Intelligent agriculture monitoring system based on big data

Abstract

The invention relates to the technical field of agricultural monitoring, and discloses an intelligent agricultural monitoring system based on big data, which comprises a division unit, an integration unit, an analysis unit, an association unit and an evaluation unit.According to the scheme, a single plant serves as a core, grids are dynamically divided according to overground and root system characteristics, different from traditional large-area monitoring, crops can be accurately focused, and the monitoring accuracy is improved. Various data such as images, environment, soil and the like are fused, big data and intelligent agriculture are combined, limitation of single data is broken through, overground and underground conditions are associated through overground physiological advance values and root depth activity coefficients, the situation that only overground parts are monitored, monitoring lag conditions occur is broken through, and growth conditions of crops are known in advance. Therefore, crop health is accurately evaluated, and an accurate detection effect is achieved.

Description

Technical Field

[0001] The present invention relates to the field of agricultural monitoring technology, and in particular to a smart agricultural monitoring system based on big data. Background Art

[0002] In the field of smart agriculture, crop monitoring mainly relies on satellite remote sensing images, drone images, and environmental and soil data. Deep learning algorithms are usually used to conduct comprehensive analysis of these multi-dimensional data to accurately capture surface conditions, such as changes in leaf color, so as to timely and comprehensively understand the growth status of crops.

[0003] However, in greenhouse cultivation environments, most of the data currently collected only focuses on the above-ground parts of crops. Although data analysis can help us understand the growth status of crops to a certain extent, the physiological changes of crops often occur before their surface morphological changes. For example, after the crop roots are deprived of oxygen for 3-5 days, symptoms such as leaf wilting will not appear. This will result in poor monitoring results and obvious lags if only the surface morphology of crops is monitored, making it impossible to understand the true growth status of crops in a timely manner. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides a smart agricultural monitoring system based on big data to solve the above problems.

[0005] The above technical objectives of the present invention are achieved through the following technical solutions:

[0006] The smart agricultural monitoring system based on big data includes:

[0007] The division unit is used to obtain the maximum diameter of the ground and the maximum extension radius of the root system of each crop in the greenhouse, determine the dynamic grid size of the single crop, and divide the planting area into multiple dynamic grids based on the dynamic grid size, each grid covering the single crop and the extension range of its root system;

[0008] The integration unit is used to collect image data, environmental data, and soil data at different depths of individual crops in each grid in real time, and integrate the three pre-processed data to obtain a big data fusion matrix for individual crops;

[0009] The analysis unit is used to analyze the big data fusion matrix to obtain the aboveground physiological advance value and the root depth activity coefficient respectively;

[0010] The correlation unit is used to perform correlation analysis on the aboveground physiological advance value and the root depth activity coefficient to obtain the crop health correlation index;

[0011] The evaluation unit is used to determine the health status of crops based on crop health correlation indexes.

[0012] Furthermore, the dynamic grid size of individual crops is determined, including:

[0013] The maximum above-ground diameter and the maximum root extension radius of each crop were calculated to obtain the proportional coefficient;

[0014] Determine the redundant root coverage radius based on the proportional coefficient and the maximum above-ground diameter of each crop plant;

[0015] Taking a single crop plant as the center, the planting area is divided into square grids with a side length of twice the root redundancy coverage radius to generate a dynamic grid size for a single crop plant.

[0016] Furthermore, the three pre-processed data are integrated to obtain a big data fusion matrix of a single crop, including:

[0017] Extract image data to obtain image feature vector;

[0018] Analyze the correlation between each data in the environmental data, and integrate the correlation to obtain a comprehensive environmental impact index;

[0019] Analyze soil data at different depths to obtain soil characteristic tensors;

[0020] The image feature vector, comprehensive environmental impact index and soil feature tensor are fused to generate a big data fusion matrix of a single crop.

[0021] Furthermore, the big data fusion matrix is ​​analyzed to obtain the ground physiological lead value, including:

[0022] Perform image extraction on the big data fusion matrix to obtain the time series image change rate;

[0023] Determine the environmental physiological response coefficient based on the time series image change rate and the comprehensive environmental impact index;

[0024] Screen the soil characteristic tensor and generate shallow soil impact weights;

[0025] Obtain historical data of crops, analyze the historical data, and obtain the historical mean of image feature vectors at the same growth stage;

[0026] The historical mean is corrected based on the comprehensive environmental impact index of the same period to obtain the dynamic physiological benchmark vector.

[0027] Furthermore, the big data fusion matrix is ​​analyzed to obtain the ground physiological lead value, which also includes:

[0028] The time series image change rate, environmental physiological response coefficient, and shallow soil impact weight are integrated to obtain the current physiological characteristic prediction value;

[0029] Compare the current physiological characteristic prediction value with the dynamic physiological reference vector to obtain the advance prediction deviation;

[0030] The lead prediction deviation is calculated to generate the above-ground physiological lead value.

[0031] Furthermore, the big data fusion matrix was analyzed to obtain the root depth activity coefficient, including:

[0032] The soil characteristic tensor is parsed to extract the nutrient availability and water conductivity of the soil at each depth, and the depth soil activity weight is obtained;

[0033] Based on the dynamic changes of temperature and light duration in the comprehensive environmental impact index, the depth soil activity weight is corrected to generate a depth correction coefficient;

[0034] Extract the image feature vector to obtain the aboveground growth rate characteristics, and fuse the aboveground growth rate characteristics with the depth correction coefficient to obtain the preliminary root activity coefficient;

[0035] Determine the root activity baseline value for the same period in history based on historical data, and calculate the deviation rate between the preliminary root activity coefficient and the root activity baseline value;

[0036] The preliminary root activity coefficient was calibrated based on the deviation rate to obtain the root depth activity coefficient.

[0037] Furthermore, the root activity baseline values ​​for the same period in history are determined based on historical data, including:

[0038] Get the current date and the crop planting date, and determine the growth stage matching factor based on the two dates;

[0039] Screening growth stage matching factors from historical data All healthy crop samples within 5% are used to construct a temporally and spatially similar sample set;

[0040] Cluster analysis is performed on the spatiotemporally similar sample sets to obtain the basic activity distribution interval;

[0041] Extract environmental data and construct environmental correction tensor;

[0042] Adjust the basic activity distribution interval according to the environmental correction tensor to generate the environmental adaptability benchmark interval;

[0043] The environmental adaptability benchmark interval was disassembled to separate the trend part and the cycle part, and the root activity benchmark value for the same historical period was calculated.

[0044] Furthermore, correlation analysis was conducted on the aboveground physiological advance value and the root depth activity coefficient to obtain the crop health correlation index, including:

[0045] Decompose the aboveground physiological advance value and root depth activity coefficient to generate a multi-dimensional growth coordination tensor;

[0046] The matching degree of above-ground and underground characteristic components in the growth coordination tensor is calculated to obtain the dynamic correlation between the growth cycle and spatial distribution.

[0047] Determine the growth correlation of healthy crops in the same period based on historical data and establish a health fluctuation feature library;

[0048] Calculate the difference between the current dynamic correlation and the health fluctuation feature library to generate the health feature deviation;

[0049] The dynamic correlation degree and health characteristic deviation degree were analyzed to obtain the crop health correlation index.

[0050] Furthermore, based on historical data, the growth correlation of healthy crops in the same period in history is determined, and a health fluctuation feature library is established, including:

[0051] Based on historical data, determine the healthy crop data at different growth stages during the same period in history;

[0052] Calculate the matching degree between healthy crop data and the growth stage of current crops and the similarity of environmental factors to generate a healthy sample screening index;

[0053] Based on the healthy sample screening index, target historical healthy crop samples are screened to obtain the growth correlation feature matrix;

[0054] The growth-related characteristic matrix was analyzed and a health fluctuation characteristic library was established.

[0055] Furthermore, the crop health status is determined based on the crop health correlation index, including:

[0056] Determine the health threshold interval based on aboveground physiological advance values, root depth activity coefficients, and historical data;

[0057] The health threshold interval is compared with the crop health association index to obtain the crop health status.

[0058] In summary, the present invention mainly has the following beneficial effects:

[0059] Dynamic grid division enables precise monitoring of individual crops. Traditional greenhouse monitoring mostly uses fixed-area sampling, which makes it difficult to take into account the growth differences of different crops. This solution dynamically calculates the grid size based on the maximum diameter above ground and the root extension radius, and combines the proportional coefficient to determine the redundant coverage radius of the root system, so that each grid accurately covers the individual crop and the root influence range. This dynamic adjustment mechanism not only avoids interference from adjacent crops, but also ensures the integrity of the environmental data around the root system.

[0060] By integrating image data, environmental data and soil data to generate a fusion matrix, and then combining big data with smart agriculture, the monitoring effect is improved. At the same time, characteristics such as the time series image change rate and the environmental physiological response coefficient are extracted, and combined with historical data correction to obtain a dynamic physiological benchmark, the above-ground physiological lead value is calculated, and the soil characteristic tensor is analyzed. Combined with the environmental dynamic correction, the root depth activity coefficient is generated, and the above-ground morphological changes are correlated with root activity. This can capture physiological abnormalities before the surface symptoms of crops appear. For example, physiological fluctuations caused by root hypoxia can be identified 3-5 days in advance, breaking through the time limitations of traditional surface morphological monitoring, providing a critical window period for timely intervention, and avoiding monitoring lags.

[0061] The matching degree of above-ground and underground characteristics is calculated through the growth coordination tensor, and the health correlation index is generated by combining with the health fluctuation feature library. It not only takes into account the physiological linkage between the above-ground part and the root system of the crop, but also incorporates the influence of environmental factors and historical data, so that the health assessment is upgraded from a single indicator judgment to a multi-dimensional dynamic correlation analysis. Compared with the traditional assessment method based only on surface features such as leaf color, the health status determined by the health correlation index more comprehensively reflects the true physiological state of the crop, thereby improving the monitoring effect. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a schematic diagram of the big data-based smart agriculture monitoring system of the present invention. DETAILED DESCRIPTION

[0063] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0064] refer to Figure 1 , a smart agricultural monitoring system based on big data, including:

[0065] The division unit is used to obtain the maximum diameter of the ground and the maximum extension radius of the root system of each crop in the greenhouse, determine the dynamic grid size of the single crop, and divide the planting area into multiple dynamic grids based on the dynamic grid size, each grid covering the single crop and the extension range of its root system;

[0066] The integration unit is used to collect image data, environmental data, and soil data at different depths of individual crops in each grid in real time, and integrate the three pre-processed data to obtain a big data fusion matrix for individual crops;

[0067] The analysis unit is used to analyze the big data fusion matrix to obtain the aboveground physiological advance value and the root depth activity coefficient respectively;

[0068] The correlation unit is used to perform correlation analysis on the aboveground physiological advance value and the root depth activity coefficient to obtain the crop health correlation index;

[0069] The evaluation unit is used to determine the health status of crops based on crop health correlation indexes.

[0070] Through dynamic grid division, it accurately covers the range of individual crops and root systems, integrates ground images, environmental data and soil data at different depths, and realizes the fusion of above-ground and underground data. By analyzing the above-ground physiological lead value and the root depth activity coefficient, the health index is obtained through correlation. It can capture physiological changes in advance, overcome the lag of only monitoring surface morphology, and reflect the real growth status of crops more timely and accurately. It is suitable for the monitoring and planting of crops in greenhouses.

[0071] In one aspect of this embodiment, determining the dynamic grid size of a single crop plant includes:

[0072] The maximum above-ground diameter and the maximum extension radius of the root system of each crop are calculated to obtain a proportionality coefficient, specifically comprising: dividing the maximum above-ground diameter of a single crop by the maximum extension radius of the root system to obtain the proportionality coefficient;

[0073] Based on the proportional coefficient and the maximum above-ground diameter of each crop, the root redundancy coverage radius is determined. Specifically, the proportional coefficient weight is set to 0.3, the maximum above-ground diameter weight is set to 0.7, the proportional coefficient is multiplied by 0.3, the maximum above-ground diameter is multiplied by 0.7, and the two products are added together. The result is the root redundancy coverage radius;

[0074] Taking a single crop plant as the center, the planting area is divided into square grids with a side length of twice the root redundant coverage radius to generate the dynamic grid size of the single crop plant. Specifically, a plane rectangular coordinate system is established with the single crop plant as the origin, the coordinates of the plant center point are obtained, the root redundant coverage radius is multiplied by 2 to obtain the side length of the square, and half of the side length is extended in the four directions of east, west, south and north with the single crop plant as the center point as the reference to determine the coordinate values ​​of the four vertices of the square. The coordinate points are connected to form a closed square area. This area is the dynamic grid size of the single crop plant, ensuring that the distance from the grid boundary to the plant center is equal everywhere and covers the root expansion range.

[0075] Through dynamic grid division, accurate coverage of the root system of crops is achieved. In greenhouse cultivation, the limitation of traditional monitoring of only the above-ground part is solved. With the help of the calculation of the redundant coverage radius of the root system, the grid can take into account the needs of root expansion, which allows monitoring to be extended to the underground root area. Combined with soil data, it can capture physiological changes such as root hypoxia in advance, avoid lags, and help growers grasp the real growth status of crops in a timely manner.

[0076] By centering the dynamic grid on the plant and keeping the distance from the boundary to the center equal, the balance and integrity of the monitoring range are ensured. The clear coordinate definition of the square grid facilitates the integration of relevant data from various big data. In smart agricultural monitoring, this structured grid can analyze data more efficiently and improve the accuracy of judging the growth status of crops.

[0077] In one case of this embodiment, the three pre-processed data are integrated to obtain a big data fusion matrix of a single crop, including:

[0078] Extracting image data to obtain image feature vectors specifically includes: gray-scaling the pre-processed image data to convert the color image into a grayscale image, using the Sobel operator to calculate the horizontal and vertical gradient values, identifying the leaf edge contour by the gradient value, using the Harris corner detection algorithm to calculate the grayscale change value of the pixel points in the image, determining the position of the leaf vertices and recording the coordinates, and counting the number of leaves. The leaf area is calculated using the pixel counting method (every 100 pixels corresponds to 1 square centimeter), and the plant height can be converted by the difference between the upper and lower edge pixels (every 200 pixels corresponds to 1 meter). The leaf number, area, plant height, and edge contour features are arranged in a fixed order to form an image feature vector.

[0079] Analyze the correlation between each data in the environmental data and integrate the correlation to obtain a comprehensive environmental impact index. Specifically, the following steps are performed: standardize the temperature, humidity, light intensity, carbon dioxide concentration and other parameters of the environmental data; use the Pearson correlation coefficient algorithm to calculate the linear correlation coefficient between each parameter to form a correlation matrix; use the hierarchical analysis method to compare the importance of each parameter; construct a judgment matrix and calculate the weight vector using the square root method; and perform a weighted summation of the correlation matrix and the weight vector to obtain a comprehensive environmental impact index, where the correlation matrix elements range from -1 to 1 and the weight vector elements sum to 1;

[0080] The soil data at different depths are analyzed to obtain soil characteristic tensors, including: 20 cm, 20 40 cm, 40 The soil depth was divided into three layers (60 cm). pH, nitrogen, phosphorus, potassium, and organic matter content were collected for each layer. Principal component analysis was used to calculate the variance contribution of each indicator. The top three principal components with a cumulative contribution rate exceeding 85% were selected. The three principal components of each layer were arranged in order from shallow to deep depth. Each principal component corresponded to a specific value, forming a three-dimensional array containing depth stratification, principal component sequence number, and eigenvalue, which is the soil characteristic tensor.

[0081] The image feature vector, comprehensive environmental impact index and soil feature tensor are fused to generate a big data fusion matrix for a single crop. Specifically, the image feature vector is converted into a row vector, the comprehensive environmental impact index is used as a single column vector, and the soil feature tensor is expanded into a two-dimensional matrix. The serial fusion algorithm is used to splice the image feature vector, the comprehensive environmental impact index and the soil feature tensor expansion matrix in the order of the image feature vector, the comprehensive environmental impact index and the soil feature tensor expansion matrix. The number of matrix rows is: the dimension value of the image feature vector plus 1, plus the number of rows after the soil feature tensor is expanded. The sum of the three is the total number of rows of the matrix. The number of matrix columns is: comparing the number of columns of the image feature vector, the number of columns of the comprehensive environmental impact index and the number of columns after the soil feature tensor is expanded, and taking the largest value as the total number of columns of the matrix. The matrix formed after determining the number of rows and columns is the big data fusion matrix.

[0082] By integrating image, environmental and soil data to generate a big data fusion matrix, the limitation of focusing only on the above-ground part in greenhouse cultivation has been broken through. The image feature vector captures above-ground information such as the number of leaves and plant height, the soil feature tensor covers key indicators at different depths, and the comprehensive environmental impact index integrates multiple environmental parameters. The fusion of the three realizes the linkage analysis of above-ground and underground, biological and environmental data of crops, providing comprehensive data support for accurate judgment of growth status, avoiding the one-sidedness of single above-ground data, and thus achieving the purpose of monitoring crops based on big data.

[0083] By combining soil data with image data, root problems can be warned in advance through abnormal soil indicators before leaves wilt, effectively solving the lag of conventional monitoring. The standardized construction of the matrix enables efficient correlation of multi-source data, making it easier to capture the correlation between physiological and morphological changes and timely grasp the true growth status of crops.

[0084] In one case of this embodiment, the big data fusion matrix is ​​analyzed to obtain the ground physiological lead value, including:

[0085] Perform image extraction on the big data fusion matrix to obtain the time series image change rate, specifically including: extracting column data corresponding to the image feature vector from the big data fusion matrix, which contains features such as leaf number, area, and plant height, and sorting them in chronological order according to the acquisition time to form a continuous time series. The sliding window method is used, and the window length is set to 24 hours and the window step is 6 hours. For each window, the average leaf area and the average plant height in the window are calculated, and then the mean difference between the two adjacent windows is calculated, specifically: the average leaf area of ​​the latter window is subtracted from the average leaf area of ​​the previous window to obtain the mean leaf area difference. The mean plant height difference can be obtained by the same calculation method as above, and the leaf area mean difference is divided by 24 hours to obtain the leaf area change per unit time, and the plant height change per unit time is divided by 24 hours. The linear regression algorithm is used to perform trend fitting on the leaf area change per unit time and the plant height change per unit time, respectively. The slope of the fitted straight line is the corresponding time series image change rate;

[0086] The environmental physiological response coefficient is determined based on the time series image change rate and the comprehensive environmental impact index. Specifically, the weight of the leaf area change rate, the plant height change rate, and the comprehensive environmental impact index are set to 0.6, 0.2, and 0.2, respectively. The leaf area change rate and plant height change rate in the time series image change rate and the comprehensive environmental impact index are multiplied by the corresponding weights and then summed. The summed result is the environmental physiological response coefficient.

[0087] The soil characteristic tensor is screened to generate shallow soil impact weights, specifically including: screening out 0 from the soil characteristic tensor The three principal component data corresponding to the 20 cm shallow layer are normalized, and the eigenvalues ​​of these three principal components are converted to the range of 0-1 through linear mapping. The information entropy of each principal component is calculated, and the information entropy of each principal component is subtracted from 1 to obtain the preliminary entropy weight. The three preliminary entropy weights are normalized so that the sum of the three is 1. The three normalized preliminary entropy weights are multiplied by the preset adjustment coefficient 1.2 respectively. The result is the shallow soil impact weight. The number of shallow soil impact weights is three, and the sum of the three shallow soil impact weights is 1;

[0088] Obtain historical data on crops, analyze the historical data, and obtain the historical mean of image feature vectors at the same growth stage. Specifically, the following steps are performed: historical data for the past three complete growth cycles of the crop are obtained from the historical data, and growth stages are divided into seedling stage (1-30 days) and jointing stage (31-60 days) according to the number of days after sowing. For the current stage of the crop, all image feature vectors of the same stage in the historical data are extracted. Features include: leaf number, area, plant height, and edge contour. The arithmetic mean algorithm is used to sum the values ​​of each feature in the same stage, and then the sum is divided by the number of samples (the sample size for each stage is no less than 50 groups). The result is the historical mean of each feature in the image feature vector at the same growth stage.

[0089] The historical mean is corrected based on the comprehensive environmental impact index of the same period to obtain a dynamic physiological baseline vector, specifically including: setting the correction coefficient to 0.8 when the comprehensive environmental impact index is 0, and to 1.2 when it is 2. When the index is between 0 and 2, the corresponding coefficient is obtained by proportional linear conversion, and the historical mean of each feature in the historical mean vector is multiplied by the converted correction coefficient. The four corrected values ​​obtained are arranged in the original order, and the formed vector is the dynamic physiological baseline vector. Among them, the proportional linear conversion includes: dividing the comprehensive environmental impact index from 0 to 2 into 20 intervals, each interval corresponds to an exponential change of 0.1, and for each increase of 0.1 in the exponential value, the correction coefficient increases by 0.02. For example, when the comprehensive environmental impact index is 0.1, the corresponding correction coefficient is 0.82, when it is 0.2, it corresponds to 0.84, and so on, until it reaches 1.2 when the index is 2.

[0090] By combining the dynamic physiological baseline vector with the time-series image change rate, advanced perception of the physiological state of crops is achieved. The dynamic physiological baseline vector is based on historical data and real-time environmental corrections, breaking through the limitations of fixed thresholds and more closely following the true growth patterns of crops. The time-series image change rate captures subtle growth trends through a sliding window. After comparison with the baseline vector, it can identify physiological deviations in advance when there are no abnormalities in leaf morphology. For example, in the early stages of root hypoxia, a slight difference in the plant height change rate from the baseline can be used as an early warning, significantly shortening the response time difference between physiological changes and manifestations.

[0091] The accuracy of physiological status assessment is improved through the quantitative integration of environmental and soil factors. The environmental physiological response coefficient is weighted to integrate image changes and environmental influences, and the shallow soil impact weight focuses on key indicators of the root active zone. The two are analyzed in conjunction with the above-ground characteristics, reflecting the relationship between soil, environment and crops. This allows the physiological lead value to not only reflect the surface growth trend, but also to be related to the soil environmental status of the root system, avoiding misjudgment of single above-ground data.

[0092] In one case of this embodiment, analyzing the big data fusion matrix to obtain the ground physiological lead value also includes:

[0093] The time series image change rate, environmental physiological response coefficient, and shallow soil influence weight are integrated to obtain the current physiological characteristic prediction value, specifically including: setting the time series image change rate weight to 0.5, the environmental physiological response coefficient weight to 0.25, and the shallow soil influence weight to 0.25, multiplying the time series image change rate by 0.5, the environmental physiological response coefficient by 0.25, and the shallow soil influence weight by 0.25, and adding the three products to obtain the current physiological characteristic prediction value;

[0094] Among them, the time-series image change rate directly reflects the real-time growth dynamics of aboveground parts such as leaf area and plant height. It is the most intuitive quantitative indicator of the physiological state of crops and plays a dominant role in the prediction of current characteristics. Therefore, it is given the highest weight of 0.5. The environmental physiological response coefficient is the coupling result of the external environment and growth dynamics. It is an indirect influencing factor and therefore has a weight of 0.25. The shallow soil impact weight corresponds to the 0-20 cm soil layer where root activity is most active. Although it affects nutrient absorption, the effect is indirectly reflected through physiological responses. Therefore, the same environmental coefficient weight is 0.25, forming a weight distribution in which self-growth is dominant, and internal and external influences are coordinated.

[0095] Comparing the current physiological characteristic prediction value with the dynamic physiological reference vector to obtain the advance prediction deviation, specifically including: using the vector difference algorithm to calculate the difference between the current physiological characteristic prediction value and the dynamic physiological reference vector according to the corresponding characteristic dimension, specifically: subtracting the number of leaves in the reference vector from the number of leaves in the prediction value to obtain the difference. According to the same calculation method as above, the area, plant height, and contour feature differences can be obtained. The differences of the four features are arranged in the original sequence, and the resulting vector is the advance prediction deviation;

[0096] The advance prediction deviation is calculated to generate the ground physiological advance value, including: for each sampling point , the time series image change rate Divide by the mean of the historical image feature vectors of the same period , and get the relative rate of change term , take the time partial derivative of the relative rate of change term and get the time rate of change term , the time base Divide by the rate of change benchmark , and get the product term , multiply the product term by the time rate of change term to obtain the normalized reference term ; Comprehensive Environmental Impact Index Divide by the adjustment coefficient of environmental response sensitivity , and obtain the environmental index ratio , the environmental index ratio term is applied to the first kind of modified Bessel function to obtain the environmental impact correction term , multiply the normalized baseline item by the environmental impact correction item to obtain the multiplication result. The multiplication results of the sampling points are summed and averaged to obtain the time series feature item ; Variable along soil depth , from 0 to the shallow soil depth threshold Integrate to obtain the soil characteristic change rate term , the soil characteristic benchmark Multiply by the shallow soil depth threshold , get the product result, and convert the time base Divide by the product, and we get ,Will Multiply by the soil characteristic change rate term to obtain the normalized benchmark , the normalized baseline amount is compared with the soil Ground-related gain coefficient Multiply them together to get the gain coefficient term ; Add the time series characteristic term and the gain coefficient term to obtain the ground physiological lead value , in specific applications, it can be achieved through the following calculation formula, for example: ;

[0097] Where, Indicates the above-ground physiological lead value, Indicates the total number of sampling points of time series data, express The index of Represents the time base quantity, Indicates the rate of change reference, represents the time partial derivative operator, Indicates the rate of change of the temporal image, represents the mean of the image feature vectors in the same period of history, represents the modified Bessel function of the first kind, represents the comprehensive environmental impact index, Indicates the adjustment coefficient of environmental response sensitivity, 1 5. Indicates soil Ground-related gain factor, 2 7, Represents the soil characteristic benchmark, represents the shallow soil depth threshold, Indicates the soil depth, represents the time rate of change of the soil characteristic tensor, represents the depth differential;

[0098] The following are Explanation of the value range:

[0099] When the root redundancy radius When the diameter is 1.5 times that of the aboveground root, it indicates that the potential of the root system to absorb water and fertilizer is not fully utilized, and the driving force of the root activity signal on aboveground physiology is weakened. The value is low, The value is 1 3;

[0100] When the root redundancy radius When the diameter of the root system is 1.5 times that of the crown, it indicates that the root system expansion range is smaller than the crown width (underground space is limited), the competition for water and nutrients between roots is intensified, and the risk of root hypoxia increases. The value is too high. The value is 4 5;

[0101] The following are Explanation of the value range:

[0102] When crops are planted densely, that is, the side length of the dynamic grid of a single crop At 50cm, because the number of plants per unit area exceeds the reasonable range, the competition for light energy, water and nutrients is fierce, and the root system is intertwined, resulting in a decrease in oxygen content in the root zone. The value is low, The value is 2 4.7;

[0103] When crops are planted sparsely, that is, the side length of the dynamic grid of a single crop When the height is 50cm, the number of plants per unit area is lower than the appropriate range, the individual growth space of crops is excessive, the water absorption resistance of the root system is reduced, but the light energy utilization rate of the group is reduced. The value is too high. The value is 4.8 7.

[0104] By integrating the rate of change of time-series images, environmental impacts and shallow soil weights, with self-growth as the main factor and internal and external factors coordinated in weight distribution, the current physiological characteristic prediction value is accurately calculated. After comparing with the dynamic physiological reference vector, it can capture subtle deviations in characteristics such as leaf number and plant height through advanced prediction deviation. Combined with the parameter adjustment of root redundant radius and planting density, it can reflect the root activity and competition status in a targeted manner, and warn of root hypoxia and other problems 3-5 days before leaf wilt, greatly shortening the time difference between physiological changes and manifestations, and solving the monitoring lag.

[0105] By integrating the time series characteristic terms and the soil gain coefficient terms, and fusing multi-dimensional data such as the time change rate, environmental correction and soil integral, a directly applicable quantitative indicator is formed. The parameter values ​​under different root redundancy radii and planting densities allow the indicator to adapt to different growth scenarios, avoiding misjudgment of a single standard. Quantitative analysis is carried out based on physiological mechanisms, allowing monitoring to move from surface morphological observation to physiological status prediction, and timely understanding of the growth status of crops in smart agriculture based on big data analysis.

[0106] In one case of this embodiment, the big data fusion matrix is ​​analyzed to obtain the root depth activity coefficient, including:

[0107] The soil characteristic tensor is parsed to extract the nutrient availability and water conductivity of the soil at each depth to obtain the deep soil activity weight. Specifically, the three principal component eigenvalues ​​of each depth layer are extracted from the soil characteristic tensor, and these eigenvalues ​​are mapped to the nutrient availability and water conductivity of the corresponding depth using a multiple linear regression algorithm. The weight of the nutrient availability is set to 0.6, and the weight of the water conductivity is set to 0.4. The nutrient availability of each depth layer is multiplied by 0.6, and the water conductivity is multiplied by 0.4. The two are added to obtain the activity score of the depth layer. The activity score of each depth layer is normalized so that the result is between 0 and 1. The obtained value is the deep soil activity weight. There are three deep soil activity weights, corresponding to the three layers of deep soil.

[0108] Based on the dynamic change values ​​of temperature and light duration in the comprehensive environmental impact index, the deep soil activity weight is corrected to generate a depth correction coefficient, which includes: using the sliding window method (window 24 hours, step length 6 hours) to process the temperature and light duration data in the comprehensive environmental impact index, specifically: calculating the average temperature and the average light duration in each window, then calculating the mean difference of adjacent windows, dividing the mean difference by 24 hours, and obtaining the dynamic change rate of temperature and the dynamic change rate of light duration. The temperature change rate weight is set to 0.6 and the light change rate weight is set to 0.4. The temperature dynamic change rate and the light duration dynamic change rate are multiplied by the corresponding weights and then added to obtain the environmental correction factor. The environmental correction factor is multiplied by the proportion of each depth (1 / 3) to obtain the basic correction value of the corresponding depth. The basic correction value of each depth is then multiplied by the soil activity weight of the depth layer. Finally, three depth correction coefficients are obtained, corresponding to the three depths of soil.

[0109] The image feature vector is extracted to obtain the aboveground growth rate characteristics, which are then integrated with the depth correction coefficient to obtain the preliminary root activity coefficient. Specifically, the following steps are performed: extracting the time series data corresponding to leaf area and plant height from the image feature vector, sorting them by acquisition time, and using the sliding window method (window 24 hours, step length 6 hours) to calculate the average leaf area and average plant height in each window. Then, the mean difference between two adjacent windows can be calculated (that is, the mean of the latter window minus the mean of the previous window), and the mean difference is divided by 24 hours to obtain the unit time change of leaf area and plant height. The linear regression algorithm is used to perform trend fitting on these two changes, and the slope of the fitting line is the aboveground growth rate characteristic. The weight of the growth rate characteristic is set to 0.3, and the three depth correction coefficients are each one-third of 0.7. The growth rate characteristic is multiplied by 0.3, and the three depth correction coefficients are multiplied by 0.7 and divided by 3. All products are then added together, and the result is the preliminary root activity coefficient.

[0110] Determine the baseline value of root activity for the same period in history based on historical data;

[0111] Calculating the deviation rate between the preliminary root activity coefficient and the root activity baseline value, specifically comprising: subtracting the root activity baseline value from the preliminary root activity coefficient to obtain a difference, and then dividing the difference by the root activity baseline value, the result of which is the deviation rate;

[0112] The preliminary root activity coefficient is calibrated based on the deviation rate to obtain the root depth activity coefficient, which includes: calculating each soil layer Exponential decay weight , each soil layer Exponential decay weight and soil activity weight and environmental correction factor Multiply to get the product term , the product terms of all soil layers are summed and then divided by the sum of the exponential decay weights to obtain the normalized depth-weighted average term ; The three-day growth acceleration rate Multiply by growth sensitivity And take the negative exponent to get the exponential response term , divide 2 by , and get the response ratio ; The depth weight of each layer Multiply by the environmental correction factor , and get the product result , sum the product results of all soil layers and get the sum result , subtract the historical benchmark value from the summation result Divide by the benchmark standard deviation , and obtain the deviation calibration term , apply the S-type function to the bias calibration term and map it to The normalized deviation calibration term is obtained ; The aboveground growth rate , the normalized depth weighted average term, the response ratio term, and the normalized deviation calibration term are multiplied to obtain the root depth activity coefficient , when applied specifically, it can be achieved through the following calculation formula, for example: ;

[0113] Where, represents the root depth activity coefficient, represents the aboveground growth rate, represents the number of soil layers, express The index of represents the base of natural logarithms, represents the depth attenuation factor (0.5), Indicates the Layer soil activity weight, , Indicates the Layer environment correction factor, , represents growth sensitivity (0.8), Indicates the three-day growth acceleration rate, represents the sigmoid function, represents the depth weight, , Indicates the historical benchmark value, represents the benchmark standard deviation;

[0114] The following are Explanation of the value range:

[0115] When dense planting occurs, competition for resources such as light, water and nutrients becomes intense. The soil structure deteriorates, resulting in the inhibition of root activity and poor soil activity. The value is low, The value is ;

[0116] When sparsely planted, resources such as light, water and nutrients can be fully utilized by crops, and the soil activity is better, so The value is too high. The value is ;

[0117] The following are Explanation of the value range:

[0118] When soil moisture content The water holding capacity of the entire planting area is 40%, and the water absorption resistance of the crop roots increases, making it difficult to absorb water. The value is low, The value is ;

[0119] When soil moisture content 40% of the water holding capacity of the entire planting area, the water absorption efficiency of the root system is improved, so The value is too high. The value is ;

[0120] The following are Explanation of the value range:

[0121] When the root depth of crops is mainly concentrated in the shallow soil layer (0–20 cm), the soil depth is mainly reflected in the surface layer. The value is low, The value is ;

[0122] When the root depth of crops is mainly concentrated in the deep soil layer, that is, 20 40cm, 40 When the soil depth is 60cm, it is mainly reflected in the deep layer. The value is too high. The value is .

[0123] By integrating multi-dimensional data such as soil activity weights, environmental correction coefficients, and above-ground growth rates, the combination of big data and smart agriculture is achieved. Through methods such as exponential decay weights and bias calibration, root activity at different soil depths is analyzed in layers. By adapting parameters such as planting density, soil moisture content, and root distribution depth, the effects of resource competition, water absorption efficiency, etc. on the root system can be specifically reflected, breaking through the limitation of only monitoring the above-ground part and allowing growers to intuitively understand potential problems such as root hypoxia.

[0124] The problem of monitoring lag is effectively solved through the dynamic calculation of the root depth activity coefficient. The dynamic changes of environment and growth are captured through a sliding window, and the deviation is calibrated using historical benchmark values, so that the coefficient can respond to root physiological changes in real time. Compared with lagging phenomena such as leaf wilting, this coefficient can provide early warning of root problems of crops through abnormal correlation between soil activity and growth rate, facilitating timely intervention in greenhouse cultivation.

[0125] In one aspect of this embodiment, determining the root activity baseline value for the same period in history based on historical data includes:

[0126] Obtain the current date and crop planting date, and based on these two dates, determine the growth stage matching factor. Specifically, this includes: calculating the difference between the current date and the crop planting date, which is the number of days of growth. Based on the number of days of each growth stage in historical data, such as the seedling stage (1-30 days) and the jointing stage (31-60 days), determine the current stage. Subtract the number of days of growth in this stage from the start of the current stage, and then divide it by the total number of days in the stage (for example, the seedling stage is 30 days). The result is the growth stage matching factor.

[0127] Screening growth stage matching factors from historical data All healthy crop samples within 5% of the original data were used to construct a spatiotemporal similarity sample set, specifically including: extracting the growth stage matching factors of all healthy crop samples from historical data, calculating the difference between each sample factor and the current growth stage matching factor, screening out samples whose absolute value of the difference does not exceed 5% of the current factor, and grouping these samples by planting year and growth stage to form a spatiotemporal similarity sample set;

[0128] Cluster analysis was performed on the spatiotemporally similar sample sets to obtain the basic activity distribution interval. Specifically, the K-means clustering algorithm was used to analyze the spatiotemporally similar sample sets. Specifically, the root activity values ​​of healthy crops in the spatiotemporally similar sample sets were used as clustering features. The number of clusters was set to 3. The Euclidean distance between the root activity value of each sample and the three initial cluster centers was calculated. The samples were divided into the cluster groups with the closest distance. The cluster centers were iteratively updated until the centers no longer changed. At this time, the root activity values ​​of each class of samples were counted. The minimum and maximum values ​​in each class were taken to form the distribution interval of that class. The three intervals were combined to obtain the basic activity distribution interval.

[0129] Extract environmental data and construct an environmental correction tensor. This involves extracting temperature, humidity, light intensity, and carbon dioxide concentration from the environmental data. Data from the past seven days are taken and the daily mean is calculated. The analytic hierarchy process is used to determine the weights of each parameter. The daily mean is multiplied by the corresponding weight to obtain the daily correction value. These correction values ​​are arranged with days as rows and parameters as columns to form a 4-column, 7-row matrix, which is the environmental correction tensor.

[0130] The basic activity distribution interval is adjusted according to the environmental correction tensor to generate an environmental adaptability benchmark interval, specifically comprising: calculating the arithmetic mean of the daily correction values ​​in the environmental correction tensor to obtain a correction coefficient; multiplying the minimum and maximum values ​​of the basic activity distribution interval by the correction coefficient respectively to obtain the adjusted minimum and maximum values ​​of the interval; the interval formed by the two values ​​is the environmental adaptability benchmark interval;

[0131] The environmental adaptability benchmark interval was disassembled to separate the trend part and the cycle part, and the root activity benchmark value for the same period in history was calculated, specifically including: using the 3-day moving average method to process the values ​​of the environmental adaptability benchmark interval, specifically: taking the arithmetic average of the values ​​of 3 consecutive days in the environmental adaptability benchmark interval to obtain the trend part, subtracting the corresponding trend part value from the original interval value, and the remaining part is the cycle part. The arithmetic average of the trend part and the cycle part were calculated respectively, and the result of adding the two is the root activity benchmark value for the same period in history.

[0132] By accurately determining the baseline value of root activity during the same period in history, the reliability of root monitoring is improved. By combining growth stage matching with K-means clustering, the basic activity interval is extracted from healthy samples, and then dynamically adjusted through the environmental correction tensor to make the baseline value adapt to the current environment. This not only reflects the growth patterns of crops, but also incorporates environmental influences, avoiding judgment bias caused by environmental differences. It can more accurately identify root abnormalities, make up for the shortcomings of monitoring based solely on ground data, and enhance the ability of smart agriculture to perceive the true growth status of crops.

[0133] In one case of this embodiment, correlation analysis is performed on the aboveground physiological advance value and the root depth activity coefficient to obtain a crop health correlation index, including:

[0134] The aboveground physiological advance value and the root depth activity coefficient are decomposed to generate a multi-dimensional growth coordination tensor, which includes: extracting the four basic sub-items of leaf number, area, plant height, and contour characteristics from the aboveground physiological advance value; for the root depth activity coefficient, 20cm, 20 40cm, 40 Three 60cm soil layers were used to extract the active sub-items corresponding to each layer. These were then classified according to three dimensions: time (daily), growth indicator type (aboveground / belowground), and soil depth. A normalization algorithm was used to convert all sub-item values ​​to the 0-1 interval to obtain a baseline value. Using the analytic hierarchy process, the importance of each sub-item in the corresponding dimension was compared pairwise to determine its respective weight. Under the cross-classification of the three dimensions, the baseline value of each sub-item was multiplied by its weight. The results for the same dimensional combinations were then summed to form a three-dimensional array, namely, a multi-dimensional growth coordination tensor.

[0135] The matching degree of the above-ground and underground characteristic components in the growth coordination tensor was calculated to obtain the dynamic correlation between the growth cycle and spatial distribution. Specifically, the above-ground characteristic components (the number of leaves, area, plant height, and contour characteristics) and the underground characteristic components (three-layer soil activity) were aligned in the growth coordination tensor according to the daily time dimension. The cosine similarity algorithm was used to calculate the similarity between the above-ground and underground components at the same time point (by comparing the degree of fit of the numerical distribution of the two) to obtain the daily matching degree. The growth cycle was divided into segments (for example, every 7 days as a cycle), and the average daily matching degree in each cycle was calculated as the cycle matching degree. The soil depth weight of the underground component (the weight of the shallow layer 0-20cm is 0.5, the weight of the middle layer 20) was used as the cycle matching degree. 40cm weight is 0.3, depth 40 The weight of 60cm is 0.2), and the period matching degree is multiplied by the corresponding depth weight and then summed. The summation result is the dynamic correlation between the growth period and spatial distribution;

[0136] Determine the growth correlation of healthy crops in the same period based on historical data and establish a health fluctuation feature library;

[0137] Calculating the difference between the current dynamic correlation and the health fluctuation feature library to generate the health feature deviation, specifically including: extracting historical dynamic correlation data matching the current growth stage from the health fluctuation feature library, calculating the arithmetic mean thereof, subtracting the arithmetic mean from the current dynamic correlation to obtain the difference, then calculating the standard deviation of the historical dynamic correlation data, dividing the difference by the standard deviation, and the result obtained is the health feature deviation;

[0138] The dynamic correlation and health feature deviation were analyzed to obtain the crop health correlation index, which specifically included: setting the dynamic correlation weight to 0.6 and the health feature deviation weight to 0.4, taking the absolute value of the health feature deviation, and then subtracting the absolute value from 1 to obtain the health feature adaptation, multiplying the dynamic correlation by 0.6 and the health feature adaptation by 0.4, and then adding the two product results to obtain the crop health correlation index.

[0139] By generating a crop health correlation index, a linkage analysis of above-ground and underground growth status is achieved. It decomposes the above-ground physiological advance value and the root depth activity coefficient into multi-dimensional sub-items, constructs a growth coordination tensor, calculates the dynamic correlation degree based on cosine similarity, and then compares it with the health fluctuation feature library to generate deviation. It integrates above-ground and underground data and can accurately reflect the growth coordination of the two. Before abnormalities occur in leaves, it warns of root problems through changes in correlation, solves the lag of monitoring only the above-ground part, provides comprehensive and timely health assessments for greenhouse cultivation, and improves the accuracy of smart agricultural monitoring.

[0140] In one case of this embodiment, the growth correlation of healthy crops in the same period in history is determined based on historical data, and a health fluctuation feature library is established, including:

[0141] Based on historical data, the healthy crop data at different growth stages during the same historical period are determined, specifically including: collecting growth data and environmental data of crops during the same historical period, screening out the recognized healthy crop data, dividing the stages by growth days, such as the seedling stage (1-30 days) and the jointing stage (31-60 days), and classifying the health data of each stage by indicators (such as the number of leaves, area, plant height, outline characteristics, etc.), and using the median algorithm to calculate the same indicator value of all healthy samples in that stage. The result is the indicator data of healthy crops at that historical period and that growth stage, and thus the healthy crop data at different growth stages during the same historical period can be determined;

[0142] Calculate the matching degree of healthy crop data with the growth stage of the current crop and the similarity of environmental factors to generate a healthy sample screening index, specifically including: about the matching degree of growth stage, specifically: calculate the difference between the current crop growth days and the growth days of historical healthy samples, for example, the current crop is on the 15th day of a certain growth stage, and a certain historical sample is on the 18th day of the same stage, the difference between the two is 3 days, subtract 1 (the absolute value of the difference divided by the typical duration of the growth stage), and then take the maximum value with 0 to get the growth stage matching degree, and can ensure that the matching degree is within It is within the range of 0-1 and reflects continuity. Regarding the similarity of environmental factors, specifically: take environmental factors such as temperature and humidity, calculate the factor mean of the current and historical samples, subtract (the absolute value of the mean difference divided by the historical standard deviation of the factor) from 1 for each factor, and limit the result to between 0-1. Calculate the arithmetic mean of all factor similarities to obtain the environmental factor similarity; set the growth stage matching weight to 0.6 and the environmental factor similarity weight to 0.4, multiply the growth stage matching and environmental factor similarity by the corresponding weights respectively, and sum them. The result is the healthy sample screening index;

[0143] Based on the healthy sample screening index, target historical healthy crop samples were screened out to obtain a growth-related feature matrix. Specifically, the following steps were performed: a screening threshold (0.8) was set, historical samples whose healthy sample screening index was not lower than the screening threshold were retained, and historical samples were used as target samples. Aboveground indicators (number of leaves, plant height, etc.) and underground indicators (root activity at each layer) of these target samples were extracted. The samples were sorted in order of growth time, and the aboveground indicators of each sample were used as rows of the matrix, and the underground indicators were used as columns of the matrix. The indicator data at different time points corresponded to the values ​​at different positions in the matrix. The matrix formed by the permutations and combinations was the growth-related feature matrix.

[0144] The growth correlation feature matrix was analyzed and a health fluctuation feature library was established. Specifically, the growth correlation feature matrix was split according to the growth stage to obtain the sub-matrix of each stage. For the above-ground and underground indicators of each stage, the mean and standard deviation of each indicator were calculated. The mean plus or minus 3 times the standard deviation was taken as the normal fluctuation range of the indicator. Each sub-matrix was divided into the growth stage, indicator type (above-ground, below-ground, and below-ground) and the health fluctuation feature library was established. The health fluctuation feature library is established by classifying the indicators (underground) and recording the mean, standard deviation and fluctuation range of each indicator.

[0145] By combining the matching degree of growth stages with the similarity of environmental factors to screen target samples, we ensure that the screened historical health samples are highly similar to the current crops. The growth correlation feature matrix can accurately reflect the correlation patterns of above-ground and underground indicators. At the same time, the health fluctuation feature library effectively improves the foresight of monitoring. The library covers the correlation data of above-ground and underground indicators at different growth stages, which can be quickly compared with the real-time data of current crops. Before the leaves wilt and other phenomena appear, root problems can be detected in advance through abnormal fluctuations in indicators, which makes up for the lag of monitoring only the above-ground part, allowing greenhouse growers to grasp the real growth status of crops in a timely manner and achieve the monitoring effect of smart agriculture.

[0146] In one case of this embodiment, determining the health status of crops according to the crop health correlation index includes:

[0147] Determining the healthy threshold interval based on the aboveground physiological advance value, root depth activity coefficient, and historical data involves: screening healthy crop samples from the past five growth cycles from historical data, extracting their aboveground physiological advance value and root depth activity coefficient, calculating the mean and standard deviation of the aboveground physiological advance value and root depth activity coefficient, and using the mean plus or minus three times the standard deviation as the normal fluctuation range of the aboveground physiological advance value and root depth activity coefficient, respectively. The intersection of the two is the healthy threshold interval;

[0148] The health threshold interval is compared with the crop health association index to obtain the crop health status, specifically including: comparing the crop health association index with the health threshold interval, if the crop health association index is within the health threshold interval, it is judged to be healthy; if the crop health association index is lower than the lower limit of the health threshold interval or higher than the upper limit of the health threshold interval, it is judged to have abnormal growth.

[0149] The health threshold interval is obtained by calculating the mean and standard deviation through the above-ground physiological advance value and root depth activity coefficient of historical health samples, taking into account the normal fluctuation range of above-ground and underground growth indicators. Comparing the health-related index with this interval can intuitively determine whether the crop is healthy or not, avoiding the subjectivity of judging based solely on surface morphology, effectively solving the problem of monitoring lag, and improving the timeliness of health judgment. The health threshold interval integrates the above-ground physiological advance value and the root depth activity coefficient. These two indicators can reflect crop physiological changes in advance. When the health-related index exceeds the interval, growth abnormalities can be determined before phenomena such as leaf wilting appear, allowing growers to intervene in time. This monitoring and judgment method based on big data can capture the true growth status of crops in a timely manner and ensure the monitoring effect of smart agriculture.

[0150] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The smart agricultural monitoring system based on big data is characterized by: include: The division unit is used to obtain the maximum diameter of the ground and the maximum extension radius of the root system of each crop in the greenhouse, determine the dynamic grid size of the single crop, and divide the planting area into multiple dynamic grids based on the dynamic grid size, each grid covering the single crop and the extension range of its root system; The integration unit is used to collect image data, environmental data, and soil data at different depths of individual crops in each grid in real time, and integrate the three pre-processed data to obtain a big data fusion matrix for individual crops; The analysis unit is used to analyze the big data fusion matrix to obtain the aboveground physiological advance value and the root depth activity coefficient respectively; The correlation unit is used to perform correlation analysis on the aboveground physiological advance value and the root depth activity coefficient to obtain the crop health correlation index; The evaluation unit is used to determine the health status of crops based on crop health correlation indexes.

2. The smart agriculture monitoring system based on big data according to claim 1 is characterized in that: Determine dynamic grid size for individual crops, including: The maximum above-ground diameter and the maximum root extension radius of each crop were calculated to obtain the proportional coefficient; Determine the redundant root coverage radius based on the proportional coefficient and the maximum above-ground diameter of each crop plant; Taking a single crop plant as the center, the planting area is divided into square grids with a side length of twice the root redundancy coverage radius to generate a dynamic grid size for a single crop plant.

3. The smart agriculture monitoring system based on big data according to claim 1 is characterized in that: The three pre-processed data are integrated to obtain the big data fusion matrix of individual crops, including: Extract image data to obtain image feature vector; Analyze the correlation between each data in the environmental data, and integrate the correlation to obtain a comprehensive environmental impact index; Analyze soil data at different depths to obtain soil characteristic tensors; The image feature vector, comprehensive environmental impact index and soil feature tensor are fused to generate a big data fusion matrix of a single crop.

4. The smart agriculture monitoring system based on big data according to claim 3 is characterized in that: Analyze the big data fusion matrix to obtain the ground physiological lead value, including: Perform image extraction on the big data fusion matrix to obtain the time series image change rate; Determine the environmental physiological response coefficient based on the time series image change rate and the comprehensive environmental impact index; Screen the soil characteristic tensor and generate shallow soil impact weights; Obtain historical data of crops, analyze the historical data, and obtain the historical mean of image feature vectors at the same growth stage; The historical mean is corrected based on the comprehensive environmental impact index of the same period to obtain the dynamic physiological benchmark vector.

5. The smart agriculture monitoring system based on big data according to claim 4 is characterized in that: Analyze the big data fusion matrix to obtain the ground physiological lead value, including: The time series image change rate, environmental physiological response coefficient, and shallow soil impact weight are integrated to obtain the current physiological characteristic prediction value; Compare the current physiological characteristic prediction value with the dynamic physiological reference vector to obtain the advance prediction deviation; The lead prediction deviation is calculated to generate the above-ground physiological lead value.

6. The smart agriculture monitoring system based on big data according to claim 3 is characterized in that: The big data fusion matrix is ​​analyzed to obtain the root depth activity coefficient, including: The soil characteristic tensor is parsed to extract the nutrient availability and water conductivity of the soil at each depth, and the depth soil activity weight is obtained; Based on the dynamic changes of temperature and light duration in the comprehensive environmental impact index, the depth soil activity weight is corrected to generate a depth correction coefficient; Extract the image feature vector to obtain the aboveground growth rate characteristics, and fuse the aboveground growth rate characteristics with the depth correction coefficient to obtain the preliminary root activity coefficient; Determine the root activity baseline value for the same period in history based on historical data, and calculate the deviation rate between the preliminary root activity coefficient and the root activity baseline value; The preliminary root activity coefficient was calibrated based on the deviation rate to obtain the root depth activity coefficient.

7. The big data-based smart agriculture monitoring system according to claim 6, characterized in that: Determine the baseline root activity value for the same period based on historical data, including: Get the current date and the crop planting date, and determine the growth stage matching factor based on the two dates; All healthy crop samples within the growth stage matching factor of ±5% were screened from historical data to construct a spatiotemporal similarity sample set; Cluster analysis is performed on the spatiotemporally similar sample sets to obtain the basic activity distribution interval; Extract environmental data and construct environmental correction tensor; Adjust the basic activity distribution interval according to the environmental correction tensor to generate the environmental adaptability benchmark interval; The environmental adaptability benchmark interval was disassembled to separate the trend part and the cycle part, and the root activity benchmark value for the same historical period was calculated.

8. The smart agriculture monitoring system based on big data according to claim 6 is characterized in that: Correlation analysis was performed on the aboveground physiological advance value and the root depth activity coefficient to obtain the crop health correlation index, including: Decompose the aboveground physiological advance value and root depth activity coefficient to generate a multi-dimensional growth coordination tensor; The matching degree of above-ground and underground characteristic components in the growth coordination tensor is calculated to obtain the dynamic correlation between the growth cycle and spatial distribution. Determine the growth correlation of healthy crops in the same period based on historical data and establish a health fluctuation feature library; Calculate the difference between the current dynamic correlation and the health fluctuation feature library to generate the health feature deviation; The dynamic correlation degree and health characteristic deviation degree were analyzed to obtain the crop health correlation index.

9. The big data-based smart agriculture monitoring system according to claim 8, characterized in that: Based on historical data, we determine the growth correlation of healthy crops in the same period in history and establish a health fluctuation feature library, including: Based on historical data, determine the healthy crop data at different growth stages during the same period in history; Calculate the matching degree between healthy crop data and the growth stage of current crops and the similarity of environmental factors to generate a healthy sample screening index; Based on the healthy sample screening index, target historical healthy crop samples are screened to obtain the growth correlation feature matrix; The growth-related characteristic matrix was analyzed and a health fluctuation characteristic library was established.

10. The smart agriculture monitoring system based on big data according to claim 8, characterized in that: Determine crop health status based on crop health related indices, including: Determine the health threshold interval based on aboveground physiological advance values, root depth activity coefficients, and historical data; The health threshold interval is compared with the crop health association index to obtain the crop health status.

Images