(X i ×W i ) Where K is the health score, Xi is the individual score of each impact factor, and Wi is the dynamic weight of each impact factor.

7. The crop growth status assessment method based on agricultural Internet of Things as described in claim 1, characterized in that: The health rating is graded as follows: a health rating of ed is considered excellent, and current management measures should be maintained, with a focus on monitoring critical indicators for high yield; a health rating of fe is considered good, and targeted adjustments should be made to factors with low individual scores; a health rating of gf is considered moderate, and intervention should be implemented to address stress conditions; a health rating of hg is considered poor, and emergency measures should be taken to address influencing factors, and data sources should be reviewed; a health rating below h is considered poor, and disaster control measures should be implemented, with optimization of the planting plan for the next quarter.

8. The crop growth status assessment method based on agricultural Internet of Things as described in claim 1, characterized in that: The stress state detection results include: abiotic stresses include drought, floods, salinity, high or low temperature, and nutrient deficiency; biotic stresses include pest and disease infection.

9. The crop growth status assessment method based on agricultural Internet of Things as described in claim 1, characterized in that: The method for predicting growth trends is as follows: analyze historical health score sequences and environmental data time series to construct a multidimensional time series matrix; use a model to output future health change curves and key node prediction results; The probability of risk is quantified, and risk markers are set for situations where the daily average decline in health exceeds the standard or for abnormal weather conditions.

10. The crop growth status assessment method based on agricultural Internet of Things as described in claim 1, characterized in that: The method for continuously optimizing model parameters and decision rules is as follows: a closed-loop feedback mechanism is constructed, monitoring data is input into the effect evaluation model, the accuracy of equipment execution is verified from a technical dimension, the physiological response of crops is evaluated from an agronomic dimension, and the input-output ratio is calculated from an economic dimension. The evaluation results are pushed to the model optimization module and the rule adjustment module respectively through a preset feedback path. In terms of model parameter optimization, newly generated effective data is used as training samples to dynamically update the weight parameters of the prediction model; in terms of decision rule optimization, a rule utility evaluation matrix is ​​established based on feedback data, and the existing rule base is iteratively updated.

Images