Agricultural water and fertilizer management method and system based on cloud platform

Abstract

The present application relates to water and fertilizer management technical field, specifically to a kind of agricultural water and fertilizer management method and system based on cloud platform, including the following steps, utilize sensor to collect crop image and extract chlorophyll and leaf area characteristics, analyze variation rate to determine crop growth process and time sensitivity, construct index type dynamic weight mapping relationship, task resource coupling index is calculated in combination with cloud node load and fertilization data fluctuation, based on index task sequencing and execution time delay deviation compensation, generate water and fertilizer management instruction.In the present application, by fusing multi-dimensional morphological features and environmental load data to construct a dynamic control model, a nonlinear correlation between crop demand and execution weight is established using an exponential function mapping mechanism, precise perception of crop growth demand and adaptive optimization of task execution timing are achieved, the task queue is flexibly adjusted according to the resource coupling state, resource competition conflicts are effectively avoided, and the efficiency of water and fertilizer collaborative management is improved.

Description

Technical Field

[0001] This invention relates to the field of water and fertilizer management technology, and in particular to an agricultural water and fertilizer management method and system based on a cloud platform. Background Technology

[0002] The field of water and fertilizer management technology involves the coordinated allocation and control of water resources and fertilizers in agricultural production. Core aspects include irrigation strategy formulation, fertilizer application control, crop water and fertilizer requirement model construction, meteorological and soil information collection and analysis, and control mechanisms for irrigation and fertilization equipment. This technology aims to improve crop yield and quality, conserve water and fertilizer resources, and reduce environmental burden, broadly encompassing automated irrigation control, precision fertilization decision-making, IoT sensing networks, and data-driven management models. In the process of water and fertilizer management, water and fertilizer demand analysis is conducted based on multi-source data such as soil moisture, crop growth stage, and meteorological conditions. This analysis, combined with supporting facilities such as drip irrigation systems, fertilizer tanks, and control valves, enables the implementation of water and fertilizer supply strategies, making it one of the key supporting technologies for sustainable agricultural development.

[0003] Traditional agricultural water and fertilizer management methods rely on manual experience or simple timed irrigation and quantitative fertilization equipment in farmland management. Fixed irrigation and fertilization frequencies and amounts are set according to crop type and growth cycle, lacking effective acquisition and comprehensive analysis of information such as real-time crop growth status, soil nutrient dynamics, and weather trends. Common methods include manually inspecting fields to determine crop status and soil moisture levels, and operating irrigation and fertilization equipment manually or semi-automatically. While some areas have timed control devices, a data-driven adjustment mechanism is lacking, resulting in a low match between water and fertilizer supply and actual demand, leading to water waste or insufficient fertilizer efficiency. Some methods use sensors to acquire basic data, but only for recording purposes, without further integration with the water and fertilizer control process, and without building an integrated management platform to analyze and process multi-source information and output strategies.

[0004] Traditional management methods rely on manual experience or simple timers to perform irrigation and fertilization tasks. They lack dynamic response mechanisms when dealing with the complex and ever-changing needs of crop growth, resulting in a serious disconnect between water and fertilizer supply and the actual physiological state of crops. Furthermore, existing equipment is unable to detect real-time fluctuations in soil environment and weather conditions, making it impossible to establish effective linkage and control strategies. This leads to spatiotemporal mismatches in the transmission and transformation of agricultural resources, causing water waste and low fertilizer utilization rates, which seriously restricts the efficiency of modern agricultural production and the level of refined management. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of existing technologies by proposing an agricultural water and fertilizer management method and system based on a cloud platform.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: an agricultural water and fertilizer management method based on a cloud platform, comprising the following steps:

[0007] S1: Obtain a continuous sequence of crop leaf images, extract the mean change of the G component in the RGB channels, use the Sobel operator to calculate the gradient difference between the leaf edge and the center pixel, and combine the mean change of the G component to calculate the chlorophyll concentration variation rate.

[0008] S2: Collect the crop leaf area growth rate per unit time, perform correlation analysis between the chlorophyll concentration variation rate and the crop leaf area growth rate, identify the density of the maximum slope change point, and extract the crop time sensitivity features.

[0009] S3: Call the crop time sensitivity feature to obtain the historical irrigation response efficiency data of the plot, substitute the crop time sensitivity feature into the exponential function weight mapping relationship to generate exponential dynamic weight parameters;

[0010] S4: Obtain the standard deviation sequence of fertilizer application amount for the same plot over multiple years, discretize the standard deviation sequence to construct a fertilizer data fluctuation density index, collect the real-time processing load value of cloud computing nodes, and perform a ratio calculation between the fertilizer data fluctuation density index and the real-time processing load value to generate a task resource coupling index.

[0011] S5: Based on the task resource coupling index, sort the water and fertilizer tasks and calculate the priority difference between adjacent tasks. For priority differences less than the preset resource competition threshold, call the exponential dynamic weight parameter to calculate the time delay deviation compensation and generate water and fertilizer management instructions.

[0012] As a further aspect of the present invention, the chlorophyll concentration variation rate includes concentration decay trend value, color gradient change rate, and nutrient deficiency warning confidence level; the crop time sensitivity characteristics include critical growth period determination value, rapid growth duration span, and fertilizer urgency coefficient; the exponential dynamic weight parameters include real-time scheduling gain value, historical response correction term, and dynamic priority ladder; the task resource coupling index includes computing power supply-demand ratio, task execution urgency, and node throughput matching score; and the water and fertilizer management instructions include regulating valve opening value, fertilizer pump injection flow rate, and target area execution sequence.

[0013] As a further aspect of the present invention, the step of obtaining the chlorophyll concentration variation rate specifically includes:

[0014] S101: Acquire crop leaf image sequences at consecutive time points through an image sensor, perform RGB color space calculation on a single frame image and separate the green component matrix, calculate the arithmetic mean of gray values ​​of pixels in the effective leaf area within the matrix, perform first-order difference operation on the arithmetic mean of gray values ​​of adjacent time points, quantify the overall drift of leaf color values ​​in the time dimension, and generate a green channel mean difference index.

[0015] S102: For the crop leaf image sequence, load the Sobel discrete differential operator to perform planar convolution processing, calculate the gradient vectors in the horizontal and vertical directions, synthesize the gradient magnitude matrix of the whole image, divide the leaf into edge region of interest and center region of interest based on pixel coordinate position, calculate the average pixel gradient intensity in the two regions respectively, perform difference calculation on the average gradient intensity of the edge region and the center region and extract the absolute value to generate regional gradient difference magnitude features.

[0016] S103: Detect the sign of the green channel mean difference index to determine the direction of concentration evolution, construct a dynamic correction coefficient using the regional gradient difference amplitude feature, perform weighted compensation on the difference index magnitude, obtain the timestamp difference at the image acquisition time, divide the compensated difference index value by the timestamp difference, and generate the chlorophyll concentration variation rate.

[0017] As a further aspect of the present invention, the process of dividing the leaf into edge regions of interest and central regions of interest based on pixel coordinates specifically involves: extracting the binarized mask of a single frame image from the crop leaf image sequence and calculating the geometric centroid coordinates of the binarized mask; traversing the effective pixels within the binarized mask to calculate the Euclidean distance from each pixel to the geometric centroid coordinates; setting a region division distance threshold, wherein the value of the region division distance threshold is set based on a preset scaling factor of the maximum inscribed circle radius of the binarized mask; marking the set of pixels with an Euclidean distance greater than the region division distance threshold as edge regions of interest, and marking the set of pixels with an Euclidean distance less than or equal to the region division distance threshold as central regions of interest.

[0018] As a further aspect of the present invention, the step of obtaining the crop time sensitivity feature specifically includes:

[0019] S201: Use a multispectral camera to acquire crop canopy images and calculate vegetation indices. Extract leaf region pixels through image segmentation and convert them into physical area values. Calculate the area increment within a unit time interval to obtain the crop leaf area growth rate. Call the chlorophyll concentration variation rate mentioned above and map the two to a unified dimension interval. Construct a dual-channel signal including morphological expansion and physiological variation. Calculate the covariance matrix of the dual-channel signal using a sliding window and establish a growth rate-related state sequence.

[0020] S202: For the growth rate associated state sequence, perform first-order difference operation to extract the slope curve, identify the abrupt change moment when the absolute value of the slope exceeds the preset steady-state threshold, calculate the time interval density between consecutive abrupt change moments, mark the time period with density value higher than the preset frequency benchmark as the high-frequency fluctuation area, count the cumulative number of slope sign flips in the area, and generate the maximum slope change point density.

[0021] S203: Call the density of the maximum slope change points and input it into the preset crop growth process classifier for pattern matching. Determine that the crop is currently in a critical state of transition from vegetative growth to reproductive growth. Extract the extreme value of the growth rate acceleration under the state. Combine the historical water stress response data to calculate the resource demand urgency index for the current period and generate crop time sensitivity features.

[0022] As a further aspect of the present invention, the method for setting the preset steady-state threshold is as follows: collect the historical growth rate sequence of crops under standard growth conditions, calculate the first derivative distribution of the historical growth rate sequence, select the confidence interval boundary value of the first derivative distribution as a benchmark, combine the noise level of the current environmental sensor to perform weighted correction on the benchmark, and set the corrected value as the preset steady-state threshold.

[0023] The specific method for setting the preset frequency benchmark is as follows: based on the crop's physiological characteristics, the minimum delay time for the crop to respond to external water and fertilizer stimuli is obtained; the maximum frequency of physiological state switching that theoretically occurs per unit time is calculated; the maximum frequency is used as the initial benchmark; the initial benchmark is adjusted by introducing an environmental disturbance filtering coefficient; and the processed value is set as the preset frequency benchmark.

[0024] As a further aspect of the present invention, the step of obtaining the exponential dynamic weight parameter specifically includes:

[0025] S301: Retrieve irrigation event logs and corresponding crop growth monitoring data for plots from the local agricultural management database over multiple years, calculate the ratio of biomass increment to water consumption per unit time after each irrigation, construct a historical water use efficiency sequence, map the sequence to a closed interval between zero and one using the max-min normalization method, remove outlier noise data in the sequence, and generate historical irrigation response efficiency coefficients.

[0026] S302: An exponential function is used to fit the water and fertilizer demand urgency curve. The time sensitivity feature is used as a power exponential variable that determines the growth rate of the curve. The function base is adjusted using the historical irrigation response efficiency coefficient. The nonlinear transformation logic between the input feature and the scheduling weight is established, and a sensitivity-driven weight calculation model is established.

[0027] S303: Input the crop time sensitivity feature value obtained from real-time monitoring into the sensitivity-driven weight calculation model, perform floating-point exponentiation to calculate the original weight value, use the Sigmoid function to limit the saturation of the calculation result to prevent the weight value from overflowing, quantize the calculated value into dimensionless scalar data, and generate exponential dynamic weight parameters.

[0028] As a further aspect of the present invention, the step of obtaining the task resource coupling index specifically includes:

[0029] S401: Retrieve fertilization record data for a specified plot of land over multiple years from the local agricultural database, remove outliers, segment the fertilization data according to the crop growth cycle, calculate the degree of deviation of the fertilization amount from the average value in each cycle, and obtain the fertilization standard deviation sequence.

[0030] S402: Based on the numerical distribution range of the standard deviation sequence of the fertilizer application amount, set equally spaced discretized windows, count the number of sequence elements falling into each window to calculate the distribution probability, and perform a weighted integral operation on the distribution probability in combination with the window center value to quantify the clustering characteristics of fertilization behavior in the time dimension and generate a fertilization data fluctuation density index.

[0031] S403: Real-time reading of the CPU load percentage, dynamic random access memory occupancy rate, and remaining bandwidth throughput of the network interface of the cloud platform computing node; measurement of end-to-end instruction transmission latency; calling the fertilization data fluctuation density index; and calculation of the task resource coupling index.

[0032] As a further aspect of the present invention, the step of obtaining the water and fertilizer management instruction specifically includes:

[0033] S501: Call the task resource coupling index, perform local sorting of multiple water and fertilizer tasks in the same irrigation area in descending order of index value, calculate the index value difference between adjacent tasks in the sorting sequence to quantify the priority decay gradient, compare the calculated difference with the preset resource competition threshold to filter out the task pairs with high conflict probability, and generate the adjacent task priority difference sequence.

[0034] S502: Call the exponential dynamic weight parameter and the priority difference sequence of adjacent tasks, retrieve the average historical response lag time and hydraulic stability duration of the associated execution valve from the equipment operation log library, use multiplication to combine the exponential dynamic weight parameter with the aggregate lag time value to obtain the dynamic compensation offset, add the offset to the original plan timestamp to reconstruct the task triggering logic, and generate the execution parameters after time delay deviation compensation.

[0035] S503: Based on the execution parameters after the time delay deviation compensation, the compensated and corrected flow demand and time axis data are converted into electrical control signals for the opening degree of the solenoid valve and the frequency of the booster pump. The signals are mapped to the register address of the field programmable logic controller. The address segment and data segment are assembled into a standard instruction package including a cyclic redundancy check code to generate water and fertilizer management instructions.

[0036] A cloud-based agricultural water and fertilizer management system, wherein the cloud-based agricultural water and fertilizer management system is used to implement the above-mentioned cloud-based agricultural water and fertilizer management method, the system comprising:

[0037] The chlorophyll monitoring module acquires a continuous sequence of crop leaf images, extracts the mean change of the G component in the RGB channels, calculates the gradient difference between the leaf edge and the center pixel using the Sobel operator, and calculates the chlorophyll concentration variation rate by combining the mean change of the G component.

[0038] The time-sensitive analysis module collects the crop leaf area growth rate per unit time, correlates the chlorophyll concentration variation rate with the crop leaf area growth rate, identifies the density of the maximum slope change point, and extracts the crop time-sensitive features.

[0039] The dynamic weight adjustment module calls the crop time sensitivity feature, obtains the historical irrigation response efficiency data of the plot, substitutes the crop time sensitivity feature into the exponential function weight mapping relationship to calculate, and generates exponential dynamic weight parameters.

[0040] The task resource analysis module obtains the standard deviation sequence of fertilizer application amount for the same plot of land over multiple years, discretizes the standard deviation sequence to construct a fertilizer data fluctuation density index, collects the real-time processing load value of cloud computing nodes, and calculates the ratio between the fertilizer data fluctuation density index and the real-time processing load value to generate a task resource coupling index.

[0041] The water and fertilizer management implementation module sorts water and fertilizer tasks based on the task resource coupling index and calculates the priority difference between adjacent tasks. For priority differences less than the preset resource competition threshold, it calls the exponential dynamic weight parameter to calculate the time delay deviation compensation and generates water and fertilizer management instructions.

[0042] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0043] In this invention, by extracting crop leaf image features and combining chlorophyll concentration variation with leaf area growth rate analysis, the crop growth process and time sensitivity are accurately determined. An exponential function is used to construct a dynamic weight mapping relationship to adaptively adjust task priorities. A resource coupling index is generated based on the load of cloud computing nodes and the fluctuation density of fertilization data. Delay deviation compensation calculations are performed on high-conflict tasks, and refined management instructions are generated. This effectively solves the problems of water and fertilizer supply being out of sync with crop demand and resource mismatch in time and space under traditional methods, and improves the accuracy of water and fertilizer coordinated regulation and resource utilization efficiency in agricultural production. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a schematic diagram of the workflow of the present invention;

[0046] Figure 2 This is a detailed flowchart of S1 of the present invention;

[0047] Figure 3 This is a detailed flowchart of the S2 process of the present invention;

[0048] Figure 4 This is a detailed flowchart of the S3 process of the present invention;

[0049] Figure 5 This is a detailed flowchart of the S4 process of the present invention;

[0050] Figure 6 This is a detailed flowchart of S5 of the present invention;

[0051] Figure 7 This is a system flowchart of the present invention. Detailed Implementation

[0052] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0053] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0054] Please see Figure 1 This invention provides a technical solution: an agricultural water and fertilizer management method based on a cloud platform, comprising the following steps:

[0055] S1: Acquire crop leaf image sequences at consecutive time points using an image sensor, extract the mean change of the G component in the RGB color value channel, use the Sobel operator to calculate the pixel gradient difference between the leaf edge region and the center region, combine the mean change of the G component to determine the chlorophyll concentration change trend, and calculate the chlorophyll concentration variation rate.

[0056] S2: Obtain the crop leaf area growth rate per unit time, perform correlation analysis between the chlorophyll concentration variation rate and the crop leaf area growth rate, determine the crop growth process status by identifying the density of the maximum slope change point, and extract crop time sensitivity features.

[0057] S3: Call the crop time sensitivity feature to obtain the historical irrigation response efficiency data of the plot, substitute the crop time sensitivity feature into the weight mapping relationship constructed by the exponential function for calculation, and generate exponential dynamic weight parameters;

[0058] S4: Obtain the standard deviation sequence of fertilizer application amount for the same plot over multiple years, divide the standard deviation sequence into discrete intervals to construct the fertilizer data fluctuation density index, collect the real-time processing load value of the cloud computing node in the region, calculate the ratio between the fertilizer data fluctuation density index and the real-time processing load value, and generate the task resource coupling index.

[0059] S5: Based on the task resource coupling index, multiple water and fertilizer tasks are locally sorted and the priority difference between adjacent tasks is calculated. For tasks with a priority difference less than a set threshold, the exponential dynamic weight parameter is invoked to perform task execution history delay deviation compensation calculation. Water and fertilizer management instructions are generated based on the compensation calculation results.

[0060] Chlorophyll concentration variation rate includes concentration decay trend value, color gradient change rate and nutrient deficiency warning confidence; crop time sensitivity characteristics include critical growth period determination value, rapid growth duration and fertilizer urgency coefficient; exponential dynamic weight parameters include real-time scheduling gain value, historical response correction term and dynamic priority ladder; task resource coupling index includes computing power supply and demand ratio, task execution urgency and node throughput matching score; water and fertilizer management instructions include control valve opening value, fertilizer pump flow rate and target area execution sequence.

[0061] Please see Figure 2 The specific steps for obtaining the chlorophyll concentration variation rate are as follows:

[0062] S101: Acquire crop leaf image sequences at consecutive time points through an image sensor, perform RGB color space calculation on a single frame image and separate the green component matrix, calculate the arithmetic mean of gray values ​​of pixels in the effective leaf area within the matrix, perform first-order difference operation on the arithmetic mean of gray values ​​of adjacent time points, quantify the overall drift of leaf color values ​​in the time dimension, and generate a green channel mean difference index.

[0063] A CMOS image sensor mounted on top of the crop continuously captures raw image data streams of the crop leaves at a preset acquisition frequency of 500 milliseconds. This sensor has a resolution of 24 megapixels and a dynamic range of 12 bits, enabling it to accurately record subtle color fluctuations under varying light conditions. For each acquired raw image frame, the image processing unit performs a de-Bayering operation to convert the RAW format data into a standard RGB full-color matrix. Subsequently, a channel separation algorithm is used to extract an independent green component matrix. This matrix is ​​a two-dimensional numerical array, where each element corresponds to the green channel grayscale value of a pixel, with values ​​ranging from 0 to 255. By traversing all values ​​in the green component matrix, background noise and interfering pixels from non-leaf areas are removed, and a set of pixels representing the effective leaf area is selected. An arithmetic mean operation is performed on the grayscale values ​​of the pixels within this set, i.e., the grayscale values ​​of all effective pixels are summed and divided by the total number of pixels to obtain the arithmetic mean of the grayscale values ​​of a single frame image. For example, at time node t1, the accumulated grayscale sum is 12,500,000, and the number of effective pixels is 100,000. Therefore, the arithmetic mean of the grayscale at that moment is calculated to be 125. Next, the arithmetic mean of the grayscale at the previous time node t0 is retrieved along the time axis and set to 128. Then, a first-order difference operation is performed: the mean of the current time t1 (125) is subtracted from the mean of the previous time t0 (128), resulting in a difference of -3. This difference quantifies the overall drift of the leaf's green color value within a 500-millisecond time interval. This difference is serialized and stored to generate a green channel mean difference index reflecting the short-term fluctuation trend of chlorophyll concentration.

[0064] S102: For crop leaf image sequences, load the Sobel discrete differential operator to perform planar convolution processing, calculate the gradient vectors in the horizontal and vertical directions, synthesize the gradient magnitude matrix of the whole image, divide the leaf into edge region of interest and center region of interest based on pixel coordinate position, calculate the average pixel gradient intensity in the two regions respectively, perform difference calculation on the average gradient intensity of the edge region and the center region and extract the absolute value to generate regional gradient difference magnitude features.

[0065] The process of dividing a leaf into edge regions of interest and a central region of interest based on pixel coordinates is as follows: First, extract the binarized mask of a single frame from the crop leaf image sequence and calculate the geometric centroid coordinates of the binarized mask. Second, traverse the effective pixels within the binarized mask to calculate the Euclidean distance from each pixel to the geometric centroid coordinates. Third, set a region division distance threshold, where the value of the region division distance threshold is set based on a preset scaling factor of the maximum inscribed circle radius of the binarized mask. Fourth, mark the set of pixels with an Euclidean distance greater than the region division distance threshold as edge regions of interest, and mark the set of pixels with an Euclidean distance less than or equal to the region division distance threshold as the central region of interest.

[0066] The image processing core invokes a pre-defined 3x3 Sobel discrete differential operator to perform planar convolution processing on the image's brightness channel in both the horizontal and vertical directions. The horizontal convolution kernel highlights the vertical edge texture in the image, calculating the horizontal gradient vector Gx; the vertical convolution kernel highlights the horizontal edge texture, calculating the vertical gradient vector Gy. Subsequently, the horizontal gradient vector Gx and the vertical gradient vector Gy at each pixel location are squared, and the square root of the sum of the two squared values ​​is taken to synthesize the gradient magnitude matrix of the entire image. This matrix intuitively reflects the texture complexity and vein clarity of the leaf surface. The process of dividing the leaf into edge regions of interest and center regions of interest based on pixel coordinates then begins. First, Otsu's adaptive thresholding method is performed on a single frame image to generate a binary mask, where the leaf region is marked as 1 and the background region is marked as 0. The first and zeroth moments of this binary mask are calculated using the image moment algorithm. By dividing the first moment by the zeroth moment, the precise pixel coordinates (Cx, Cy) of the leaf's geometric centroid are obtained. Subsequently, all valid pixels marked as 1 within the binarized mask are traversed, and the Euclidean distance from each pixel (Px, Py) to the geometric centroid (Cx, Cy) is calculated using the distance formula. A region segmentation distance threshold is set. First, the largest inscribed circle of the binarized mask is identified, and its radius is measured. For example, if the radius of the largest inscribed circle is 200 pixels, and the preset scaling factor is set to 1.2, the calculated region segmentation distance threshold is 240 pixels. All pixels with a calculated Euclidean distance greater than 240 pixels are classified as edge regions of interest, which typically cover the leaf tip and leaf margin serrations. Pixels with an Euclidean distance less than or equal to 240 pixels are classified as central regions of interest, covering the midrib and central mesophyll tissue. After segmentation, the gradient magnitudes of all pixels within the edge regions of interest are statistically analyzed, and their arithmetic mean is calculated as the average gradient intensity of the edge region. Similarly, the average gradient intensity of the central region is calculated. The average gradient intensity of the edge region is set to 45, and the average gradient intensity of the center region is set to 25. The difference is calculated by subtracting 25 from 45 to get 20, and the absolute value is extracted to finally generate the regional gradient difference amplitude feature used to characterize the heterogeneity of leaf texture.

[0067] S103: Call the green channel mean difference index and the regional gradient difference amplitude feature, detect the numerical sign of the green channel mean difference index to determine the direction of concentration evolution, construct a dynamic correction coefficient using the regional gradient difference amplitude feature, perform weighted compensation on the difference index magnitude, obtain the timestamp difference at the image acquisition time, divide the compensated difference index value by the timestamp difference, and generate the chlorophyll concentration variation rate.

[0068] The process of constructing dynamic correction coefficients using regional gradient difference amplitude characteristics specifically includes: obtaining a preset standard gradient heterogeneity benchmark value, wherein the standard gradient heterogeneity benchmark value is set by statistically calculating the arithmetic mean of the regional gradient difference amplitude characteristics of multiple groups of healthy crops under the same light conditions; calculating the positive deviation ratio between the regional gradient difference amplitude characteristics and the standard gradient heterogeneity benchmark value; and inputting the positive deviation ratio as an independent variable into a preset S-shaped growth function for mapping operation to output dynamic correction coefficients, ensuring that the compensation intensity is improved when the gradient difference increases significantly.

[0069] The process of applying weighted compensation to the magnitude of the difference index specifically includes: calculating the absolute value of the green channel mean difference index; multiplying the absolute value by the dynamic correction coefficient to obtain the compensation increment; and adding the compensation increment to the absolute value to generate the compensated difference index value.

[0070] The algorithm calls upon the green channel mean difference index and the regional gradient difference amplitude feature. First, it checks the sign of the green channel mean difference index. If the index is negative, it indicates a decreasing trend in chlorophyll concentration; if it is positive, it indicates an accumulation trend. Then, it constructs a dynamic correction coefficient using the regional gradient difference amplitude feature. This process first obtains a preset standard gradient heterogeneity baseline value from an agricultural database. This baseline value is set as follows: 100 sets of image data of the same healthy crop variety under standard greenhouse light conditions (light intensity 20,000 lux) are selected, and the arithmetic mean of their regional gradient difference amplitude features is calculated. For example, if the result is 15, the regional gradient difference amplitude feature of the current image is obtained and set to 20. Then, the positive deviation ratio between this value and the baseline value of 15 is calculated, i.e., (20 minus 15) divided by 15, resulting in 0.33. The ratio 0.33 is used as the independent variable x to input a preset S-shaped growth function. The function is y = 1 plus 1 divided by (1 plus e to the power of -5x). This function maps the input to a dynamic correction coefficient, ensuring that when the gradient difference is significantly greater than the baseline (meaning an increased risk of leaf edge curling or scorching), the compensation intensity is non-linearly increased. The dynamic correction coefficient of the mapped output is set to 1.8. The process of performing weighted compensation on the magnitude of the difference index then unfolds: the absolute value of the green channel mean difference index (e.g., the aforementioned -3) is calculated, which is 3. The absolute value 3 is multiplied by the dynamic correction coefficient 1.8 to obtain the compensation increment 5.4. The compensation increment 5.4 is superimposed back onto the original absolute value 3 to obtain the compensated difference index value 8.4. Finally, the difference between the timestamp t1 of the current image acquisition time and the timestamp t0 of the previous frame is obtained, for example, if the time difference is 0.5 seconds. The compensated value 8.4 is divided by 0.5 to generate a chlorophyll concentration variation rate of 16.8 per second.

[0071] Please see Figure 3 The specific steps for obtaining crop time sensitivity features are as follows:

[0072] S201: Use a multispectral camera to acquire crop canopy images and calculate vegetation indices. Extract leaf region pixels through image segmentation and convert them into physical area values. Calculate the area increment within a unit time interval to obtain the crop leaf area growth rate. Call the chlorophyll concentration variation rate and map the two to a unified dimension interval. Construct a dual-channel signal including morphological expansion and physiological variation. Calculate the covariance matrix of the dual-channel signal using a sliding window and establish a growth rate-related state sequence.

[0073] The process of calculating the covariance matrix of dual-channel signals using a sliding window is as follows: obtain the total duration of the crop growth cycle and the sampling frequency of image acquisition; determine the minimum time unit based on the sampling frequency; set the time span of the sliding window to be an integer multiple of the minimum time unit to cover the complete diurnal variation cycle of crop photosynthesis; set the movement step size of the sliding window to be a specific proportion of the time span; and calculate the variance of the crop leaf area growth rate and the chlorophyll concentration variation rate, as well as their covariance, within each sliding window to construct the covariance matrix.

[0074] Crops canopy images were acquired using a multispectral camera equipped with a red-edge band. The Normalized Difference Vegetation Index (NDVI) was calculated, and the total number of pixels in the leaf region was extracted using an image segmentation algorithm. Combining this with the camera's imaging distance and focal length parameters, the pixel count was converted into a physical area value (square centimeters). For example, if the total leaf area measured at time t1 was 500 square centimeters and at time t2 (1 hour interval) it was 502 square centimeters, the area increment per unit time interval was calculated as 2 square centimeters. Dividing this by the 1-hour time interval yielded a leaf area growth rate of 2 square centimeters per hour. Then, the chlorophyll concentration variation rate generated in the previous steps was used. Since the two rates have different dimensions (area rate is a morphological quantity, concentration rate is a physiological quantity), Z-score normalization was used to map them to a unified dimensional interval with a mean of 0 and a standard deviation of 1. A dual-channel signal sequence including morphological expansion and physiological variation was constructed. The process of calculating the covariance matrix of the dual-channel signal using a sliding window is as follows: the total duration of the crop growth cycle (e.g., 90 days) and the image acquisition sampling frequency (e.g., once per hour) were obtained, and the minimum time unit was determined to be 1 hour. The sliding window span is set to 24 times the smallest time unit, i.e., 24 hours, to cover the complete diurnal rhythm of crop photosynthesis. The sliding window step size is set to 25% of the time span, i.e., 6 hours. Within each 24-hour time window, 24 sets of data points are included. The variances and covariances of the crop leaf area growth rate sequence and the chlorophyll concentration variation rate sequence are calculated. For example, within a certain window, the area rate variance is 0.5, the concentration rate variance is 0.8, and the covariance is 0.4. A 2x2 covariance matrix is ​​constructed, and the eigenvalues ​​of these matrices are arranged in chronological order to establish a growth rate-related state sequence.

[0075] S202: For the growth rate-related state sequence, perform first-order difference operation to extract the slope curve, identify the abrupt change moment when the absolute value of the slope exceeds the preset steady-state threshold, calculate the time interval density between consecutive abrupt change moments, mark the time period with density value higher than the preset frequency benchmark as the high-frequency fluctuation area, count the cumulative number of slope sign flips in the area, and generate the maximum slope change point density.

[0076] The specific method for setting the preset steady-state threshold is as follows: collect the historical growth rate sequence of crops under standard growth environment, calculate the first derivative distribution of the historical growth rate sequence, select the confidence interval boundary value of the first derivative distribution as the benchmark, combine the current environmental sensor noise level to perform weighted correction on the benchmark, and set the corrected value as the preset steady-state threshold.

[0077] The specific method for setting the preset frequency benchmark is as follows: based on the crop's physiological characteristics, obtain the minimum delay time for the crop to respond to external water and fertilizer stimuli, calculate the maximum frequency of physiological state switching theoretically occurring per unit time, use the maximum frequency as the initial benchmark, and adjust the initial benchmark by introducing an environmental disturbance filtering coefficient, and set the processed value as the preset frequency benchmark.

[0078] Calculate the difference between state values ​​at adjacent time points and extract the slope curve. Identify abrupt changes where the absolute value of the slope exceeds a preset steady-state threshold. This preset steady-state threshold is set as follows: collect a historical growth rate sequence of the crop under standard growing conditions (e.g., data from the past 3 years), calculate the first derivative distribution of this sequence, and select the 95% confidence interval boundary value of the distribution (e.g., 0.05) as the benchmark. Simultaneously, read the noise level of current environmental sensors (e.g., wind speed sensors), set the noise figure to 1.1, multiply the benchmark 0.05 by 1.1 to obtain a corrected steady-state threshold of 0.055, scan the slope curve, and mark all points with absolute values ​​greater than 0.055 as abrupt changes. Calculate the time interval density between consecutive abrupt changes. The preset frequency benchmark is set as follows: based on crop physiological characteristics, obtain the minimum delay time (e.g., 10 minutes) for the crop's response to water and fertilizer stimuli, and calculate the theoretical maximum frequency of physiological state switching within a unit time (e.g., 1 hour) as 6 times (60 divided by 10). Using 6 as the initial benchmark, an environmental disturbance filtering coefficient (e.g., 0.8, excluding minor environmental fluctuations) is introduced. 6 is multiplied by 0.8 to obtain 4.8, which is rounded down to set the preset frequency benchmark to 4. Time periods with a time interval density higher than 4 per hour are marked as high-frequency fluctuation zones. Within this zone, the cumulative number of times the statistical slope sign flips (from positive to negative or vice versa) is calculated. For example, if the slope flips 8 times within one hour, a maximum slope change point density of 8 is generated.

[0079] S203: Call the density of the maximum slope change points, input it into the preset crop growth process classifier for pattern matching, determine that the crop is currently in a critical state of transition from vegetative growth to reproductive growth, extract the extreme value of growth rate acceleration under the state, combine it with historical water stress response data to calculate the resource demand urgency index for the current period, and generate crop time sensitivity features.

[0080] The crop growth process classifier is a neural network model based on a multilayer perceptron. Its internal structure includes: an input layer with 5 neurons, receiving slope density, current accumulated temperature, sunshine duration, soil moisture change rate, and historical growth stage codes; a first hidden layer with 64 neurons, using the ReLU activation function to enhance nonlinear feature extraction; a second hidden layer with 32 neurons, also using the ReLU activation function; and an output layer with 3 neurons, corresponding to three states: vegetative growth, reproductive growth, and critical transition period, outputting a probability distribution using the Softmax function. For example, with an input density of 8, the model outputs a critical transition period probability of 0.85, indicating that the crop is currently in a critical state transitioning from vegetative to reproductive growth. In this state, the extreme value of the growth rate acceleration (e.g., the maximum value of the second derivative) is extracted, and combined with historical water stress response data, the resource demand urgency index for the current period is calculated. Setting the acceleration extreme value to 0.2, historical data shows that the sensitivity coefficient of this variety to water deficit during the critical period is 0.9, thus calculating the resource demand urgency index to 0.18, ultimately generating the crop time sensitivity feature.

[0081] Please see Figure 4 The specific steps for obtaining the exponential dynamic weight parameters are as follows:

[0082] S301: Retrieve irrigation event logs and corresponding crop growth monitoring data for plots from the local agricultural management database over multiple years, calculate the ratio of biomass increment to water consumption per unit time after each irrigation, construct a historical water use efficiency sequence, map the sequence to a closed interval between zero and one using the max-min normalization method, remove outlier noise data in the sequence, and generate historical irrigation response efficiency coefficients.

[0083] The process of removing outlier data from the sequence is as follows: statistically analyze the distribution frequency of all numerical points in the historical water use efficiency sequence, calculate the interquartile range and median of the sequence; multiply the interquartile range by a preset dispersion tolerance coefficient to obtain the fluctuation tolerance value, the dispersion tolerance coefficient is set according to the total sample size of the dataset and exhibits a logarithmic decay relationship; add the median and the fluctuation tolerance value to obtain the upper cutoff threshold, subtract the fluctuation tolerance value from the median to obtain the lower cutoff threshold, and remove values ​​that exceed the range of the upper and lower cutoff thresholds;

[0084] Irrigation event logs and corresponding crop growth monitoring data for the plot over the past 5 years were retrieved from the local agricultural management database. For each historical irrigation event, the ratio of biomass increase (dry weight increase, in grams) to water consumption (transpiration, in liters) within 24 hours after irrigation was calculated. For example, if biomass increased by 50 grams and water consumption was 20 liters after an irrigation, the ratio would be 2.5 grams per liter. A historical water use efficiency sequence containing thousands of data points was constructed. Subsequently, the maximum (e.g., 5.0) and minimum (e.g., 1.0) values ​​in the sequence were found using the min-max normalization method, and all values ​​were mapped to closed intervals between 0 and 1. The process of removing outlier data from the sequence is as follows: The distribution frequency of all numerical points in the historical water use efficiency sequence is statistically analyzed. The interquartile range and median of the sequence are calculated. The dataset contains 1000 samples (a relatively large number). The dispersion tolerance coefficient is set with a logarithmic decay relationship based on the total number of samples. For example, if the coefficient k is set to 3 and the base value is set to 4.5, then 4.5 - 3 = 1.5. The median is set to 0.5, the interquartile range to 0.2, and the fluctuation tolerance value is calculated as 0.2 multiplied by 1.5, which equals 0.3. The upper cutoff threshold is calculated as 0.5 plus 0.3, which equals 0.8, and the lower cutoff threshold is calculated as 0.5 minus 0.3, which equals 0.2. The sequence is traversed, and all values ​​greater than 0.8 or less than 0.2 are removed. The cleaned historical irrigation response efficiency coefficients are generated, as shown in Table 1, which displays a comparison of some data before and after cleaning.

[0085] Table 1 Historical Water Use Efficiency Data Cleaning Comparison Table

[0086] ;

[0087] S302: Calling crop time sensitivity features and historical irrigation response efficiency coefficients, using an exponential function to fit the water and fertilizer demand urgency curve, using time sensitivity features as the power exponential variable that determines the growth rate of the curve, using historical irrigation response efficiency coefficients to adaptively fine-tune the function base to reflect the specific absorption capacity of the plot, establishing the nonlinear transformation logic between input features and scheduling weights, and establishing a sensitivity-driven weight calculation model.

[0088] The process of adaptively fine-tuning the function base using historical irrigation response efficiency coefficients is as follows: First, the ideal water conversion rate in a pre-set standard growth model is obtained as a benchmark anchor point. Second, the deviation of the historical irrigation response efficiency coefficients from the benchmark anchor point is calculated. Third, a linear mapping relationship is constructed with the deviation as the independent variable and the base correction factor as the dependent variable. The slope of the linear mapping relationship is determined by the correlation coefficient between historical yield data and irrigation amount. Fourth, the calculated base correction factor is superimposed on the preset initial base to change the steepness of the exponential function's growth.

[0089] Using crop time sensitivity features (e.g., 0.18) and historical irrigation response efficiency coefficients (e.g., the washed mean of 0.6), an exponential function is used to fit the water and fertilizer demand urgency curve, with the basic form being Y equal to A raised to the power of X. The time sensitivity feature 0.18 is used as the exponential variable X determining the curve's growth rate. The process of adaptively fine-tuning the function base A using the historical irrigation response efficiency coefficient is as follows: The ideal water conversion rate in a pre-set standard growth model is obtained as the baseline anchor point, for example, 0.8. The deviation of the historical response efficiency coefficient 0.6 from the baseline anchor point 0.8 is calculated, i.e., (0.6 minus 0.8) divided by 0.8 equals -0.25. A linear mapping relationship is constructed, with the slope determined by the correlation coefficient between historical yield data and irrigation amount (e.g., if the correlation coefficient is 0.8, then the slope is 0.8). The base correction factor is calculated as -0.25 multiplied by 0.8, which equals -0.2. This factor is superimposed on the preset initial base (e.g., 2.0) to obtain the adjusted base of 1.8. Finally, a sensitivity-driven weight calculation model was established, with the core logic that the weight W equals 1.8 to the power of 0.18. This process establishes the non-linear transformation logic between input features and scheduling weights.

[0090] S303: Input the crop time sensitivity feature values ​​obtained from real-time monitoring into the sensitivity-driven weight calculation model, perform floating-point exponentiation to calculate the original weight values, use the Sigmoid function to limit the saturation of the calculation results to prevent the weight values ​​from overflowing, quantize the calculated values ​​into dimensionless scalar data for task queue sorting, and generate exponential dynamic weight parameters.

[0091] The process of using the Sigmoid function to limit the saturation of the calculation results is as follows: a center offset parameter is set, which is dynamically assigned based on the average value of the historical average scheduling weights; a curve slope parameter is set, which is set inversely based on the reciprocal of the total amount of water and fertilizer resources currently available in the system; the original weight values ​​are shifted using the center offset parameter, and the shifted values ​​are scaled using the curve slope parameter; the processed values ​​are then substituted into the logistic equation for normalization calculation.

[0092] The crop time sensitivity feature value obtained from real-time monitoring (e.g., 0.25) is input into the sensitivity-driven weight calculation model, and a floating-point exponentiation operation is performed, i.e., 1.8 raised to the power of 0.25, resulting in approximately 1.158, which is used as the original weight value. The process of using the Sigmoid function to limit the saturation of the calculation result is as follows: A center offset parameter is set, which is dynamically assigned based on the historical average scheduling weight (e.g., 1.1), i.e., the offset is 1.1. A curve slope parameter is set, which is set inversely based on the reciprocal of the total amount of water and fertilizer resources currently available in the system (e.g., 1000 liters), for example, the slope parameter k equals 10000 divided by 1000, which equals 10. The original weight value 1.158 is shifted using the center offset parameter, i.e., 1.158 minus 1.1 equals 0.058. The shifted value is scaled using the curve slope parameter, i.e., 0.058 multiplied by 10 equals 0.58. Substituting the processed value 0.58 into the logistic equation (i.e., 1 divided by (1 plus e to the power of -0.58)), the result is approximately 0.641. This value 0.641 is then quantized into a dimensionless scalar data for task queue sorting, generating exponential dynamic weight parameters.

[0093] Please see Figure 5 The specific steps for obtaining the task resource coupling index are as follows:

[0094] S401: Retrieve fertilization record data for a specified plot of land over many years from the local agricultural database, remove zero or negative outliers caused by sensor malfunctions, segment the fertilization data according to the crop growth cycle, calculate the degree of deviation of the fertilization amount from the average value in each cycle, and obtain the fertilization standard deviation sequence.

[0095] Fertilization records for the past five years for a designated plot of land are retrieved from the local agricultural database, and outliers (zero or negative values) caused by sensor malfunctions are removed. Fertilization data are segmented and statistically analyzed according to crop growth cycles (e.g., seedling stage, jointing stage, grain-filling stage). During the grain-filling stage, the deviation of each record point's fertilizer application value from the average value for that period is calculated. For example, if the average fertilizer application is 10 kg / mu, and a record shows 12 kg / mu, the deviation is 2. The sum of squared deviations is calculated, divided by the total number of records, and the square root is taken to obtain the standard deviation sequence of fertilizer application rates; for example, a standard deviation of 1.5 kg / mu.

[0096] S402: Based on the numerical distribution range of the standard deviation sequence of fertilizer application, set equally spaced discretized windows, count the number of sequence elements falling into each window to calculate the distribution probability, and perform a weighted integral operation on the distribution probability in combination with the window center value to quantify the clustering characteristics of fertilization behavior in the time dimension and generate a fertilization data fluctuation density index.

[0097] Based on the numerical distribution range of the fertilizer application standard deviation sequence (e.g., 0 to 5), equally spaced discretized windows are set, for example, a window width of 0.5, for a total of 10 windows. The number of sequence elements falling into each window is counted, and the distribution probability is calculated. For example, the probability of falling within the range of 1.0 to 1.5 is 0.3. A weighted integral operation is performed on the distribution probability using the window center value (e.g., 1.25), that is, the center value of each window is multiplied by the probability, to quantify the clustering characteristics of fertilization behavior in the time dimension. The integral result is set to 1.8 to generate a fertilization data fluctuation density index.

[0098] S403: Real-time reading of the CPU load percentage, dynamic random access memory utilization, and remaining bandwidth throughput of the network interface of the cloud platform computing nodes; measurement of end-to-end instruction transmission latency; and retrieval of fertilizer data fluctuation density index using the formula:

[0099] ;

[0100] The computational resource coupling index is used to obtain the task.

[0101] in, This represents the task-resource coupling index. The normalized value representing the fertilization data fluctuation density index is obtained by normalizing the calculated fertilization data fluctuation density index to its maximum value, mapping it to the interval between zero and one. The normalized ratio representing the remaining bandwidth throughput of the network interface is obtained by dividing the remaining available bandwidth of the node in real time by the maximum total bandwidth capacity of the node's physical links, using a network monitoring probe. The normalized value representing the real-time processing load of a cloud computing node is obtained by directly reading the real-time CPU core utilization percentage and converting it into a decimal form between zero and one. The normalized coefficient representing the instruction transmission delay is obtained by dividing the end-to-end communication delay (obtained through ping testing or heartbeat packet detection) by the maximum allowable transmission delay threshold of the system. The normalization constant representing the minimum system response time is obtained by taking the system's inherent hardware interrupt response baseline time and dividing it by the system's maximum allowable transmission delay threshold to maintain dimensional consistency. The normalized value representing the dynamic random access memory utilization rate is obtained by reading the number of used memory pages in the system memory manager and dividing it by the total physical memory capacity.

[0102] The system reads the CPU load percentage, dynamic random access memory (DRAM) utilization, and remaining bandwidth throughput of the network interface of the cloud platform computing nodes in real time. End-to-end instruction transmission latency is measured via ping tests or heartbeat packet detection. To ensure calculation accuracy, the following specific operating parameters are collected in real time: the real-time CPU core utilization percentage of the cloud node is 60% (i.e., 0.6); the percentage of DRAM pages used, as reported by the memory manager, is 70% (i.e., 0.7); and the remaining available bandwidth of the node read by the network probe is 50Mbps (the maximum total bandwidth capacity of the node's physical link is set to 100Mbps). Regarding communication latency, the measured end-to-end latency is 20ms, while the maximum allowable transmission latency threshold is set to 100ms; the system's inherent hardware interrupt response baseline time is also obtained as 5ms. Furthermore, the fertilization data fluctuation density index generated in step S402 is used, with its original calculated value set to 1.8, while the historical maximum baseline value for this index is 3.6. Based on the collected raw data, the normalized values ​​of each parameter are first calculated according to preset rules. Normalized value of fertilization data fluctuation density index The acquisition method involves normalizing the maximum value of the fertilization data fluctuation density index:

[0103] ;

[0104] Normalized ratio of remaining bandwidth throughput of network interface The method for obtaining this information is to divide the node's remaining available bandwidth by the maximum total bandwidth capacity of the node's physical links.

[0105] ;

[0106] Normalized value of real-time processing load of cloud computing nodes The method of obtaining the data is to directly read the real-time CPU core usage percentage and convert it into a decimal form between zero and one.

[0107] ;

[0108] Normalized coefficient of instruction transmission delay The method of obtaining this information is to divide the end-to-end communication delay by the maximum allowable transmission delay threshold of the system.

[0109] ;

[0110] Normalized constant of minimum system response time The method of obtaining this information is to divide the system's inherent hardware interrupt response baseline time by the system's maximum allowable transmission delay threshold.

[0111] ;

[0112] Normalized value of dynamic random access memory utilization The method to obtain this number is to read the number of used memory pages and divide it by the total physical memory capacity.

[0113] ;

[0114] After parameter normalization, the task resource coupling index is obtained by using the fertilization data fluctuation density index and the parameters calculated above, and then by formula calculation. :

[0115] ;

[0116] Substitute the above specific values ​​into the formula to perform the final calculation:

[0117] ;

[0118] The final computational resource coupling index is 2.69.

[0119] Please see Figure 6 The specific steps for obtaining water and fertilizer management instructions are as follows:

[0120] S501: Call the task resource coupling index, perform local sorting of multiple water and fertilizer tasks in the same irrigation area in descending order of index value, calculate the difference of index value between adjacent tasks in the sorted sequence to quantify the priority decay gradient, compare the calculated difference with the preset resource competition threshold to filter out the pairs of tasks to be processed with high conflict probability, and generate a sequence of priority difference between adjacent tasks.

[0121] The specific method for setting the resource contention threshold is as follows: call the historical task scheduling log of the cloud computing node, count the frequency peak of instruction conflicts under full load operation, obtain the maximum physical bandwidth capacity of the current communication channel, calculate the average bit size of a single water and fertilizer task instruction data packet, divide the maximum physical bandwidth capacity by the average bit size to obtain the theoretical concurrency limit, use the frequency peak to perform weighted smoothing on the theoretical concurrency limit, and determine the processed value as the resource contention threshold.

[0122] The task resource coupling index is set to 2.69. Multiple water and fertilizer tasks within the same irrigation area are then locally sorted in descending order of their index values. For example, task A has an index of 2.69, task B has an index of 2.50, and task C has an index of 2.10. The difference in index values ​​between adjacent tasks in the sorted sequence is calculated; for example, the difference between A and B is 0.19, and the difference between B and C is 0.40. The calculated difference is compared with a preset resource contention threshold. The resource contention threshold is set as follows: The historical task scheduling logs of the cloud computing node are retrieved, and the peak frequency of instruction conflicts under full load operation is counted (e.g., 5 times per second). The maximum physical bandwidth capacity of the current communication channel is obtained (e.g., 10 Mbps). The average bit size of a single water and fertilizer task instruction data packet is calculated (e.g., 1 Kb, or 1024 bits). The maximum bandwidth of 10,240,000 bits per second is divided by 1024 bits to obtain the theoretical concurrency limit of 10,000 times per second. The theoretical concurrency limit is weighted and smoothed using five frequency peak values ​​(e.g., by taking the logarithm or a scaling factor, with a weight of 0.00001). A simplified logic is used here: the frequency peak value of 5 is divided by the theoretical limit of 10000, yielding 0.0005. This value is then amplified by a safety factor (e.g., by a factor of 200) to ultimately determine the resource contention threshold as 0.1. Comparison reveals that the difference between A and B (0.19) is greater than the threshold of 0.1, and the difference between B and C (0.40) is also greater than the threshold of 0.1. Therefore, tasks A and B, and B and C are determined to be high-probability conflict pairs, generating a sequence of priority difference values ​​for adjacent tasks.

[0123] S502: Call the exponential dynamic weight parameter and the priority difference sequence of adjacent tasks, retrieve the average historical response lag time and hydraulic stability duration of the associated execution valve from the equipment operation log library, use multiplication to combine the exponential dynamic weight parameter and the aggregate lag time value to obtain the dynamic compensation offset, add the offset to the original plan timestamp to reconstruct the task triggering logic, and generate the execution parameters after time delay deviation compensation.

[0124] The process of combining the exponential dynamic weight parameters with the aggregated lag time value using multiplication to obtain the dynamic compensation offset is as follows: the average historical response lag time is defined as the reference physical delay; the numerical part of the exponential dynamic weight parameters is extracted as the time sensitivity gain; the reference physical delay and the time sensitivity gain are multiplied to obtain the preliminary compensation value; the ratio of the hydraulic stability time to the preset standard fluid stability period is obtained; the nonlinear correction factor is constructed using this ratio; and the preliminary compensation value is multiplied by the nonlinear correction factor to generate the dynamic compensation offset.

[0125] The process involves calling an exponential dynamic weight parameter (e.g., 0.641) and a sequence of priority differences between adjacent tasks (e.g., 0.19). The average historical response lag time (e.g., 2 seconds) and hydraulic stabilization duration (e.g., 5 seconds) of the associated execution valve are retrieved from the equipment operation log database. The process of deriving the dynamic compensation offset using multiplication is as follows: the average historical response lag time of 2 seconds is defined as the baseline physical delay. The numerical part of the exponential dynamic weight parameter, 0.641, is extracted as the time sensitivity gain. The baseline physical delay of 2 seconds is multiplied by the time sensitivity gain of 0.641 to obtain a preliminary compensation value of 1.282 seconds. The ratio of the hydraulic stabilization duration of 5 seconds to the preset standard fluid stabilization period (e.g., 4 seconds), i.e., 1.25, is used to construct a nonlinear correction factor (e.g., taking the square root of the ratio, approximately 1.118). The preliminary compensation value of 1.282 seconds is multiplied by the nonlinear correction factor 1.118 to generate a dynamic compensation offset of approximately 1.433 seconds. The offset of 1.433 seconds is superimposed on the original planned timestamp (e.g., 10:00:00), the task triggering logic is reconstructed, and the execution parameters after time delay deviation compensation are generated (i.e., 10:00:01.433 trigger).

[0126] S503: Based on the execution parameters after time delay deviation compensation, the compensated and corrected flow demand and time axis data are converted into electrical control signals for the opening degree of the solenoid valve and the frequency of the booster pump. The signals are mapped to the register address of the field programmable logic controller, and the address segment and data segment are assembled into a standard instruction package including a cyclic redundancy check code to generate water and fertilizer management instructions.

[0127] The process of assembling the address segment and data segment into a standard instruction package including a cyclic redundancy check (CR) code is as follows: a data frame template is constructed according to a preset industrial fieldbus protocol standard; register addresses are converted into hexadecimal codes and filled into the address field; electrical control signals are converted into binary bit streams and filled into the load field; a modulo-2 division operation is performed on the joint bit sequence of the address field and the load field using a generator polynomial; the remainder is used as the CR code; the CR code is appended to the load field to generate the standard instruction package.

[0128] Based on the execution parameters after time delay deviation compensation, the compensated and corrected flow demand (e.g., 15 liters / minute) and time axis data are converted into electrical control signals. Specifically, the flow demand is mapped to control signals for the solenoid valve opening (e.g., 75% opening corresponds to 7.5V) and the booster pump frequency (e.g., 45Hz). These signals are mapped to the register addresses of the field-programmable logic controller (FPGA), for example, the valve opening is written to address 0x4001, and the pump frequency is written to address 0x4002. The process of assembling the address segment and data segment into a standard instruction packet including a cyclic redundancy check (CRC) code is as follows: a data frame template is constructed according to industrial fieldbus protocol standards such as Modbus RTU. Register address 0x4001 is converted to hexadecimal encoding and filled into the address field. The electrical control signal (the value corresponding to 75%) is converted into a binary bit stream and filled into the payload field. A modulo-2 division operation is performed on the joint bit sequence of the address field and payload field using a generator polynomial (e.g., CRC-16 standard polynomial 0x8005). During the calculation, the polynomial and the data sequence are XORed and shifted to obtain a 16-bit remainder as a cyclic redundancy check code (e.g., 0xA1B2). After appending this check code 0xA1B2 to the payload field, the final water and fertilizer management instruction packet is generated and sent to the execution end via the RS485 interface.

[0129] Please see Figure 7 A cloud-based agricultural water and fertilizer management system, used to execute the aforementioned cloud-based agricultural water and fertilizer management method, includes:

[0130] The chlorophyll monitoring module acquires a continuous sequence of crop leaf images, extracts the mean change of the G component in the RGB channels, uses the Sobel operator to calculate the gradient difference between the leaf edge and the center pixel, and combines the mean change of the G component to calculate the chlorophyll concentration variation rate.

[0131] The time-sensitive analysis module collects the crop leaf area growth rate per unit time, performs correlation analysis between the chlorophyll concentration variation rate and the crop leaf area growth rate, identifies the density of the maximum slope change point, and extracts the crop time-sensitive features.

[0132] The dynamic weight adjustment module calls the crop time sensitivity feature, obtains the historical irrigation response efficiency data of the plot, substitutes the crop time sensitivity feature into the exponential function weight mapping relationship to calculate, and generates exponential dynamic weight parameters.

[0133] The task resource analysis module obtains the standard deviation sequence of fertilizer application amount for the same plot of land over multiple years, discretizes the standard deviation sequence to construct the fertilizer data fluctuation density index, collects the real-time processing load value of cloud computing nodes, and calculates the ratio between the fertilizer data fluctuation density index and the real-time processing load value to generate the task resource coupling index.

[0134] The water and fertilizer management implementation module sorts water and fertilizer tasks based on the task resource coupling index and calculates the priority difference between adjacent tasks. For priority differences less than the preset resource competition threshold, it calls the exponential dynamic weight parameter to calculate the time delay deviation compensation and generate water and fertilizer management instructions.

[0135] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. A and B can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it can also represent an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0136] In this invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of a single item or a plurality of items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be a single item or multiple items.

[0137] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0138] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0139] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0140] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0141] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0142] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0143] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0144] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of protection of the described technical solutions.

Claims

1. A cloud-based agricultural water and fertilizer management method, characterized in that, Includes the following steps: S1: Obtain a continuous sequence of crop leaf images, extract the mean change of the G component in the RGB channels, use the Sobel operator to calculate the gradient difference between the leaf edge and the center pixel, and combine the mean change of the G component to calculate the chlorophyll concentration variation rate. S2: Collect the crop leaf area growth rate per unit time, perform correlation analysis between the chlorophyll concentration variation rate and the crop leaf area growth rate, identify the density of the maximum slope change point, and extract the crop time sensitivity features. The specific steps for obtaining the crop time sensitivity features are as follows: S201: Use a multispectral camera to acquire crop canopy images and calculate vegetation indices. Extract leaf region pixels through image segmentation and convert them into physical area values. Calculate the area increment within a unit time interval to obtain the crop leaf area growth rate. Call the chlorophyll concentration variation rate mentioned above and map the two to a unified dimension interval. Construct a dual-channel signal including morphological expansion and physiological variation. Calculate the covariance matrix of the dual-channel signal using a sliding window and establish a growth rate-related state sequence. S202: For the growth rate associated state sequence, perform first-order difference operation to extract the slope curve, identify the abrupt change moment when the absolute value of the slope exceeds the preset steady-state threshold, calculate the time interval density between consecutive abrupt change moments, mark the time period with density value higher than the preset frequency benchmark as the high-frequency fluctuation area, count the cumulative number of slope sign flips in the area, and generate the maximum slope change point density. S203: Call the density of the maximum slope change points and input it into a preset crop growth process classifier for pattern matching. The crop growth process classifier is a neural network model based on a multilayer perceptron. Its input layer receives the density of the maximum slope change points, the current accumulated temperature, sunshine hours, soil moisture change rate, and historical growth stage codes. Its output layer outputs the probability distribution of the three states corresponding to vegetative growth, reproductive growth, and critical transition period. It determines that the crop is currently in the critical state of transition from vegetative growth to reproductive growth, extracts the extreme value of the growth rate acceleration under the state, and combines it with the sensitivity coefficient of the crop to water deficit in the historical water stress response data. The extreme value of the growth rate acceleration is multiplied by the sensitivity coefficient to calculate the resource demand urgency index for the current period and generate the crop time sensitivity feature. S3: Call the crop time sensitivity feature to obtain the historical irrigation response efficiency data of the plot, substitute the crop time sensitivity feature into the exponential function weight mapping relationship to generate exponential dynamic weight parameters; S4: Obtain the standard deviation sequence of fertilizer application amount for the same plot over multiple years, discretize the standard deviation sequence to construct a fertilizer data fluctuation density index, collect the real-time processing load value of cloud computing nodes, and perform a ratio calculation between the fertilizer data fluctuation density index and the real-time processing load value to generate a task resource coupling index. S5: Based on the task resource coupling index, sort the water and fertilizer tasks and calculate the priority difference between adjacent tasks. For priority differences less than the preset resource competition threshold, call the exponential dynamic weight parameter to calculate the time delay deviation compensation and generate water and fertilizer management instructions.

2. The agricultural water and fertilizer management method based on a cloud platform according to claim 1, characterized in that, The chlorophyll concentration variation rate includes concentration decay trend value, color gradient change rate, and nutrient deficiency warning confidence level. The crop time sensitivity characteristics include critical growth period determination value, rapid growth duration, and fertilizer urgency coefficient. The exponential dynamic weight parameters include real-time scheduling gain value, historical response correction term, and dynamic priority ladder. The task resource coupling index includes computing power supply and demand ratio, task execution urgency, and node throughput matching score. The water and fertilizer management instructions include control valve opening value, fertilizer pump flow rate, and target area execution sequence.

3. The agricultural water and fertilizer management method based on a cloud platform according to claim 1, characterized in that, The specific steps for obtaining the chlorophyll concentration variation rate are as follows: S101: Acquire crop leaf image sequences at consecutive time points through an image sensor, perform RGB color space calculation on a single frame image and separate the green component matrix, calculate the arithmetic mean of gray values ​​of pixels in the effective leaf area within the matrix, perform first-order difference operation on the arithmetic mean of gray values ​​of adjacent time points, quantify the overall drift of leaf color values ​​in the time dimension, and generate a green channel mean difference index. S102: For the crop leaf image sequence, load the Sobel discrete differential operator to perform planar convolution processing, calculate the gradient vectors in the horizontal and vertical directions, synthesize the gradient magnitude matrix of the whole image, divide the leaf into edge region of interest and center region of interest based on pixel coordinate position, calculate the average pixel gradient intensity in the two regions respectively, perform difference calculation on the average gradient intensity of the edge region and the center region and extract the absolute value to generate regional gradient difference magnitude features. S103: Detect the sign of the green channel mean difference index to determine the direction of concentration evolution, construct a dynamic correction coefficient using the regional gradient difference amplitude feature, perform weighted compensation on the difference index magnitude, obtain the timestamp difference at the image acquisition time, divide the compensated difference index value by the timestamp difference, and generate the chlorophyll concentration variation rate.

4. The agricultural water and fertilizer management method based on a cloud platform according to claim 3, characterized in that, The process of dividing the leaf into edge region of interest and center region of interest based on pixel coordinate position specifically involves: extracting the binarization mask of a single frame image in the crop leaf image sequence and calculating the geometric centroid coordinates of the binarization mask; traversing the effective pixels within the binarization mask to calculate the Euclidean distance from each pixel to the geometric centroid coordinates; Set a region division distance threshold, wherein the value of the region division distance threshold is set based on a preset scaling factor of the maximum inscribed circle radius of the binarized mask; The set of pixels whose Euclidean distance is greater than the region segmentation distance threshold is marked as the edge region of interest, and the set of pixels whose Euclidean distance is less than or equal to the region segmentation distance threshold is marked as the center region of interest.

5. The agricultural water and fertilizer management method based on a cloud platform according to claim 1, characterized in that, The specific method for setting the preset steady-state threshold is as follows: collect the historical growth rate sequence of crops under standard growth environment, calculate the first derivative distribution of the historical growth rate sequence, select the confidence interval boundary value of the first derivative distribution as the benchmark, combine the current environmental sensor noise level to perform weighted correction on the benchmark, and set the corrected value as the preset steady-state threshold. The specific method for setting the preset frequency benchmark is as follows: based on the crop's physiological characteristics, the minimum delay time for the crop to respond to external water and fertilizer stimuli is obtained; the maximum frequency of physiological state switching that theoretically occurs per unit time is calculated; the maximum frequency is used as the initial benchmark; the initial benchmark is adjusted by introducing an environmental disturbance filtering coefficient; and the processed value is set as the preset frequency benchmark.

6. The agricultural water and fertilizer management method based on a cloud platform according to claim 1, characterized in that, The specific steps for obtaining the exponential dynamic weight parameters are as follows: S301: Retrieve irrigation event logs and corresponding crop growth monitoring data for plots from the local agricultural management database over multiple years, calculate the ratio of biomass increment to water consumption per unit time after each irrigation, construct a historical water use efficiency sequence, map the sequence to a closed interval between zero and one using the max-min normalization method, remove outlier noise data in the sequence, and generate historical irrigation response efficiency coefficients. S302: An exponential function is used to fit the water and fertilizer demand urgency curve. The time sensitivity feature is used as a power exponential variable that determines the growth rate of the curve. The function base is adjusted using the historical irrigation response efficiency coefficient. The nonlinear transformation logic between the input feature and the scheduling weight is established, and a sensitivity-driven weight calculation model is established. S303: Input the crop time sensitivity feature value obtained from real-time monitoring into the sensitivity-driven weight calculation model, perform floating-point exponentiation to calculate the original weight value, use the Sigmoid function to limit the saturation of the calculation result to prevent the weight value from overflowing, quantize the calculated value into dimensionless scalar data, and generate exponential dynamic weight parameters.

7. The agricultural water and fertilizer management method based on a cloud platform according to claim 1, characterized in that, The specific steps for obtaining the task resource coupling index are as follows: S401: Retrieve fertilization record data for a specified plot of land over multiple years from the local agricultural database, remove outliers, segment the fertilization data according to the crop growth cycle, calculate the degree of deviation of the fertilization amount from the average value in each cycle, and obtain the fertilization standard deviation sequence. S402: Based on the numerical distribution range of the standard deviation sequence of the fertilizer application amount, set equally spaced discretized windows, count the number of sequence elements falling into each window to calculate the distribution probability, and perform a weighted integral operation on the distribution probability in combination with the window center value to quantify the clustering characteristics of fertilization behavior in the time dimension and generate a fertilization data fluctuation density index. S403: Real-time reading of the CPU load percentage, dynamic random access memory occupancy rate, and remaining bandwidth throughput of the network interface of the cloud platform computing node; measurement of end-to-end instruction transmission latency; calling the fertilization data fluctuation density index; and calculation of the task resource coupling index.

8. The agricultural water and fertilizer management method based on a cloud platform according to claim 1, characterized in that, The specific steps for obtaining the water and fertilizer management instructions are as follows: S501: Call the task resource coupling index, perform local sorting of multiple water and fertilizer tasks in the same irrigation area in descending order of index value, calculate the index value difference between adjacent tasks in the sorting sequence to quantify the priority decay gradient, compare the calculated difference with the preset resource competition threshold to filter out the task pairs with high conflict probability, and generate the adjacent task priority difference sequence. S502: Call the exponential dynamic weight parameter and the priority difference sequence of adjacent tasks, retrieve the average historical response lag time and hydraulic stability duration of the associated execution valve from the equipment operation log library, use multiplication to combine the exponential dynamic weight parameter with the aggregate lag time value to obtain the dynamic compensation offset, add the offset to the original plan timestamp to reconstruct the task triggering logic, and generate the execution parameters after time delay deviation compensation. S503: Based on the execution parameters after the time delay deviation compensation, the compensated and corrected flow demand and time axis data are converted into electrical control signals for the opening degree of the solenoid valve and the frequency of the booster pump. The signals are mapped to the register address of the field programmable logic controller. The address segment and data segment are assembled into a standard instruction package including a cyclic redundancy check code to generate water and fertilizer management instructions.

9. A cloud-based agricultural water and fertilizer management system, characterized in that, The system is used to implement the cloud-based agricultural water and fertilizer management method according to any one of claims 1-8, and the system includes: The chlorophyll monitoring module acquires a continuous sequence of crop leaf images, extracts the mean change of the G component in the RGB channels, calculates the gradient difference between the leaf edge and the center pixel using the Sobel operator, and calculates the chlorophyll concentration variation rate by combining the mean change of the G component. The time-sensitive analysis module collects the crop leaf area growth rate per unit time, correlates the chlorophyll concentration variation rate with the crop leaf area growth rate, identifies the density of the maximum slope change point, and extracts the crop time-sensitive features. The dynamic weight adjustment module calls the crop time sensitivity feature, obtains the historical irrigation response efficiency data of the plot, substitutes the crop time sensitivity feature into the exponential function weight mapping relationship to calculate, and generates exponential dynamic weight parameters. The task resource analysis module obtains the standard deviation sequence of fertilizer application amount for the same plot of land over multiple years, discretizes the standard deviation sequence to construct a fertilizer data fluctuation density index, collects the real-time processing load value of cloud computing nodes, and calculates the ratio between the fertilizer data fluctuation density index and the real-time processing load value to generate a task resource coupling index. The water and fertilizer management implementation module sorts water and fertilizer tasks based on the task resource coupling index and calculates the priority difference between adjacent tasks. For priority differences less than the preset resource competition threshold, it calls the exponential dynamic weight parameter to calculate the time delay deviation compensation and generates water and fertilizer management instructions.

Images