A big data-based seedling raising system and method

Abstract

This invention belongs to the field of seedling technology and discloses a seedling system and method based on big data. The system includes the following steps: S1: Real-time collection of health parameters within a unit seedling area using a data acquisition device. This invention, by setting a necessary resource replenishment prediction model, predicts the amount of water replenishment, nutrient solution replenishment, and light enhancement required for a single application. It can dynamically recommend the optimal replenishment plan based on the actual needs of the seedlings. By fitting a power function between the replenishment coefficient and the improvement time, it can estimate the number of days required from intervention to health recovery. The predicted transplanting time for each area to be intervened is compared with the target transplanting date to automatically identify areas at risk of delay. These areas are then sorted by urgency, and resources are dynamically allocated. When resources are limited, priority is calculated based on the current health score to ensure that the areas most in need and most likely to recover receive priority support. Early warning information and seedling progress reports are generated.

Description

Technical Field

[0001] This invention relates to the field of seedling technology, and more specifically, to a seedling system and method based on big data. Background Technology

[0002] With the rapid development of modern agriculture, seedling raising, as the initial stage of crop production, directly affects subsequent growth and yield. Traditional seedling management mainly relies on manual experience, using observation of seedling appearance and simple environmental measurements to determine irrigation, fertilization, and light regulation.

[0003] Traditional seedling cultivation often employs a timed and quantitative water and fertilizer supply model, which cannot be dynamically adjusted according to the actual needs of seedlings. This leads to a waste of water resources and nutrient solutions, and even causes diseases due to over-irrigation. When environmental changes occur or seedlings show abnormal growth, it is often difficult for humans to detect and respond in time, and measures are often taken only when the problem becomes serious, missing the best intervention opportunity. Existing systems can only react to the current state and cannot predict future growth trends and key time nodes (such as the transplanting time), which makes it difficult to formulate precise production plans. This often results in seedlings waiting for the field or the field waiting for seedlings, affecting the overall planting efficiency. Existing technologies monitor environmental parameters such as temperature, humidity, and light in real time through sensors and automatically control equipment based on thresholds. However, these systems are still at the level of single-factor regulation and have failed to couple the environment with the physiological state of crops. Furthermore, they lack deep integration of multi-source data and intelligent decision-making. Summary of the Invention

[0004] To address the problems in the background art, this invention proposes a seedling cultivation system and method based on big data.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a seedling cultivation system based on big data, comprising the following modules: The data acquisition module is used to collect health parameters in real time within the unit's seedling area through acquisition devices; The health assessment module is used to generate a health score for a unit seedling area based on health parameters, and to calculate the rate of change of the health score for a unit seedling area based on the health score. The health grading module is used to classify the health of a unit seedling area based on the health score and the rate of change of the health score, and to take corresponding measures for unit seedling areas of different levels. The necessary resource replenishment prediction module is used to acquire historical necessary resource replenishment data and build a necessary resource replenishment prediction model, and use the necessary resource replenishment prediction model to predict the necessary resource replenishment in the area to be intervened. The improvement time prediction module is used to calculate the replenishment coefficient based on the predicted required resource replenishment amount, and then obtain the expected improvement time based on the replenishment coefficient. The transplanting time prediction module is used to acquire historical transplanting time data and build a transplanting time prediction model. The transplanting time prediction model is used to predict the transplanting time of the area to be intervened, and the scheduling is optimized based on the predicted transplanting time.

[0006] Furthermore, the seedling base is equipped with multiple seedling greenhouses, all containing the same crop. Each seedling greenhouse is divided into multiple seedling areas. The health parameters include the average stem diameter, plant height, number of leaves, leaf area, and relative chlorophyll content of the crop within each seedling area. Using a laser rangefinder or high-precision image recognition technology, the diameter of the stem at the base of the plant is measured at a fixed time each day. The average of three measurements is taken as the diameter value for that day, and the stem diameter is then normalized.

[0007] In the formula, For normalized stem diameter, The measured stem diameter is... and These are the minimum and maximum diameters that this crop variety may have during the seedling stage; Plant height is obtained by scanning the top of the plant using an ultrasonic sensor or lidar, and then the plant height is normalized.

[0008] In the formula, To normalize plant height, This is the measured plant height. This represents the ideal plant height for this variety during the seedling stage. and These are the lower and upper limits of plant height; The plant images were captured using a high-definition camera. A deep learning object detection algorithm was used to identify and count the number of leaves, which was then normalized.

[0009] In the formula, To normalize the number of leaves, This represents the actual number of leaves. This represents the maximum expected number of leaves during the seedling stage of this variety. Leaf contours are extracted using image segmentation techniques, pixel areas are calculated and converted to obtain leaf area, and then the leaf area is normalized.

[0010] In the formula, To normalize the leaf area, This is the measured leaf area. This represents the maximum leaf area during the seedling stage of this variety; The relative chlorophyll content of leaves was obtained by measuring the relative chlorophyll value using a hyperspectral imager, and then normalized.

[0011] In the formula, Normalized relative chlorophyll content, This represents the measured relative chlorophyll content. and These represent the minimum and maximum relative chlorophyll content of this crop variety during the seedling stage.

[0012] Furthermore, the process of generating a health score for a unit seedling area based on health parameters includes: Health score S:

[0013] In the formula, , , , and These are weighting coefficients, obtained through training based on historical data; Gradient descent is used to train the weight coefficients, with the objective function being to maximize the fit between the health score and the actual growth status of the crop during the seedling stage, and the mean squared error being used as the metric. The actual loss function is set with a learning rate of 0.001. The iteration stops when the loss function value is ≤0.01, thus obtaining the optimal combination of weight coefficients. The process of calculating the rate of change of health score per unit seedling area based on health score includes: Set a monitoring cycle. At the end of each monitoring cycle, calculate the health score of each seedling area. For each seedling area, save the historical health score data for the most recent few days. Plot the time sequence on the x-axis and the corresponding health score on the y-axis. Perform a univariate linear regression on these data points to obtain a straight line that reflects the change of the score over time. The slope of this line is the rate of change of the health score within that time period.

[0014] Furthermore, the process of classifying the health of a unit seedling area based on health scores and the rate of change of health scores includes: Based on historical health score data, set appropriate first health score threshold, second health score threshold and third health score threshold. The historical health score data refers to the data set of health scores of the previous production line. Compare the current health score of the production line with the health score threshold. Based on historical health score change rate data, set appropriate first health score change rate threshold, second health score change rate threshold and third health score change rate threshold. The historical health score change rate data refers to the data set of health score change rates of the previous production line. Compare the current health score change rate of the production line with the health score change rate threshold. If the current health score of a unit's seedling area is greater than or equal to the first health score threshold and the current health score change rate is greater than or equal to the first health score change rate threshold, it is judged as an excellent area and no intervention is required; only routine environmental monitoring is needed. If the second health score threshold is less than or equal to the current health score of the unit seedling area, which is less than the first health score threshold, and the current health score change rate is greater than or equal to the second health score change rate threshold, it is judged as a good area. The change trend should be monitored, and environmental parameters can be adjusted appropriately to promote its development towards excellence. If the third health score threshold is less than or equal to the current health score of the unit seedling area which is less than the second health score threshold, or the current health score change rate is less than the second health score change rate threshold, it is judged as a sub-healthy area, and the cause needs to be analyzed and targeted measures taken. If the current health score of a unit's seedling area is less than the third health score threshold, or the current health score change rate is less than the third health score change rate threshold, it is determined to be an area requiring intervention, and the necessary resource supplementation intervention process must be carried out immediately.

[0015] Furthermore, the process of acquiring historical data on necessary resource replenishment and constructing a necessary resource replenishment prediction model, and then using this model to predict the necessary resource replenishment amount for the area to be intervened in, includes: The necessary resource replenishment intervention process refers to replenishing the area to be intervened with necessary resources. Necessary resources refer to water resources, nutrient solution resources and light resources. The amount of necessary resource replenishment refers to the amount of water replenishment, the amount of nutrient solution replenishment, and the amount of light intensity increase in a single supply. Factors affecting the amount of necessary resource replenishment include: the current health score of the area to be intervened, the rate of change of the current health score, the current frequency of resource replenishment, the current single water supply, the current single nutrient solution volume, the current light intensity, the current substrate moisture content, and the current nutrient solution pH value; The current replenishment frequency refers to the average number of times the area to be intervened has been irrigated or fertilized in the past week, which can be obtained from the execution module log; The current single water supply volume refers to the water supply volume of the most recent irrigation; The current amount of nutrient solution used in a single application refers to the amount of nutrient solution used in the most recent fertilization. Current light intensity refers to photosynthetically active radiation measured in real time; The current substrate moisture content is obtained by measuring the volumetric moisture content using a soil moisture sensor. The current pH value of the nutrient solution is obtained by pH measurement. Acquire historical data on the amount of necessary resources to be replenished for a single area to be intervened in different monitoring periods. The historical data on the amount of necessary resources to be replenished includes the current health score, the rate of change of the current health score, the current frequency of resource replenishment, the current single water supply, the current single nutrient solution volume, the current light intensity, the current substrate moisture content, the current nutrient solution pH value, and the historical amount of necessary resources to be replenished for the single area to be intervened in the monitoring period. Based on the current health score, current health score change rate, current replenishment frequency, current single water supply, current single nutrient solution volume, current light intensity, current substrate moisture content, current nutrient solution pH value, and corresponding historical necessary resource replenishment data of the corresponding areas to be intervened, a necessary resource replenishment prediction set is generated and divided into the first training set and the first test set. The 8-dimensional static feature vectors are sorted according to their influence on crop growth to construct a one-dimensional sequence of length 8, giving the features ordered growth correlation characteristics, so that the one-dimensional convolution can extract the synergistic influence between features. The convolution kernel size 3 is used to capture the local coupling relationship between three adjacent features. The pooling window 2 uses mean pooling to retain feature coupling information and avoid loss of effective information. The first convolutional neural network is constructed. The first convolutional neural network is a one-dimensional convolutional neural network containing two convolutional layers, two pooling layers, a flattening layer, and two fully connected layers. A Dropout layer is set after the first fully connected layer. The network is trained using the Adam optimizer, the loss function is mean squared error, and the training termination is determined by early stopping. The current health score, current health score change rate, current resource replenishment frequency, current single water supply, current single nutrient solution volume, current light intensity, current substrate moisture content, and current nutrient solution pH value from the different historical necessary resource replenishment data in the first training set are used as the input data of the first convolutional neural network, and the corresponding historical necessary resource replenishment amounts in the first training set are used as the output data of the first convolutional neural network. The first convolutional neural network is trained to obtain the first initial convolutional neural network. The first initial convolutional neural network is validated using the first test set. The first initial convolutional neural network, whose output is less than or equal to the preset first test error threshold, is used as the required resource replenishment prediction model. The current health score, current health score change rate, current resource replenishment frequency, current single water supply volume, current single nutrient solution volume, current light intensity, current substrate moisture content, and current nutrient solution pH value of each area to be intervened are input into the required resource replenishment prediction model to obtain the predicted required resource replenishment amount for each area to be intervened, namely the predicted single water supply replenishment amount, the predicted single nutrient solution replenishment amount, and the predicted light intensity increase amount.

[0016] Furthermore, the process of calculating the replenishment coefficient based on the predicted required resource replenishment includes: In order to comprehensively assess the urgency of resource replenishment, a replenishment coefficient K is obtained based on the predicted single water supply replenishment amount, the predicted single nutrient solution replenishment amount, and the predicted increase in light intensity. The predicted single water supply replenishment amount is normalized:

[0017] In the formula, To normalize the predicted single water supply replenishment amount, This is the actual predicted single water supply replenishment amount. The largest single water supply replenishment volume that has appeared in historical data; The predicted amount of nutrient solution to be replenished at one time was normalized:

[0018] In the formula, To normalize the predicted amount of nutrient solution to be added at one time, This is to predict the actual amount of nutrient solution to be supplemented per feeding. The largest single nutrient solution replenishment volume recorded in historical data; The predicted increase in light intensity was normalized:

[0019] In the formula, To normalize the predicted increase in light intensity, This represents the actual predicted increase in light intensity. This represents the maximum increase in light intensity observed in historical data. Supplementary coefficient K:

[0020] In the formula, , and The weighting coefficient can be determined based on the sensitivity of resources to crop growth; The process of obtaining the expected improvement time based on the supplementary coefficient includes: During operation, historical data of necessary resource interventions in the area to be intervened are continuously recorded. Each historical record includes the supplementary coefficient at the time of intervention and the actual number of days from the start of the intervention to the improvement of the health score of the area to be intervened to the target value, i.e. the improvement time. A large number of historical data points are plotted on a coordinate system with the supplementary coefficient as the x-axis and the improvement time as the y-axis. Through regression analysis, it is found that there is a power function relationship between the two. The least squares method is used to fit the data points to obtain the empirical formula for the crop variety at the current growth stage. The power function expression is T= T represents the expected improvement time (days), K is the supplementary coefficient, and a, b, and c are the fitting constants for crop growth stages. The objective is to minimize the sum of squared residuals. The initial parameter ranges are set as a∈(0, 10], b∈(-5, 0), and c∈(0, 5]. The convergence condition is that the sum of squared residuals S≤0.05 or 500 iterations. After fitting, the coefficient of determination is used. Validate if the value is ≥0.85; if not, expand the parameter range and refit. When intervention is needed in a certain area, first calculate its current supplementary coefficient, and then substitute the supplementary coefficient into the fitted power function curve to obtain the expected improvement time.

[0021] Furthermore, the process of acquiring historical transplanting time data and constructing a transplanting time prediction model, and then using this model to predict the transplanting time for the area to be intervened, includes: Crops in the seedling greenhouse need to reach the transplantable standard before the specified time. Crops in other levels of unit seedling areas will definitely grow faster than those in the areas to be intervened. Therefore, it is only necessary to consider the estimated transplanting time of all areas to be intervened. The transplanting time refers to the time required for crops to reach the transplantable standard from the present. Factors influencing the transplanting time in the area to be intervened include: the current health score of the area to be intervened, the expected improvement time, the current growth stage, the historical average temperature for the same period, the historical average light intensity for the same period, and the current photosynthetic rate; The current growth stage is represented by a numerical code; Historical average temperature for the same period refers to the average temperature over the past week. Historical average sunshine duration refers to the average sunshine intensity over the past week. The current photosynthetic rate was obtained by actual measurement using a photosynthesis instrument; Historical transplanting time data of a single area to be intervened in different monitoring periods are obtained. The historical transplanting time data includes the current health score, expected improvement time, current growth stage, historical average temperature, historical average light intensity, current photosynthetic rate, and historical transplanting time of the single area to be intervened. Based on the current health score, expected improvement time, current growth stage, historical average temperature, historical average light intensity and current photosynthetic rate of the corresponding area to be intervened in different historical transplanting time data, a transplanting time prediction set is generated and divided into a second training set and a second test set. The 6-dimensional static feature vectors are sorted according to the weights determined by crop growth to construct a one-dimensional sequence of length 6, giving the features ordered growth-related attributes. The one-dimensional convolutional kernel size of 3 is used to capture the local synergistic effect of three adjacent features. The pooling window 2 uses mean pooling to retain feature coupling information, taking into account both feature dimension compression and information integrity. A second convolutional neural network is constructed. The second convolutional neural network is a one-dimensional convolutional neural network containing two convolutional layers, two pooling layers, a flattening layer, and two fully connected layers. A Dropout layer is set after the first fully connected layer. The network is trained using the Adam optimizer, the loss function is mean squared error, and the training termination is determined by early stopping. The current health score, expected improvement time, current growth stage, historical average temperature, historical average light intensity, and current photosynthetic rate from different historical transplanting time data in the second training set are used as input data for the second convolutional neural network, and the corresponding historical transplanting time in the second training set is used as output data for the second convolutional neural network. The second convolutional neural network is trained to obtain the second initial convolutional neural network. The second initial convolutional neural network is validated using the second test set. The second initial convolutional neural network that outputs a second test error threshold less than or equal to the preset second test error threshold is used as the transplanting time prediction model. The current health score, expected improvement time, current growth stage, historical average temperature, historical average light intensity, and current photosynthetic rate of each area to be intervened are input into the transplanting time prediction model to obtain the predicted transplanting time for each area to be intervened.

[0022] Furthermore, the process of optimizing scheduling based on the predicted transplanting time includes: Based on the transplanting time prediction model, the predicted transplanting time for each area to be intervened is obtained. Combined with the target transplanting date in the production plan, a unified target transplanting date is set for the seedling greenhouse. The remaining available days between the current date and the target transplanting date are calculated. For the area to be intervened, its predicted transplanting time is compared with the remaining available days. If the predicted transplanting time is less than or equal to the remaining available days, it indicates that the area is expected to reach the transplanting standard on schedule and no extra attention is required. If the predicted transplanting time is greater than the remaining available days, it is determined that the area has a risk of delay and resources need to be prioritized to accelerate its growth. Calculate the number of days beyond the expected transplanting time for each area at risk of delay, i.e., the predicted transplanting time minus the remaining available days. Sort these areas in descending order of the number of days beyond the expected transplanting time. The larger the number of days beyond the expected transplanting time, the more severe the delay, and the more likely the area should receive additional water, nutrient solution, and light resources. When multiple areas requiring intervention need resources simultaneously and resources are limited, a comprehensive priority is calculated by combining the current health scores of each area. The current health score is multiplied by the reciprocal of the predicted transplanting time, and the higher the value, the higher the priority, ensuring that areas with a good health foundation and the potential for rapid recovery receive priority support. Set a safety margin, such as three days. When the number of days exceeding the safety margin in a certain area to be intervened exceeds the number of days, an early warning notification will be automatically sent to the administrator, indicating that the area to be intervened may be seriously delayed and suggesting that backup seedlings be prepared in advance or the transplanting plan be adjusted. The predicted transplanting time and target date of all areas to be intervened are compared and a seedling progress report is generated to provide a basis for decision-making on the adjustment of subsequent batch sowing time and the optimization of environmental parameters.

[0023] A seedling cultivation method based on big data includes the following steps: S1: Collect health parameters in real time within the unit's seedling area using data acquisition equipment, and then normalize the health parameters. S2: Generate a health score for each seedling area based on the processed health parameters, set a monitoring cycle, and calculate the rate of change of the health score for each monitoring cycle. S3: The health of the unit seedling area is classified into four levels: excellent, good, sub-healthy, and needing intervention, based on the health score and the rate of change of the health score. Corresponding measures are taken for the unit seedling area at different levels. S4: Establish a prediction model for the necessary resource replenishment in the area to be intervened, which is used to predict the necessary resource replenishment in the area to be intervened, namely, predict the amount of water replenishment per supply, predict the amount of nutrient solution replenishment per supply, and predict the increase in light intensity. S5: The replenishment coefficient is obtained based on the predicted single water supply replenishment amount, the predicted single nutrient solution replenishment amount, and the predicted increase in light intensity. The expected improvement time is then obtained based on the replenishment coefficient. S6: Construct a transplanting time prediction model for the area to be intervened based on the expected improvement time, and schedule resources according to the predicted transplanting time.

[0024] The technical effects and advantages of the seedling cultivation system and method based on big data of the present invention are as follows: (1) By setting up a prediction model for the amount of necessary resource replenishment, the amount of water replenishment, the amount of nutrient solution replenishment and the amount of light enhancement can be predicted. The optimal replenishment plan can be dynamically recommended according to the actual needs of the seedlings. By fitting the power function of the replenishment coefficient and the improvement time, the number of days required from intervention to health recovery can be estimated. The predicted transplanting time of each area to be intervened is compared with the target transplanting date. Areas with the risk of delay are automatically identified and sorted by urgency. Resources are dynamically allocated. When resources are limited, priority is calculated in combination with the current health score to ensure that the areas that need the most and have the best chance of recovery receive priority support. Early warning information and seedling progress reports are generated to guide administrators to adjust the sowing plan or prepare backup seedlings in advance.

[0025] (2) By setting a health score, the growth vitality, nutritional status and potential stress of seedlings can be fully reflected, overcoming the one-sidedness of traditional single-indicator assessment. By calculating the rate of change of the health score, the acceleration or deceleration of the growth trend can be accurately captured. Based on the health score and the rate of change of the health score, the seedling area is divided into four levels: excellent, good, sub-healthy and needing intervention. Differentiated management measures are taken for different levels. For excellent areas, only routine monitoring is maintained to avoid excessive intervention. For good areas, the environment is appropriately fine-tuned to promote its transformation to excellent. For sub-healthy areas, the causes are analyzed and targeted adjustments are initiated. For areas needing intervention, the precise resource replenishment process is immediately initiated. This dynamic grading mechanism realizes the allocation of resources on demand, concentrating limited water, fertilizer and light on the areas that need them most, avoiding the waste and inefficiency caused by traditional uniform management. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the system of the present invention; Figure 2 This is a schematic diagram of the method flow of the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0028] Reference Figure 1 A seedling cultivation system based on big data includes the following modules: The data acquisition module is used to collect health parameters in real time within the unit's seedling area through acquisition devices; The health assessment module is used to generate a health score for a unit seedling area based on health parameters, and to calculate the rate of change of the health score for a unit seedling area based on the health score. The health grading module is used to classify the health of a unit seedling area based on the health score and the rate of change of the health score, and to take corresponding measures for unit seedling areas of different levels. The necessary resource replenishment prediction module is used to acquire historical necessary resource replenishment data and build a necessary resource replenishment prediction model, and use the necessary resource replenishment prediction model to predict the necessary resource replenishment in the area to be intervened. The improvement time prediction module is used to calculate the replenishment coefficient based on the predicted required resource replenishment amount, and then obtain the expected improvement time based on the replenishment coefficient. The transplanting time prediction module is used to acquire historical transplanting time data and build a transplanting time prediction model. The transplanting time prediction model is used to predict the transplanting time of the area to be intervened, and the scheduling is optimized based on the predicted transplanting time.

[0029] It should be further explained that, in the specific implementation process, the seedling base is equipped with multiple seedling greenhouses, all of which contain the same crop. The seedling greenhouses are divided into multiple unit seedling areas. The health parameters include the average stem diameter, plant height, number of leaves, leaf area and relative chlorophyll content of the crop in the unit seedling area. Using a laser rangefinder or high-precision image recognition technology, the diameter of the stem at the base of the plant is measured at a fixed time each day. The average of three measurements is taken as the diameter value for that day, and the stem diameter is then normalized.

[0030] In the formula, For normalized stem diameter, The measured stem diameter is... and These are the minimum and maximum diameters that this crop variety may have during the seedling stage, specifically 2 mm and 12 mm, respectively. Plant height is obtained by scanning the top of the plant using an ultrasonic sensor or lidar, and then the plant height is normalized.

[0031] In the formula, To normalize plant height, This is the measured plant height. The ideal plant height for this variety during the seedling stage is 13cm. and These are the lower and upper limits for plant height, specifically 10cm and 20cm respectively. The plant image is captured using a high-definition camera. The number of leaves is then identified and counted using a deep learning object detection algorithm (such as YOLO). The leaf count is then normalized.

[0032] In the formula, To normalize the number of leaves, This represents the actual number of leaves. This is the maximum expected number of leaves for this variety during the seedling stage, specifically 12 leaves; Leaf contours are extracted using image segmentation techniques, pixel areas are calculated and converted to obtain leaf area, and then the leaf area is normalized.

[0033] In the formula, To normalize the leaf area, This is the measured leaf area. This is the maximum leaf area of ​​this variety during the seedling stage, specifically 200 square centimeters; The relative chlorophyll content of leaves was obtained by measuring the relative chlorophyll value using a hyperspectral imager, and then normalized.

[0034] In the formula, Normalized relative chlorophyll content, This represents the measured relative chlorophyll content. and These represent the minimum and maximum relative chlorophyll content of this crop variety during the seedling stage, specifically 20 and 60, respectively.

[0035] It should be further explained that, in the specific implementation process, the process of generating a health score for each seedling area based on health parameters includes: Health score S:

[0036] In the formula, , , , and These are weighting coefficients, obtained from training on historical data, and set to 0.3, 0.15, 0.15, 0.15, and 0.25 respectively. The weight coefficients are trained using gradient descent. The core objective of the training is to maximize the fit between the health score and the actual growth status of the crop during the seedling stage, with the mean squared error as the loss function. The formula is as follows: Si is the calculated health score, which is actually the actual growth status score of the crop in the corresponding unit seedling area. n is the historical data sample size. The training learning rate is set to 0.001. The weight coefficients are continuously updated during the iteration process until the loss function value is ≤0.01 and the iteration stops, thus obtaining the optimal weight coefficient combination for the seedling period of this crop variety. If the health parameters of a certain seedling area are: It is 8mm. It is 13cm. 8 pieces It is 120 square centimeters. If the score is 40, then the health score S of the seedling area of ​​this unit is 0.6155; The process of calculating the rate of change of health score per unit seedling area based on health score includes: Set a monitoring cycle. At the end of each monitoring cycle, calculate the health score of each seedling area. For each seedling area, save the historical health score data for the most recent few days. Plot the time sequence on the x-axis and the corresponding health score on the y-axis. Perform a univariate linear regression on these data points to obtain a straight line that reflects the change of the score over time. The slope of this line is the rate of change of the health score within that time period.

[0037] It should be further explained that, in the specific implementation process, the process of classifying the health of a unit seedling area based on the health score and the rate of change of the health score includes: Based on historical health score data, set appropriate first health score threshold, second health score threshold and third health score threshold. The historical health score data refers to the data set of health scores of the previous production line. Compare the current health score of the production line with the health score threshold. Based on historical health score change rate data, set appropriate first health score change rate threshold, second health score change rate threshold and third health score change rate threshold. The historical health score change rate data refers to the data set of health score change rates of the previous production line. Compare the current health score change rate of the production line with the health score change rate threshold. If the current health score of a unit's seedling area is greater than or equal to the first health score threshold and the current health score change rate is greater than or equal to the first health score change rate threshold, it is judged as an excellent area and no intervention is required; only routine environmental monitoring is needed. If the second health score threshold is less than or equal to the current health score of the unit seedling area, which is less than the first health score threshold, and the current health score change rate is greater than or equal to the second health score change rate threshold, it is judged as a good area. The change trend should be monitored, and environmental parameters can be adjusted appropriately to promote its development towards excellence. If the third health score threshold is less than or equal to the current health score of the unit seedling area which is less than the second health score threshold, or the current health score change rate is less than the second health score change rate threshold, it is judged as a sub-healthy area, and the cause needs to be analyzed and targeted measures taken. If the current health score of a unit's seedling area is less than the third health score threshold, or the current health score change rate is less than the third health score change rate threshold, it is determined to be an area requiring intervention, and the necessary resource supplementation intervention process must be carried out immediately.

[0038] It should be further explained that, in the specific implementation process, the process of obtaining historical data on the amount of necessary resource replenishment and constructing a prediction model for the amount of necessary resource replenishment, and then using this prediction model to predict the amount of necessary resource replenishment for the area to be intervened in, includes: The necessary resource replenishment intervention process refers to replenishing the area to be intervened with necessary resources. Necessary resources refer to water resources, nutrient solution resources and light resources. The amount of necessary resource replenishment refers to the amount of water replenishment, the amount of nutrient solution replenishment, and the amount of light intensity increase in a single supply. Factors affecting the amount of necessary resource replenishment include: the current health score of the area to be intervened, the rate of change of the current health score, the current frequency of resource replenishment, the current single water supply, the current single nutrient solution volume, the current light intensity, the current substrate moisture content, and the current nutrient solution pH value; The current replenishment frequency refers to the average number of times the area to be intervened has been irrigated or fertilized in the past week, which can be obtained from the execution module log; The current single water supply volume refers to the water supply volume of the most recent irrigation; The current amount of nutrient solution used in a single application refers to the amount of nutrient solution used in the most recent fertilization. Current light intensity refers to photosynthetically active radiation measured in real time; The current substrate moisture content is obtained by measuring the volumetric moisture content using a soil moisture sensor. The current pH value of the nutrient solution is obtained by pH measurement. Acquire historical data on the amount of necessary resources to be replenished for a single area to be intervened in different monitoring periods. The historical data on the amount of necessary resources to be replenished includes the current health score, the rate of change of the current health score, the current frequency of resource replenishment, the current single water supply, the current single nutrient solution volume, the current light intensity, the current substrate moisture content, the current nutrient solution pH value, and the historical amount of necessary resources to be replenished for the single area to be intervened in the monitoring period. Based on the current health score, current health score change rate, current replenishment frequency, current single water supply, current single nutrient solution volume, current light intensity, current substrate moisture content, current nutrient solution pH value, and corresponding historical necessary resource replenishment data of the corresponding areas to be intervened, a necessary resource replenishment prediction set is generated and divided into the first training set and the first test set. Construct a first convolutional neural network, taking the current health score, current health score change rate, current resource replenishment frequency, current single water supply, current single nutrient solution volume, current light intensity, current substrate water content, and current nutrient solution pH value from different historical necessary resource replenishment data in the first training set as the input data of the first convolutional neural network, and taking the corresponding historical necessary resource replenishment amount in the first training set as the output data of the first convolutional neural network. The first convolutional neural network adopts a convolutional structure suitable for one-dimensional temporal or feature data. The specific design is as follows: the 8-dimensional features are ordered according to their impact on growth as follows: current health score → health score change rate → substrate moisture content → single water supply → nutrient solution pH value → single nutrient solution volume → light intensity → frequency of resource replenishment, forming a one-dimensional sequence of length 8. The original max pooling is replaced with mean pooling. Dimensional compression is achieved by calculating the average feature value within the pooling window, effectively preserving the coupled and synergistic information of adjacent features such as substrate moisture content and single water supply. This allows the one-dimensional convolution to possess a clear physical meaning for seedling growth in static feature processing. The input layer receives an 8-dimensional feature vector. The first convolutional layer contains 64 filters with a kernel size of 3 and a stride of 1, using ReLU activation and "same" padding to extract local features. It is then connected to the first max pooling layer with a pooling window size of 2 to reduce feature dimensionality. The second convolutional layer contains 32 filters, also using a kernel size of 3, a stride of 1, and ReLU activation to further abstract features. The second pooling layer then performs... Max pooling (window 2) is performed, followed by flattening the feature map into a one-dimensional vector, which is then input into a fully connected layer with 128 neurons using ReLU activation. A Dropout layer is added after this layer, randomly discarding 30% of the neurons to prevent overfitting. This is followed by a fully connected layer with 64 neurons, also using ReLU activation and Dropout (ratio 0.3). The output layer has 3 neurons, corresponding to three predicted values, using a linear activation function. The model is compiled using the Adam optimizer with an initial learning rate of 0.001 and a mean squared error loss function. An early stopping strategy is used during training: the training set is divided into 80% training and 20% validation. The validation set loss is monitored, and training is terminated early if the loss does not decrease for 10 consecutive epochs. The model with the lowest validation set loss is saved. The preset first test error threshold is a validation set mean squared error of less than 0.01. Models that reach this threshold are used as the final required resource supplementation prediction model. To enhance generalization ability, L2 regularization is applied to all fully connected layers with a regularization coefficient of 0.001. The first convolutional neural network is trained to obtain the first initial convolutional neural network. The first initial convolutional neural network is validated using the first test set. The first initial convolutional neural network, whose output is less than or equal to the preset first test error threshold, is used as the required resource replenishment prediction model. The current health score, current health score change rate, current resource replenishment frequency, current single water supply volume, current single nutrient solution volume, current light intensity, current substrate moisture content, and current nutrient solution pH value of each area to be intervened are input into the required resource replenishment prediction model to obtain the predicted required resource replenishment for each area to be intervened, namely the predicted single water supply replenishment, the predicted single nutrient solution replenishment, and the predicted light intensity increase. In an embodiment of the present invention, the predicted required resource replenishment amount for all areas to be intervened is obtained through a required resource replenishment amount prediction model. The predicted required resource replenishment amount is related to the current health score, the current rate of change of the current health score, the current frequency of resource replenishment, the current single water supply amount, the current single nutrient solution amount, the current light intensity, the current substrate moisture content, and the current nutrient solution pH value. The current health score directly affects the amount of necessary resource replenishment to be predicted. The higher the current health score, the worse the health status of crops in the area to be intervened, and the more necessary resource replenishment to be predicted. Therefore, the current health score is positively correlated with the amount of necessary resource replenishment. The current rate of change in health scores directly affects the amount of resources required for prediction. The lower the rate of change in health scores, the more urgent the trend of resource reversal, and the more resources are predicted to be required for prediction. Therefore, the current rate of change in health scores is negatively correlated with the amount of resources required for prediction. Currently, replenishing resources too infrequently may lead to resource shortages, while replenishing them too frequently may result in waste. The current single water supply volume reflects the current water management habits. If the current water supply volume is already high but the health score is still low, it may be necessary to increase the nutrient solution rather than simply adding water. The current amount of nutrient solution per feeding reflects the current nutrient solution management habits. If the current amount of nutrient solution is already high but the health score is still low, it may be necessary to increase water rather than simply adding nutrient solution. The strength or weakness of the current light intensity directly affects the amount of resources required for prediction. The weaker the current light intensity, the greater the required resources are predicted. Therefore, the current light intensity and the required resources are negatively correlated. If the current substrate moisture content is too low, water needs to be added; if it is too high, it may inhibit root respiration. If the pH value of the nutrient solution deviates from the suitable range (e.g., 5.5~6.5), the nutrient availability will be reduced.

[0039] It should be further explained that, in the specific implementation process, the process of calculating the replenishment coefficient based on the predicted required resource replenishment amount includes: In order to comprehensively assess the urgency of resource replenishment, a replenishment coefficient K is obtained based on the predicted single water supply replenishment amount, the predicted single nutrient solution replenishment amount, and the predicted increase in light intensity. The predicted single water supply replenishment amount is normalized:

[0040] In the formula, To normalize the predicted single water supply replenishment amount, This is the actual predicted single water supply replenishment amount. The maximum single water supply replenishment volume that has appeared in historical data is 5 liters; The predicted amount of nutrient solution to be replenished at one time was normalized:

[0041] In the formula, To normalize the predicted amount of nutrient solution to be added at one time, This is to predict the actual amount of nutrient solution to be supplemented per feeding. The maximum single nutrient solution replenishment volume recorded in historical data is 3 liters; The predicted increase in light intensity was normalized:

[0042] In the formula, To normalize the predicted increase in light intensity, This represents the actual predicted increase in light intensity. This represents the largest increase in light intensity observed in historical data, specifically 200. ; Supplementary coefficient K:

[0043] In the formula, , and The weighting coefficients can be determined based on the sensitivity of resources to crop growth. For example, the contribution ratios of water, nutrients, and light to the improvement of health scores can be determined through experiments and set to 0.4, 0.4, and 0.2, respectively. like It is 2. It is 1.5. If the value is 80, then the supplementary coefficient K is 0.44; The process of obtaining the expected improvement time based on the supplementary coefficient includes: During operation, historical data on each necessary resource intervention in the area to be intervened is continuously recorded. Each historical record includes the supplementary coefficient at the time of intervention and the actual number of days from the start of the intervention to the improvement of the health score of the area to be intervened to the target value (e.g., 0.6), i.e. the improvement time. A large number of historical data points are plotted on a coordinate system with the supplementary coefficient as the x-axis and the improvement time as the y-axis. Through regression analysis, it is found that there is a power function relationship between the two. The least squares method is used to fit the data points to obtain the empirical formula for the crop variety at the current growth stage. Power function expression: T= T is the expected improvement time (days), K is the supplementary coefficient, a is the amplitude coefficient, b is the negative power exponent (the larger K is, the shorter T is), c is the basic correction value, and a, b, and c are the specific fitting constants for the current growth stage of the crop.

[0044] Least squares fitting: Using K as the independent variable and the actual improvement time as the dependent variable, the objective is to minimize the sum of squared residuals. Initial parameters are a∈(0, 10], b∈(-5, 0), c∈(0, 5], and the iteration stops at S≤0.05 or after 500 iterations. A value ≥0.85 is used to verify the fitting effect; if the value is not met, the parameter range is expanded and the fitting is refitted.

[0045] Parameter update: As the crop enters a new growth stage, data is re-collected and parameters a, b, and c are updated through fitting. When intervention is needed for a certain area, the current supplement coefficient is first calculated. Then, the supplement coefficient is substituted into the fitted power function curve to obtain the expected improvement time. For example, if the fitted curve shows that the larger the supplement coefficient, the shorter the improvement time, then the number of days required to raise the health score to the target value can be estimated using the current supplement coefficient.

[0046] It should be further explained that, in the specific implementation process, the process of acquiring historical transplanting time data and constructing a transplanting time prediction model, and then using the transplanting time prediction model to predict the transplanting time of the area to be intervened, includes: Crops in the seedling greenhouse need to reach the transplantable standard before the specified time. Crops in other levels of unit seedling areas will definitely grow faster than those in the areas to be intervened. Therefore, it is only necessary to consider the estimated transplanting time of all areas to be intervened. The transplanting time refers to the time required for crops to reach the transplantable standard from the present. Factors influencing the transplanting time in the area to be intervened include: the current health score of the area to be intervened, the expected improvement time, the current growth stage, the historical average temperature for the same period, the historical average light intensity for the same period, and the current photosynthetic rate; The current growth stage is represented by a numerical code (e.g., 1 = cotyledon stage, 2 = true leaf stage, 3 = rapid growth stage). Historical average temperature for the same period refers to the average temperature over the past week. Historical average sunshine duration refers to the average sunshine intensity over the past week. The current photosynthetic rate was obtained by actual measurement using a photosynthesis instrument; Historical transplanting time data of a single area to be intervened in different monitoring periods are obtained. The historical transplanting time data includes the current health score, expected improvement time, current growth stage, historical average temperature, historical average light intensity, current photosynthetic rate, and historical transplanting time of the single area to be intervened. Based on the current health score, expected improvement time, current growth stage, historical average temperature, historical average light intensity and current photosynthetic rate of the corresponding area to be intervened in different historical transplanting time data, a transplanting time prediction set is generated and divided into a second training set and a second test set. A second convolutional neural network is constructed. The current health score, expected improvement time, current growth stage, historical average temperature, historical average light intensity, and current photosynthetic rate from different historical transplanting time data in the second training set are used as the input data of the second convolutional neural network, and the corresponding historical transplanting time in the second training set is used as the output data of the second convolutional neural network. The second convolutional neural network adopts the same architecture as the first convolutional neural network, but the input features are 6-dimensional and the output is 1-dimensional (predicted transplanting time). The specific structure is as follows: the 6-dimensional features are weighted according to transplanting time and ordered as follows: current health score → expected improvement time → current growth stage → current photosynthetic rate → historical average light intensity → historical average temperature, constructing a 6-dimensional sequence. The pooling layer is simultaneously replaced with mean pooling to preserve the local synergistic relationship between the expected improvement time and the current health score, and between the photosynthetic rate and the historical average light intensity, thus solving the problem of static feature convolution having no physical meaning. The input layer receives 6-dimensional features, followed by the following layers: first convolutional layer (64 filters, kernel size 3, ReLU, same padding), first pooling layer (max pooling, window 2), second convolutional layer (32 filters, kernel size 3, ReLU), second pooling layer (max pooling, window 2), flattening layer, and fully connected layer (128 neurons, ReLU, Dropout). The training parameters were the same as the first model: Adam optimizer (learning rate 0.001), batch size 32, loss function MSE, early stopping mechanism (stopping if the validation set loss does not decrease for 10 consecutive epochs), and validation set error threshold set to 0.05 days. Finally, the model with the best performance on the validation set was selected as the transplant time prediction model. The second convolutional neural network is trained to obtain the second initial convolutional neural network. The second initial convolutional neural network is validated using the second test set. The second initial convolutional neural network that outputs a second test error threshold less than or equal to the preset second test error threshold is used as the transplanting time prediction model. The current health score, expected improvement time, current growth stage, historical average temperature, historical average light intensity and current photosynthetic rate of each area to be intervened are input into the transplanting time prediction model to obtain the predicted transplanting time for each area to be intervened. In an embodiment of the present invention, the predicted transplanting time for all areas to be intervened is obtained by a transplanting time prediction model. The predicted transplanting time is related to the current health score, the expected improvement time, the current growth stage, the historical average temperature, the historical average light intensity, and the current photosynthetic rate. The current health score directly affects the predicted transplanting time. The lower the current health score, the longer the predicted transplanting time. Therefore, the current health score is negatively correlated with the transplanting time. The length of the expected improvement period directly affects the length of the predicted transplanting period. The longer the expected improvement period, the longer the predicted transplanting period. Therefore, the expected improvement period and the transplanting period are linearly positively correlated. In this embodiment, the transplanting time consists of two parts: the expected improvement time and the normal growth time after improvement. These times do not overlap, and the formula is: ,in To predict the transplanting time (in days). The expected improvement time (in days) is given by β, which is the error correction factor. To improve the normal growth time (days); Based on the current growth stage of the crop, the positive value is 8-10 days for the cotyledon stage, 5-7 days for the true leaf stage, and 2-3 days for the rapid growth stage. β is calculated based on the error rate of the predicted improvement time model from the past 30 monitoring historical data. If ε≤5%, then β=1.0; if 5%<ε≤10%, then β=1.05; if ε>10%, then β=1.1, thus achieving adaptive compensation for prediction errors of intermediate variables. Meanwhile, during the training of the transplanting time prediction model, the improved error fluctuation range (±10%) is incorporated into the second training set, enabling the model to learn the characteristic patterns under error fluctuations, further improving the overall robustness of the two-level prediction architecture, and ensuring that even if there is a small error in the improved time prediction, the transplanting time prediction result remains accurate. The later the current growth stage, the shorter the remaining time. The level of the historical average temperature for the same period directly affects the predicted transplanting time. The higher the historical average temperature for the same period, the shorter the predicted transplanting time. Therefore, the predicted improvement time and the transplanting time are positively and negatively correlated. The strength of the historical average sunlight directly affects the predicted transplanting time. The stronger the historical average sunlight, the shorter the predicted transplanting time. Therefore, the historical average sunlight and the transplanting time are positively and negatively correlated. The current photosynthetic rate directly affects the predicted transplanting time. The faster the current photosynthetic rate, the shorter the predicted transplanting time. Therefore, the current photosynthetic rate and transplanting time are positively and negatively correlated.

[0047] It should be further explained that, in the specific implementation process, the process of optimizing scheduling based on the predicted transplanting time includes: Based on the transplanting time prediction model, the predicted transplanting time for each area to be intervened is obtained. Combined with the target transplanting date in the production plan, a unified target transplanting date is set for the seedling greenhouse. The remaining available days between the current date and the target transplanting date are calculated. For the area to be intervened, its predicted transplanting time is compared with the remaining available days. If the predicted transplanting time is less than or equal to the remaining available days, it indicates that the area is expected to reach the transplanting standard on schedule and no extra attention is required. If the predicted transplanting time is greater than the remaining available days, it is determined that the area has a risk of delay and resources need to be prioritized to accelerate its growth. Calculate the number of days beyond the expected transplanting time for each area at risk of delay, i.e., the predicted transplanting time minus the remaining available days. Sort these areas in descending order of the number of days beyond the expected transplanting time. The larger the number of days beyond the expected transplanting time, the more severe the delay, and the more likely the area should receive additional water, nutrient solution, and light resources. When multiple areas requiring intervention need resources simultaneously and resources are limited, a comprehensive priority is calculated by combining the current health scores of each area. The current health score is multiplied by the reciprocal of the predicted transplanting time, and the higher the value, the higher the priority, ensuring that areas with a good health foundation and the potential for rapid recovery receive priority support. Set a safety margin, such as three days. When the number of days exceeding the safety margin in a certain area to be intervened exceeds the number of days, an early warning notification will be automatically sent to the administrator, indicating that the area to be intervened may be seriously delayed and suggesting that backup seedlings be prepared in advance or the transplanting plan be adjusted. The predicted transplanting time and target date of all areas to be intervened are compared and a seedling progress report is generated to provide a basis for decision-making on the adjustment of subsequent batch sowing time and the optimization of environmental parameters.

[0048] A seedling cultivation method based on big data includes the following steps: S1: Collect health parameters in real time within the unit's seedling area using data acquisition equipment, and then normalize the health parameters. S2: Generate a health score for each seedling area based on the processed health parameters, set a monitoring cycle, and calculate the rate of change of the health score for each monitoring cycle. S3: The health of the unit seedling area is classified into four levels: excellent, good, sub-healthy, and needing intervention, based on the health score and the rate of change of the health score. Corresponding measures are taken for the unit seedling area at different levels. S4: Establish a prediction model for the necessary resource replenishment in the area to be intervened, which is used to predict the necessary resource replenishment in the area to be intervened, namely, predict the amount of water replenishment per supply, predict the amount of nutrient solution replenishment per supply, and predict the increase in light intensity. S5: The replenishment coefficient is obtained based on the predicted single water supply replenishment amount, the predicted single nutrient solution replenishment amount, and the predicted increase in light intensity. The expected improvement time is then obtained based on the replenishment coefficient. S6: Construct a transplanting time prediction model for the area to be intervened based on the expected improvement time, and schedule resources according to the predicted transplanting time.

[0049] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0050] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A seedling cultivation system based on big data, characterized in that, Includes the following modules: The data acquisition module is used to collect health parameters in real time within the unit's seedling area through acquisition devices; The health assessment module is used to generate a health score for a unit seedling area based on health parameters, and to calculate the rate of change of the health score for a unit seedling area based on the health score. The health grading module is used to classify the health of a unit seedling area based on the health score and the rate of change of the health score, and to take corresponding measures for unit seedling areas of different levels. The necessary resource replenishment prediction module is used to acquire historical necessary resource replenishment data and build a necessary resource replenishment prediction model, and use the necessary resource replenishment prediction model to predict the necessary resource replenishment in the area to be intervened. The improvement time prediction module is used to calculate the replenishment coefficient based on the predicted required resource replenishment amount, and then obtain the expected improvement time based on the replenishment coefficient. The transplanting time prediction module is used to acquire historical transplanting time data and build a transplanting time prediction model. The transplanting time prediction model is used to predict the transplanting time of the area to be intervened, and the scheduling is optimized based on the predicted transplanting time.

2. The seedling cultivation system based on big data according to claim 1, characterized in that, The seedling base has multiple seedling greenhouses, all containing the same crop. Each greenhouse is divided into multiple seedling areas. The health parameters include the average stem diameter, plant height, number of leaves, leaf area, and relative chlorophyll content of the crop within each seedling area. These health parameters are normalized.

3. The seedling cultivation system based on big data according to claim 2, characterized in that, The process of generating a health score for a unit seedling area based on health parameters includes: Health score S: In the formula, , , , and These are weighting coefficients, obtained through training based on historical data; Gradient descent is used to train the weight coefficients, with the objective function being to maximize the fit between the health score and the actual growth status of the crop during the seedling stage, and the mean squared error being used as the metric. The actual loss function is set with a learning rate of 0.

001. The iteration stops when the loss function value is ≤0.01, thus obtaining the optimal combination of weight coefficients. The process of calculating the rate of change of health score per unit seedling area based on health score includes: Set a monitoring cycle. At the end of each monitoring cycle, calculate the health score of each seedling area. For each seedling area, save the historical health score data for the most recent few days. Plot the time sequence on the x-axis and the corresponding health score on the y-axis. Perform a univariate linear regression on these data points to obtain a straight line that reflects the change of the score over time. The slope of this line is the rate of change of the health score within that time period.

4. The seedling cultivation system based on big data according to claim 3, characterized in that, The process of classifying the health of a unit seedling area based on health scores and the rate of change in health scores includes: Based on historical health score data, set appropriate first health score threshold, second health score threshold and third health score threshold. The historical health score data refers to the data set of health scores of the previous production line. Compare the current health score of the production line with the health score threshold. Based on historical health score change rate data, set appropriate first health score change rate threshold, second health score change rate threshold and third health score change rate threshold. The historical health score change rate data refers to the data set of health score change rates of the previous production line. Compare the current health score change rate of the production line with the health score change rate threshold. If the current health score of a unit's seedling area is greater than or equal to the first health score threshold and the current health score change rate is greater than or equal to the first health score change rate threshold, it is judged as an excellent area and no intervention is required; only routine environmental monitoring is needed. If the second health score threshold is less than or equal to the current health score of the unit seedling area, which is less than the first health score threshold, and the current health score change rate is greater than or equal to the second health score change rate threshold, it is judged as a good area. The change trend should be monitored, and environmental parameters can be adjusted appropriately to promote its development towards excellence. If the third health score threshold is less than or equal to the current health score of the unit seedling area which is less than the second health score threshold, or the current health score change rate is less than the second health score change rate threshold, it is judged as a sub-healthy area, and the cause needs to be analyzed and targeted measures taken. If the current health score of a unit's seedling area is less than the third health score threshold, or the current health score change rate is less than the third health score change rate threshold, it is determined to be an area requiring intervention, and the necessary resource supplementation intervention process must be carried out immediately.

5. The seedling cultivation system based on big data according to claim 4, characterized in that, The process of acquiring historical data on necessary resource replenishment and constructing a necessary resource replenishment prediction model, and then using this model to predict the necessary resource replenishment amount for the area to be intervened in, includes: The necessary resource replenishment intervention process refers to replenishing the area to be intervened with necessary resources. Necessary resources refer to water resources, nutrient solution resources and light resources. The amount of necessary resource replenishment refers to the amount of water replenishment, the amount of nutrient solution replenishment, and the amount of light intensity increase in a single supply. Factors affecting the amount of necessary resource replenishment include: the current health score of the area to be intervened, the rate of change of the current health score, the current frequency of resource replenishment, the current single water supply, the current single nutrient solution volume, the current light intensity, the current substrate moisture content, and the current nutrient solution pH value; Acquire historical data on the amount of necessary resources to be replenished for a single area to be intervened in different monitoring periods. The historical data on the amount of necessary resources to be replenished includes the current health score, the rate of change of the current health score, the current frequency of resource replenishment, the current single water supply, the current single nutrient solution volume, the current light intensity, the current substrate moisture content, the current nutrient solution pH value, and the historical amount of necessary resources to be replenished for the single area to be intervened in the monitoring period. Based on the current health score, current health score change rate, current replenishment frequency, current single water supply, current single nutrient solution volume, current light intensity, current substrate moisture content, current nutrient solution pH value, and corresponding historical necessary resource replenishment data of the corresponding areas to be intervened, a necessary resource replenishment prediction set is generated and divided into the first training set and the first test set. The 8-dimensional static feature vectors are sorted according to their influence on crop growth to construct a one-dimensional sequence of length 8, giving the features ordered growth correlation characteristics, so that the one-dimensional convolution can extract the synergistic influence between features. The convolution kernel size 3 is used to capture the local coupling relationship between three adjacent features. The pooling window 2 uses mean pooling to retain feature coupling information and avoid loss of effective information. The first convolutional neural network is constructed. The first convolutional neural network is a one-dimensional convolutional neural network containing two convolutional layers, two pooling layers, a flattening layer, and two fully connected layers. A Dropout layer is set after the first fully connected layer. The network is trained using the Adam optimizer, the loss function is mean squared error, and the training termination is determined by early stopping. The current health score, current health score change rate, current resource replenishment frequency, current single water supply, current single nutrient solution volume, current light intensity, current substrate moisture content, and current nutrient solution pH value from the different historical necessary resource replenishment data in the first training set are used as the input data of the first convolutional neural network, and the corresponding historical necessary resource replenishment amounts in the first training set are used as the output data of the first convolutional neural network. The first convolutional neural network is trained to obtain the first initial convolutional neural network. The first initial convolutional neural network is validated using the first test set. The first initial convolutional neural network, whose output is less than or equal to the preset first test error threshold, is used as the required resource replenishment prediction model. The current health score, current health score change rate, current resource replenishment frequency, current single water supply volume, current single nutrient solution volume, current light intensity, current substrate moisture content, and current nutrient solution pH value of each area to be intervened are input into the required resource replenishment prediction model to obtain the predicted required resource replenishment amount for each area to be intervened, namely the predicted single water supply replenishment amount, the predicted single nutrient solution replenishment amount, and the predicted light intensity increase amount.

6. The seedling cultivation system based on big data according to claim 5, characterized in that, The process of calculating the replenishment coefficient based on the predicted required resource replenishment includes: In order to comprehensively assess the urgency of resource replenishment, a replenishment coefficient K is obtained based on the predicted single water supply replenishment amount, the predicted single nutrient solution replenishment amount, and the predicted increase in light intensity. The predicted single water supply replenishment amount is normalized: In the formula, To normalize the predicted single water supply replenishment amount, This is the actual predicted single water supply replenishment amount. The largest single water supply replenishment volume that has appeared in historical data; The predicted amount of nutrient solution to be replenished at one time was normalized: In the formula, To normalize the predicted amount of nutrient solution to be added at one time, This is to predict the actual amount of nutrient solution to be supplemented per feeding. The largest single nutrient solution replenishment volume that appeared in historical data; The predicted increase in light intensity was normalized: In the formula, To normalize the predicted increase in light intensity, This represents the actual predicted increase in light intensity. This represents the maximum increase in light intensity observed in historical data. Supplementary coefficient K: In the formula, , and The weighting coefficient can be determined based on the sensitivity of resources to crop growth; The process of obtaining the expected improvement time based on the supplementary coefficient includes: During operation, historical data of necessary resource interventions in the area to be intervened are continuously recorded. Each historical record includes the supplementary coefficient at the time of intervention and the actual number of days from the start of the intervention to the improvement of the health score of the area to be intervened to the target value, i.e. the improvement time. A large number of historical data points are plotted on a coordinate system with the supplementary coefficient as the x-axis and the improvement time as the y-axis. Through regression analysis, it is found that there is a power function relationship between the two. The least squares method is used to fit the data points to obtain the empirical formula for the crop variety at the current growth stage. The power function expression is T= T represents the expected improvement time (days), K is the supplementary coefficient, and a, b, and c are the fitting constants for crop growth stages. The objective is to minimize the sum of squared residuals. The initial parameter ranges are set as a∈(0, 10], b∈(-5, 0), and c∈(0, 5]. The convergence condition is that the sum of squared residuals S≤0.05 or 500 iterations. After fitting, the coefficient of determination is used. Validate if the value is ≥0.85; if not, expand the parameter range and refit. When intervention is needed in a certain area, first calculate its current supplementary coefficient, and then substitute the supplementary coefficient into the fitted power function curve to obtain the expected improvement time.

7. The seedling cultivation system based on big data according to claim 6, characterized in that, The process of acquiring historical transplanting time data and constructing a transplanting time prediction model, and then using this model to predict the transplanting time for the area to be intervened, includes: Crops in the seedling greenhouse need to reach the transplantable standard before the specified time. Crops in other levels of unit seedling areas will definitely grow faster than those in the areas to be intervened. Therefore, it is only necessary to consider the estimated transplanting time of all areas to be intervened. The transplanting time refers to the time required for crops to reach the transplantable standard from the present. Factors influencing the transplanting time in the area to be intervened include: the current health score of the area to be intervened, the expected improvement time, the current growth stage, the historical average temperature for the same period, the historical average light intensity for the same period, and the current photosynthetic rate; Historical transplanting time data of a single area to be intervened in different monitoring periods are obtained. The historical transplanting time data includes the current health score, expected improvement time, current growth stage, historical average temperature, historical average light intensity, current photosynthetic rate, and historical transplanting time of the single area to be intervened. Based on the current health score, expected improvement time, current growth stage, historical average temperature, historical average light intensity and current photosynthetic rate of the corresponding area to be intervened in different historical transplanting time data, a transplanting time prediction set is generated and divided into a second training set and a second test set. The 6-dimensional static feature vectors are sorted according to the weights determined by crop growth to construct a one-dimensional sequence of length 6, giving the features ordered growth-related attributes. The one-dimensional convolutional kernel size of 3 is used to capture the local synergistic effect of three adjacent features. The pooling window 2 uses mean pooling to retain feature coupling information, taking into account both feature dimension compression and information integrity. A second convolutional neural network is constructed. The second convolutional neural network is a one-dimensional convolutional neural network containing two convolutional layers, two pooling layers, a flattening layer, and two fully connected layers. A Dropout layer is set after the first fully connected layer. The network is trained using the Adam optimizer, the loss function is mean squared error, and the training termination is determined by early stopping. The current health score, expected improvement time, current growth stage, historical average temperature, historical average light intensity, and current photosynthetic rate from different historical transplanting time data in the second training set are used as input data for the second convolutional neural network, and the corresponding historical transplanting time in the second training set is used as output data for the second convolutional neural network. The second convolutional neural network is trained to obtain the second initial convolutional neural network. The second initial convolutional neural network is validated using the second test set. The second initial convolutional neural network that outputs a second test error threshold less than or equal to the preset second test error threshold is used as the transplanting time prediction model. The current health score, expected improvement time, current growth stage, historical average temperature, historical average light intensity, and current photosynthetic rate of each area to be intervened are input into the transplanting time prediction model to obtain the predicted transplanting time for each area to be intervened.

8. The seedling cultivation system based on big data according to claim 7, characterized in that, The process of optimizing scheduling based on predicted transplanting times includes: Based on the transplanting time prediction model, the predicted transplanting time for each area to be intervened is obtained. Combined with the target transplanting date in the production plan, a unified target transplanting date is set for the seedling greenhouse. The remaining available days between the current date and the target transplanting date are calculated. For the area to be intervened, its predicted transplanting time is compared with the remaining available days. If the predicted transplanting time is less than or equal to the remaining available days, it indicates that the area is expected to reach the transplanting standard on schedule and no extra attention is required. If the predicted transplanting time is greater than the remaining available days, it is determined that the area has a risk of delay and resources need to be prioritized to accelerate its growth. Calculate the number of days beyond the expected transplanting time for each area at risk of delay, i.e., the predicted transplanting time minus the remaining available days. Sort these areas in descending order of the number of days beyond the expected transplanting time. The larger the number of days beyond the expected transplanting time, the more severe the delay, and the more likely the area should receive additional water, nutrient solution, and light resources. When multiple areas requiring intervention need resources simultaneously and resources are limited, a comprehensive priority is calculated by combining the current health scores of each area. The current health score is multiplied by the reciprocal of the predicted transplanting time, and the higher the value, the higher the priority, ensuring that areas with a good health foundation and the potential for rapid recovery receive priority support. Set a safety margin, such as three days. When the number of days exceeding the safety margin in a certain area to be intervened exceeds the number of days, an early warning notification will be automatically sent to the administrator, indicating that the area to be intervened may be seriously delayed and suggesting that backup seedlings be prepared in advance or the transplanting plan be adjusted. The predicted transplanting time and target date of all areas to be intervened are compared and a seedling progress report is generated to provide a basis for decision-making on the adjustment of subsequent batch sowing time and the optimization of environmental parameters.

9. A seedling cultivation method based on big data, implemented based on the big data-based seedling cultivation system according to any one of claims 1-8, characterized in that, Includes the following steps: S1: Collect health parameters in real time within the unit's seedling area using data acquisition equipment, and then normalize the health parameters. S2: Generate a health score for each seedling area based on the processed health parameters, set a monitoring cycle, and calculate the rate of change of the health score for each monitoring cycle. S3: The health of the unit seedling area is classified into four levels: excellent, good, sub-healthy, and needing intervention, based on the health score and the rate of change of the health score. Corresponding measures are taken for the unit seedling area at different levels. S4: Establish a prediction model for the necessary resource replenishment in the area to be intervened, which is used to predict the necessary resource replenishment in the area to be intervened, namely, predict the amount of water replenishment per supply, predict the amount of nutrient solution replenishment per supply, and predict the increase in light intensity. S5: The replenishment coefficient is obtained based on the predicted single water supply replenishment amount, the predicted single nutrient solution replenishment amount, and the predicted increase in light intensity. The expected improvement time is then obtained based on the replenishment coefficient. S6: Construct a transplanting time prediction model for the area to be intervened based on the expected improvement time, and schedule resources according to the predicted transplanting time.

Images