Isatis root planting environment monitoring method and system based on neural network

Abstract

The invention discloses an isatis root planting environment monitoring method and system based on a neural network, and relates to the technical field of traditional Chinese medicinal material growth monitoring, and the method comprises the steps: extracting environment feature data; traversing all the environment feature data, searching for adjacent environment feature data of the target environment feature data, comparing the difference value between the target environment feature data and the adjacent environment feature data, and generating new target environment feature data as a key factor influencing the growth of the radix isatidis based on a comparison result; and predicting the growth demand of the isatis root at a future moment by using the neural network prediction model, generating an irrigation and fertilization regulation scheme, and realizing environment monitoring of isatis root planting. The isatis root planting management system can monitor and adjust environmental conditions in real time, is beneficial to improving the yield and pesticide effect of isatis roots, ensures the sustainability and stability of the planting process, and improves the planting management benefits of isatis roots to the maximum extent.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of Chinese herbal medicine growth monitoring, in particular to a GAPS-based environment monitoring method and system for planting GAPS. BACKGROUND

[0002] GAPS is a common Chinese herbal medicine, which is derived from the dried roots of Brassica campestris L. var. parachinensis Bailey, usually harvested in autumn and used as medicine after processing. It is planted in various parts of China and divided into two types: northern GAPS and southern GAPS. Northern GAPS comes from the roots of Brassica campestris L., while southern GAPS is derived from the roots and rhizomes of the plant Plumbago zeylanica L. GAPS is cold in nature and slightly sweet in taste, with bitter aftertaste, and has the effects of clearing heat and detoxifying, preventing colds, and benefiting the throat. It is widely used to treat symptoms such as fever with skin rashes, dark purple tongue, and hoarse throat. In modern agriculture, the planting of GAPS has high requirements for environmental conditions, such as temperature, humidity, and soil composition, which directly affect its growth and medicinal effect. Therefore, GAPS-based environment monitoring technology based on neural networks has emerged, which monitors various indicators of the planting environment in real time, such as soil moisture, light intensity, and climate change data, uses a neural network model to predict growth requirements, and optimizes planting management to improve yield and medicinal material quality.

[0003] However, the existing technology in GAPS planting environment monitoring often relies on traditional manual monitoring and empirical control, which is difficult to capture environmental changes in real time and accurately, lacks in-depth analysis of environmental data, cannot fully identify key factors affecting growth, and lacks scientific basis in irrigation and fertilization control, which can easily cause resource waste or poor results. In addition, it is not convenient to realize dynamic prediction and automatic adjustment, which is difficult to cope with complex and variable environmental conditions, limiting the improvement of planting benefits.

[0004] To address the problems in the related art, no effective solutions have been proposed so far. SUMMARY

[0005] To address the problems in the related art, no effective solutions have been proposed so far.

[0006] To achieve the above purposes, the present application is implemented by the following technical solutions:

[0007] According to a first aspect of the present invention, a method for monitoring the planting environment of Isatis indigotica based on a neural network is provided, comprising:

[0008] Environmental data on Isatis indigotica cultivation was obtained, and the environmental data was preprocessed to extract environmental feature data.

[0009] Traverse all environmental feature data, find neighboring environmental feature data of the target environmental feature data, compare the difference between the target environmental feature data and the neighboring environmental feature data, and based on the comparison results, generate new target environmental feature data as the key factor affecting the growth of Isatis indigotica.

[0010] A neural network prediction model was constructed using key factors, and this model was used to predict the future growth needs of Isatis indigotica, generating irrigation and fertilization control schemes to achieve environmental monitoring of Isatis indigotica cultivation.

[0011] Furthermore, all environmental feature data are traversed to find neighboring environmental feature data of the target environmental feature data, and the differences between the target environmental feature data and neighboring environmental feature data are compared. Based on the comparison results, new target environmental feature data is generated as key factors affecting the growth of Isatis indigotica, including:

[0012] Initialize the parameters and maximum number of iterations of the anomaly analysis algorithm, and set the starting point of the anomaly analysis as the initial environmental characteristics of Isatis indigotica growth, with the goal of identifying key factors affecting growth.

[0013] The search algorithm is used to traverse all environmental feature data, find the neighboring environmental feature data of the target environmental feature data, and check whether the difference between the neighboring environmental feature data and the target environmental feature data exceeds the preset value.

[0014] If the difference between the neighboring environmental feature data and the target environmental feature data is greater than the preset value, a weighting coefficient is introduced to generate new target environmental feature data; otherwise, the preset value is adjusted and new target environmental feature data is generated by combining the new weighting coefficient.

[0015] Check for anomalies in the correlation between the new target environmental feature data and the neighboring environmental feature data. If no anomalies are found, add the new target environmental feature data to the analysis model. If anomalies are found, readjust the preset values ​​and weight coefficients to update the target environmental feature data.

[0016] Determine if the maximum number of iterations has been reached. If it has, output the optimal target environment characteristic data as the key factor affecting the growth of Isatis indigotica; otherwise, continue iterative analysis.

[0017] Furthermore, a search algorithm is used to traverse all environmental feature data, find neighboring environmental feature data of the target environmental feature data, and check whether the difference between the neighboring environmental feature data and the target environmental feature data exceeds a preset value, including:

[0018] Initialize the parameters of the search algorithm and the maximum number of iterations;

[0019] Starting with the target environment feature data, construct the environment search structure and initialize the expandable state set, and begin to filter and expand the paths that match the target environment feature data from the candidate environment feature set;

[0020] Using a screening algorithm, environmental feature data that best matches the combination of target environmental features is selected from the current candidate environmental feature set for expansion, and the optimal expansion path matching the target environmental feature data is gradually established.

[0021] Recursively generate other candidate environmental feature data for the current state and extend them to the remaining branches of the previous state to form a diverse neighboring environmental feature data space.

[0022] The differences between the environmental feature data in each extended path and the target environmental feature data are evaluated. If the difference exceeds a preset threshold, conflict information is generated for further screening.

[0023] Based on the conflict information, environmental feature data that does not meet the preset conditions are eliminated, the scalable candidate environmental feature data set is updated, and the search continues until the best neighboring environmental feature data is found.

[0024] Furthermore, by using a screening algorithm to select environmental feature data from the current candidate environmental feature set that best matches the combination of target environmental features, and expanding the selection, the optimal expansion path for matching the target environmental feature data is gradually established, including:

[0025] Set the parameters and maximum number of iterations for the sieving algorithm, and initialize the search state set;

[0026] Based on the differences between candidate environmental feature data and target environmental feature data, the fitness value of each candidate environmental feature data is calculated as the basis for selecting expansion paths;

[0027] Calculate the search space size of the current candidate environment feature data according to the preset formula, and determine the scalable data volume;

[0028] All candidate environmental feature data are sorted in a non-dominated manner, and then divided into primary and secondary data groups. The primary data group with the smallest difference from the target environmental feature data is selected for expansion.

[0029] Calculate the fitness value of the newly generated candidate environment feature data and compare it with the parent data to select the optimal expansion path. Update the scalable candidate environment feature data set. If the maximum number of iterations is reached, terminate the search and output the optimal expansion path; otherwise, continue iterating.

[0030] Furthermore, all candidate environmental feature data are non-dominated and ranked, and then divided into primary and secondary data groups. The primary data group, which has the smallest difference from the target environmental feature data, is prioritized for expansion, including:

[0031] Fitness assessment is performed on all candidate environmental feature data, and the difference between each candidate environmental feature data and the target environmental feature data is calculated as the basis for ranking and grouping.

[0032] Based on the fitness values, the candidate environmental feature data are non-dominatedly sorted and divided into primary data group and secondary data group.

[0033] Select candidate environmental feature data that has the smallest difference from the target environmental feature data from the main data set, and use it as the starting point for the expansion path to ensure the matching of the expansion with the target.

[0034] The selected main data sets are expanded to generate new candidate environmental feature data, and the search continues to gradually approach the target environmental feature data.

[0035] Furthermore, a neural network prediction model is constructed using key factors, and this model is used to predict the future growth requirements of Isatis indigotica, generating irrigation and fertilization control plans to achieve environmental monitoring of Isatis indigotica cultivation, including:

[0036] The key factors were divided into training and testing sets. Principal component analysis was used to decompose the key factors, extract multi-dimensional growth characteristics of Isatis indigotica, and establish multiple neural network prediction models based on each type of Isatis indigotica growth characteristics.

[0037] The parameters of multiple neural network prediction models are optimized using a parameter optimization algorithm to obtain multiple optimized neural network prediction models;

[0038] Multiple optimized neural network prediction models are used to predict the growth requirements of Isatis indigotica at future times. The prediction results of each neural network prediction model are superimposed and reconstructed to obtain a comprehensive prediction result. Based on the comprehensive prediction result, irrigation and fertilization control schemes are formulated to realize environmental monitoring of Isatis indigotica planting.

[0039] Furthermore, parameter optimization algorithms are used to optimize the parameters of multiple neural network prediction models, resulting in several optimized neural network prediction models, including:

[0040] Set the parameters and maximum number of iterations for the parameter optimization algorithm, and determine the optimization objective function and its constraints for multiple neural network prediction models;

[0041] Initialize multiple individual neural network prediction models, each representing a set of parameter configurations for a neural network prediction model, and generate the initial neural network prediction model structure;

[0042] Based on the parameter configuration of the current neural network prediction model, train each neural network prediction model and evaluate its performance, record the current global best neural network prediction model and its corresponding parameter configuration, and update the global best neural network prediction model through random search.

[0043] If the fitness of the current neural network prediction model is greater than that of the best neural network prediction model in the previous iteration, the parameter configuration is updated by a forward operation; otherwise, a new parameter space is explored by a roll operation.

[0044] If the maximum number of iterations is reached, a breeding operation is performed to generate a new generation of neural network prediction model individuals and update the parameter configuration. The parameter configuration of the optimal neural network prediction model is output, and the next round of optimization is carried out.

[0045] After the optimization process is completed, multiple optimized neural network prediction models and their optimal parameter configurations are output.

[0046] According to a second aspect of the present invention, a neural network-based monitoring system for the planting environment of Isatis indigotica is provided, the system comprising:

[0047] The data acquisition module is used to acquire environmental data of Isatis indigotica cultivation, and to preprocess the environmental data to extract environmental feature data.

[0048] The anomaly analysis module is used to traverse all environmental feature data, find neighboring environmental feature data of the target environmental feature data, compare the difference between the target environmental feature data and the neighboring environmental feature data, and generate new target environmental feature data as the key factors affecting the growth of Isatis indigotica based on the comparison results.

[0049] The model prediction and control module is used to build a neural network prediction model using key factors, and to use the neural network prediction model to predict the growth needs of Isatis indigotica in the future, generate irrigation and fertilization control plans, and realize environmental monitoring of Isatis indigotica planting.

[0050] According to a third aspect of the present invention, an electronic device is provided, the electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program implementing the above-described method when executed by the processor.

[0051] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the above-described method.

[0052] The beneficial effects of this invention are as follows:

[0053] 1. This invention can accurately acquire and process environmental data to identify key factors affecting the growth of Isatis indigotica, optimize planting environment management, and generate scientific irrigation and fertilization control plans by predicting future growth needs, effectively improving resource utilization efficiency and reducing unnecessary waste. Simultaneously, real-time monitoring and adjustment of environmental conditions helps increase the yield and efficacy of Isatis indigotica, ensuring the sustainability and stability of the planting process and maximizing the benefits of Isatis indigotica planting and management.

[0054] 2. This invention utilizes anomaly analysis, search, and screening algorithms to dynamically model the growth environment characteristics of Isatis indigotica. It continuously optimizes target environmental characteristic data during the iterative process, accurately identifying key factors affecting growth. Through non-dominated ranking and difference assessment, it ensures effective screening and path expansion of candidate environmental characteristic data, improving the accuracy and robustness of environmental monitoring and providing reliable support for intelligent planting and scientific management of Isatis indigotica.

[0055] 3. This invention extracts multi-dimensional growth characteristics of Isatis indigotica using principal component analysis and accurately predicts future growth requirements based on a neural network prediction model. Parameter optimization algorithms are used to optimize the parameters of multiple neural network models, improving the accuracy and robustness of the predictions. The comprehensive prediction results provide scientific control schemes for irrigation and fertilization, thereby improving the accuracy of environmental monitoring and management efficiency in Isatis indigotica cultivation and supporting precision agriculture decision-making. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the 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.

[0057] Figure 1 This is a flowchart of a method for monitoring the planting environment of Isatis indigotica based on a neural network according to an embodiment of the present invention;

[0058] Figure 2 This is a schematic diagram of a neural network-based monitoring system for the planting environment of Isatis indigotica according to an embodiment of the present invention.

[0059] Figure 3 This is a schematic diagram of the hardware operating environment according to an embodiment of the present invention;

[0060] Figure 4 This is a flowchart of an anomaly analysis algorithm according to an embodiment of the present invention;

[0061] Figure 5 This is a flowchart of the search algorithm according to an embodiment of the present invention.

[0062] In the picture:

[0063] 1. Data acquisition module; 2. Anomaly analysis module; 3. Model prediction and control module. Detailed Implementation

[0064] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0065] In the description of this invention, unless otherwise stated, "a plurality of" means two or more. Furthermore, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0066] According to an embodiment of the present invention, a method and system for monitoring the planting environment of Isatis indigotica based on neural networks are provided.

[0067] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the method for monitoring the planting environment of Isatis indigotica based on neural networks according to an embodiment of the present invention includes:

[0068] S1. Obtain environmental data for Isatis indigotica cultivation, preprocess the environmental data, and extract environmental feature data.

[0069] Specifically, environmental data on Isatis indigotica cultivation was acquired, and the environmental data was preprocessed to extract environmental characteristic data, including:

[0070] Environmental data on Isatis indigotica cultivation is obtained, and the duplicate data, missing values ​​and outliers in the environmental data are denoised, filtered and smoothed to obtain complete environmental data. The complete real-time environmental data is arranged in chronological order to form a multidimensional environmental dataset.

[0071] Specifically, to achieve environmental data cleaning and feature extraction, the system first preprocesses duplicate data, missing values, and outliers after data collection. The following techniques are used: for duplicate data, a hash fingerprint deduplication and timestamp merging strategy is employed; for missing values, interpolation methods (linear interpolation, time-weighted interpolation) or nearest neighbor completion are used; for outliers, the Z-score method is used for initial screening, combined with sliding window statistical analysis (such as the 3σ principle) to identify short-term abrupt changes, and then Savitzky-Golay smoothing filtering or Locally Weighted Regression (LOESS) is used to smooth the time-series data.

[0072] After preprocessing, a time-series dataset containing multidimensional environmental parameters is constructed. The structure is a two-dimensional time-parameter matrix, arranged in ascending order of collection time, with each row representing the complete environmental state at a certain moment. All environmental data are normalized so that the parameters of each dimension fall within the range of [0, 1] to eliminate the scale effect.

[0073] Using each environmental data point in the multidimensional environmental dataset as the center point, calculate the feature similarity between that environmental data point and other environmental data points;

[0074] Set a similarity threshold, select environmental data points that match the similarity to the center point, and form a subset of environmental features;

[0075] The most representative feature groups are selected from the subset of environmental features as environmental feature data.

[0076] Specifically, using environmental data at a specific moment as the center point, the Euclidean distance function is used to calculate the feature similarity between this data and environmental data at other time points. A similarity threshold (e.g., Euclidean distance less than 0.25) is set as the nearest neighbor selection criterion, and several similar samples are selected from all samples to form a subset of environmental features. Principal component analysis (PCA) is applied to the feature variables in this subset to extract the main changing factors, or important features are selected based on indicators such as the Gini coefficient and information gain, ultimately forming a representative set of data as the environmental feature data input to the neural network model.

[0077] It should be noted that in this invention, using each environmental data point in the multidimensional environmental dataset as the center point means that the system sequentially uses the environmental data vector corresponding to each sampling time in the time series data as the analysis benchmark. For example, if the environmental parameters collected at a time point include temperature, humidity, light intensity, soil pH, and carbon dioxide concentration, a five-dimensional vector is formed, which serves as the center point at that time. The system sequentially performs this process on all time point data, forming a sliding center point strategy, thereby capturing environmental change characteristics within continuous time windows.

[0078] Specifically, environmental data for Isatis indigotica cultivation is obtained through methods such as IoT sensor monitoring, remote sensing and drone monitoring, and meteorological station data.

[0079] Specifically, environmental data includes meteorological data, soil data, moisture data, air data, etc.

[0080] Specifically, environmental characteristic data includes temperature characteristics, moisture characteristics, nutrient characteristics, and light characteristics.

[0081] S2. Traverse all environmental feature data, find neighboring environmental feature data of the target environmental feature data, and compare the difference between the target environmental feature data and the neighboring environmental feature data. Based on the comparison results, generate new target environmental feature data as the key factor affecting the growth of Isatis indigotica.

[0082] Specifically, key factors include temperature, humidity and moisture, light intensity, soil properties, nutrient supply, air quality and CO2 concentration, atmospheric pressure and wind speed, and irrigation volume.

[0083] like Figure 4 As shown, in this optional embodiment, all environmental feature data are traversed to find neighboring environmental feature data of the target environmental feature data, and the difference values ​​between the target environmental feature data and the neighboring environmental feature data are compared. Based on the comparison results, new target environmental feature data is generated as a key factor affecting the growth of Isatis indigotica, including:

[0084] Initialize the parameters and maximum number of iterations of the anomaly analysis algorithm, and set the starting point of the anomaly analysis as the initial environmental characteristics of Isatis indigotica growth, with the goal of identifying key factors affecting growth.

[0085] The search algorithm is used to traverse all environmental feature data, find the neighboring environmental feature data of the target environmental feature data, and check whether the difference between the neighboring environmental feature data and the target environmental feature data exceeds the preset value.

[0086] If the difference between the neighboring environmental feature data and the target environmental feature data is greater than the preset value, a weighting coefficient is introduced to generate new target environmental feature data; otherwise, the preset value is adjusted and new target environmental feature data is generated by combining the new weighting coefficient.

[0087] Check for anomalies in the correlation between the new target environmental feature data and the neighboring environmental feature data. If no anomalies are found, add the new target environmental feature data to the analysis model. If anomalies are found, readjust the preset values ​​and weight coefficients to update the target environmental feature data.

[0088] Determine if the maximum number of iterations has been reached. If it has, output the optimal target environment characteristic data as the key factor affecting the growth of Isatis indigotica; otherwise, continue iterative analysis.

[0089] Specifically, the parameters of the anomaly analysis algorithm are initialized, and the maximum number of iterations is set. The starting point for analysis is set as the initial environmental characteristic data of Isatis indigotica growth, including relevant data such as temperature, humidity, and light. The goal is to identify the key environmental factors affecting Isatis indigotica growth. Based on the target environmental characteristic data, a search algorithm is used to traverse all environmental characteristic data, find neighboring environmental characteristic data, and calculate the differences between these data and the target environmental characteristic data. If the difference exceeds a preset threshold, the algorithm introduces weight coefficients to generate new target environmental characteristic data; if the difference is small, the threshold and weight coefficients are adjusted, and the data is updated in conjunction with the new environmental characteristic data. The correlation between the newly generated target environmental characteristic data and neighboring environmental characteristic data is checked to ensure there are no anomalies. If there are no anomalies between the new data and neighboring data, it is added to the analysis model for further optimization. If anomalies exist, the system readjusts the threshold and weight coefficients and continues to update the target environmental characteristic data until the optimal environmental characteristic data that meets the conditions is found. It is determined whether the maximum number of iterations has been reached. If it has, the optimal target environmental characteristic data is output as a key factor affecting Isatis indigotica growth for subsequent decision-making; if it has not been reached, the target data continues to be iterated and updated. This allows for continuous adjustment and optimization of environmental characteristic data, ensuring that the identified key factors accurately reflect the environmental conditions affecting the growth of Isatis indigotica.

[0090] Specifically, the anomaly analysis algorithm is the Fast Expanding Random Tree algorithm, an algorithm used for path planning and state space exploration. It finds the optimal path by randomly generating nodes in the space and expanding the tree structure. In this invention, this algorithm is used to traverse the environmental feature data space, find neighboring environmental feature data of the target environmental feature data, and analyze the differences between these data by rapidly expanding the tree. If the difference exceeds a preset value, new target environmental feature data is generated by adjusting the weights, and the data is updated and optimized using the expansion process of the algorithm, ultimately identifying the key environmental factors affecting the growth of Isatis indigotica.

[0091] like Figure 5 As shown, in this optional embodiment, a search algorithm is used to traverse all environmental feature data, find neighboring environmental feature data of the target environmental feature data, and check whether the difference between the neighboring environmental feature data and the target environmental feature data exceeds a preset value, including:

[0092] Initialize the parameters of the search algorithm and the maximum number of iterations;

[0093] Starting with the target environment feature data, construct the environment search structure and initialize the expandable state set, and begin to filter and expand the paths that match the target environment feature data from the candidate environment feature set;

[0094] Using a screening algorithm, environmental feature data that best matches the combination of target environmental features is selected from the current candidate environmental feature set for expansion, and the optimal expansion path matching the target environmental feature data is gradually established.

[0095] Recursively generate other candidate environmental feature data for the current state and extend them to the remaining branches of the previous state to form a diverse neighboring environmental feature data space.

[0096] The differences between the environmental feature data in each extended path and the target environmental feature data are evaluated. If the difference exceeds a preset threshold, conflict information is generated for further screening.

[0097] Based on the conflict information, environmental feature data that does not meet the preset conditions are eliminated, the scalable candidate environmental feature data set is updated, and the search continues until the best neighboring environmental feature data is found.

[0098] Specifically, the search algorithm parameters and maximum iteration count are initialized, and the starting point of the search is set as the target environmental feature data (e.g., initial environmental parameters such as temperature and humidity). An environmental search structure is constructed, and an expandable state set is initialized. Data most similar to the target environmental feature data is selected from the candidate environmental feature set as the basis for path expansion. Using a screening algorithm, environmental feature data that best matches the target environmental feature data is selected from the current candidate environmental feature set, gradually constructing the optimal expansion path to ensure that the path meets the requirements of the target feature data. Other candidate environmental feature data for the current state are recursively generated and expanded to other branches of the previous state, gradually constructing a diverse space of neighboring environmental feature data. Based on this, a difference evaluation is performed on each expansion path, comparing the differences between the environmental feature data in each path and the target environmental feature data. If the difference value of a path exceeds a preset threshold, the system generates conflict information, marking the path as ineligible. Conflict information is used to filter out environmental feature data that does not meet the requirements and update the expandable candidate data set. This process continues until the optimal neighboring environmental feature data is found, or the maximum number of iterations is reached. The system outputs optimal neighboring environmental characteristic data, which serves as a key factor influencing the growth of Isatis indigotica, providing a precise basis for subsequent environmental monitoring and control measures. This ensures the optimization and accurate matching of environmental data, helping to improve the growth conditions of Isatis indigotica and the level of intelligent farmland management.

[0099] Specifically, the search algorithm is the Conflict A* algorithm, an improved version of the A* search algorithm, used to avoid conflicts in path planning. It selects the optimal path by evaluating the cost and conflict information of each path during the search process. Unlike the standard A* algorithm, the Conflict A* algorithm introduces conflict detection and handling during path evaluation, avoiding invalid path expansion. In this invention, the algorithm is used to traverse environmental feature data and find the nearest neighbor data that is closest to the target environmental feature data. By calculating the difference value of each path, if the difference exceeds a preset threshold, the algorithm generates conflict information and eliminates paths that do not meet the conditions, ultimately finding the optimal neighbor environmental feature data.

[0100] In this optional embodiment, the selection of environmental feature data that best matches the target environmental feature combination from the current candidate environmental feature set using a screening algorithm, and the gradual establishment of the optimal expansion path matching the target environmental feature data, includes:

[0101] Set the parameters and maximum number of iterations for the sieving algorithm, and initialize the search state set;

[0102] Based on the differences between candidate environmental feature data and target environmental feature data, the fitness value of each candidate environmental feature data is calculated as the basis for selecting expansion paths;

[0103] Calculate the search space size of the current candidate environment feature data according to the preset formula, and determine the scalable data volume;

[0104] All candidate environmental feature data are sorted in a non-dominated manner, and then divided into primary and secondary data groups. The primary data group with the smallest difference from the target environmental feature data is selected for expansion.

[0105] Calculate the fitness value of the newly generated candidate environment feature data and compare it with the parent data to select the optimal expansion path. Update the scalable candidate environment feature data set. If the maximum number of iterations is reached, terminate the search and output the optimal expansion path; otherwise, continue iterating.

[0106] Specifically, the parameters and maximum number of iterations of the screening algorithm are set, and the search state set is initialized to ensure that the algorithm can gradually expand paths that match the target environment feature data during the search process. Based on the differences between candidate environment feature data and target environment feature data, the fitness value of each candidate data is calculated; these values ​​will serve as the basis for filtering and expanding paths. The search space size of the current candidate data is calculated according to a preset formula to determine the amount of data that can be expanded, providing a reasonable range for subsequent searches and path expansion. All candidate environment feature data are non-dominated and sorted, dividing them into primary and secondary data groups. The primary data group with the smallest difference from the target environment feature data is prioritized for expansion, ensuring the highest matching degree. As expansion progresses, new candidate environment feature data is generated, and its fitness value is calculated and further compared with the parent data to select the optimal expansion path. In each iteration, the set of expandable candidate environment feature data is updated. If the maximum number of iterations is reached, the search terminates and the optimal expansion path is output; otherwise, iteration continues, optimizing the path and improving the matching degree. Through this process, the optimal expansion path is gradually established, which helps to identify environmental data that best matches the target environmental characteristics data, and ultimately obtains the key environmental factors that affect the growth of Isatis indigotica.

[0107] Specifically, the screening algorithm is the dandelion algorithm, a heuristic algorithm that simulates the natural dispersal process of dandelions. It gradually finds the optimal solution through information exchange and fitness competition among individuals in a population. In this invention, this algorithm is used to select data that best matches the target environment feature data from candidate environment feature data for expansion. The algorithm calculates fitness values, performs non-dominated sorting, prioritizes expanding the data group with the smallest differences, and continuously updates the expandable data set through multiple iterations until the optimal expansion path is found or the maximum number of iterations is reached.

[0108] In this optional embodiment, all candidate environmental feature data are non-dominatedly sorted, and all candidate environmental feature data are divided into primary data groups and secondary data groups. The primary data group with the smallest difference from the target environmental feature data is preferentially selected for expansion, including:

[0109] Fitness assessment is performed on all candidate environmental feature data, and the difference between each candidate environmental feature data and the target environmental feature data is calculated as the basis for ranking and grouping.

[0110] Specifically, the formula for calculating the difference between each candidate environmental feature data and the target environmental feature data is as follows:

[0111] ;

[0112] In the formula, C ah represents the total difference measure between the a-th candidate environmental feature data and the target environmental feature data; ab k represents the b-th feature value in the a-th candidate environmental feature data; b ω represents the b-th feature value in the target environment feature data; b λ represents the weight of the b-th eigenvalue; λ represents the harmonic factor, which takes values ​​in [0, 1] and is used to control the weight ratio of the squared difference term and the absolute difference term; m represents the number of eigenvalues.

[0113] Based on the fitness values, the candidate environmental feature data are non-dominatedly sorted and divided into primary data group and secondary data group.

[0114] Select candidate environmental feature data that has the smallest difference from the target environmental feature data from the main data set, and use it as the starting point for the expansion path to ensure the matching of the expansion with the target.

[0115] The selected main data sets are expanded to generate new candidate environmental feature data, and the search continues to gradually approach the target environmental feature data.

[0116] Specifically, fitness evaluation is performed on all candidate environmental feature data. The difference between each candidate data point and the target environmental feature data is calculated and used as the basis for subsequent ranking and grouping. A lower fitness value indicates a smaller difference between the candidate data and the target environmental feature data, and a better match to the target requirements. Based on the fitness value, all candidate environmental feature data are non-dominatedly ranked and divided into primary and secondary data groups. The primary data group contains data with the smallest difference from the target environmental feature data, while the secondary data group contains data with larger differences. The candidate environmental feature data with the smallest difference is selected from the primary data group as the starting point for the expansion path, ensuring that subsequent path expansion maintains a high degree of matching with the target environmental feature data. The selected primary data group is then expanded to generate new candidate environmental feature data, and the search process continues. This process gradually approaches the target environmental feature data, continuously generating new environmental feature data through iterative optimization, making the expansion path gradually conform to the target data. This ensures that the data that best matches the target environmental features is selected from the candidate data first, and by continuously optimizing the path, the accuracy of the environmental feature data is improved, ultimately providing an accurate reference for the key environmental factors for the growth of Isatis indigotica.

[0117] S3. Construct a neural network prediction model using key factors, and use the neural network prediction model to predict the growth needs of Isatis indigotica in the future, generate irrigation and fertilization control schemes, and realize environmental monitoring of Isatis indigotica planting.

[0118] In this optional embodiment, a neural network prediction model is constructed using key factors, and the neural network prediction model is used to predict the future growth requirements of Isatis indigotica, generating irrigation and fertilization control schemes to achieve environmental monitoring of Isatis indigotica cultivation, including:

[0119] The key factors were divided into training and testing sets. Principal component analysis was used to decompose the key factors, extract multi-dimensional growth characteristics of Isatis indigotica, and establish multiple neural network prediction models based on each type of Isatis indigotica growth characteristics.

[0120] The parameters of multiple neural network prediction models are optimized using a parameter optimization algorithm to obtain multiple optimized neural network prediction models;

[0121] Multiple optimized neural network prediction models are used to predict the growth requirements of Isatis indigotica at future times. The prediction results of each neural network prediction model are superimposed and reconstructed to obtain a comprehensive prediction result. Based on the comprehensive prediction result, irrigation and fertilization control schemes are formulated to realize environmental monitoring of Isatis indigotica planting.

[0122] Specifically, key factors influencing Isatis indigotica growth are divided into training and testing sets, and principal component analysis (PCA) is used to decompose these key factors. Multi-dimensional Isatis indigotica growth characteristics are extracted using PCA, and these characteristics are used to construct multiple neural network prediction models. Each model is trained based on different growth characteristics, ensuring model diversity and accuracy. Parameter optimization algorithms are used to optimize the parameters of each neural network prediction model, adjusting the parameters to improve the model's predictive ability. Multiple optimized neural network prediction models are obtained, ensuring they can accurately predict the future growth needs of Isatis indigotica. These optimized neural network models are then used to predict future growth needs. The prediction results of each model are superimposed and reconstructed to obtain a comprehensive prediction result. Based on this result, reasonable irrigation and fertilization control schemes are formulated to ensure that Isatis indigotica receives optimal environmental conditions in future growth stages. The implementation effects of these control schemes are tracked and fed back in real time through an environmental monitoring system, further optimizing management measures. This process enables precise adjustment of the Isatis indigotica growth environment, achieving intelligent agricultural management.

[0123] In this optional embodiment, the parameters of multiple neural network prediction models are optimized using a parameter optimization algorithm to obtain multiple optimized neural network prediction models, including:

[0124] Set the parameters and maximum number of iterations for the parameter optimization algorithm, and determine the optimization objective function and its constraints for multiple neural network prediction models;

[0125] Initialize multiple individual neural network prediction models, each representing a set of parameter configurations for a neural network prediction model, and generate the initial neural network prediction model structure;

[0126] Based on the parameter configuration of the current neural network prediction model, train each neural network prediction model and evaluate its performance, record the current global best neural network prediction model and its corresponding parameter configuration, and update the global best neural network prediction model through random search.

[0127] Specifically, based on the parameter configuration of the current neural network prediction model, each neural network prediction model is trained and its performance is evaluated. The current globally optimal neural network prediction model and its corresponding parameter configuration are recorded. The globally optimal neural network prediction model is updated through random search, including:

[0128] Based on the parameter configuration of the current neural network prediction model, the neural network prediction model is trained, and the performance index of the neural network prediction model is calculated as the basis for fitness evaluation.

[0129] After training is complete, record the best-performing neural network prediction model and its corresponding parameter configuration.

[0130] If the current best neural network prediction model exceeds the existing global optimal solution, then replace the record and update the parameter configuration of the neural network prediction model to the new global optimal solution.

[0131] While updating the global optimum, a new parameter space is explored using a random search method to avoid local optima;

[0132] The process of repeated training, evaluation, and search is used to continuously update the globally optimal neural network prediction model and its parameter configuration.

[0133] If the fitness of the current neural network prediction model is greater than that of the best neural network prediction model in the previous iteration, the parameter configuration is updated by a forward operation; otherwise, a new parameter space is explored by a roll operation.

[0134] If the maximum number of iterations is reached, a breeding operation is performed to generate a new generation of neural network prediction model individuals and update the parameter configuration. The parameter configuration of the optimal neural network prediction model is output, and the next round of optimization is carried out.

[0135] After the optimization process is completed, multiple optimized neural network prediction models and their optimal parameter configurations are output.

[0136] Specifically, the parameters of the parameter optimization algorithm are set, such as the maximum number of iterations, and the optimization objective function and constraints of the neural network prediction model are defined. These objective functions are used to evaluate the performance of each neural network prediction model to ensure its accurate prediction of the growth requirements of Isatis indigotica. Multiple neural network prediction models are initialized, each representing a set of parameter configurations, and an initial neural network structure is generated. Based on the current parameter configuration, each neural network model is trained and its performance is evaluated, recording the best-performing neural network model and its corresponding parameter configuration. Through random search, the optimization algorithm continuously adjusts the parameters and updates the globally optimal neural network prediction model. If the fitness of the current model is higher than the optimal model of the previous iteration, the parameter configuration is optimized through a forward operation; if the fitness is lower, new possibilities are explored in the parameter space through a roll operation. When the maximum number of iterations is reached, a breeding operation is performed to generate a new generation of neural network models, and their parameter configurations are updated to continue the next round of optimization. After multiple iterations, multiple optimized neural network prediction models and their optimal parameter configurations are output, ensuring that each model can accurately predict the growth requirements of Isatis indigotica.

[0137] Specifically, the parameter optimization algorithm is a bacterial colony optimization algorithm, a heuristic algorithm that simulates the process of bacterial communities searching for food. Through communication and competition among individual bacteria, the algorithm can find the optimal solution in the search space. In this invention, this algorithm is used to optimize the parameters of a neural network prediction model. By initializing and training multiple neural network individuals, evaluating their fitness, and continuously updating the parameter configuration using forward, tumbling, and reproduction operations, the optimal neural network model parameters are finally obtained, improving prediction performance.

[0138] like Figure 2 As shown, according to another embodiment of the present invention, a neural network-based monitoring system for the planting environment of Isatis indigotica is also provided, the system comprising:

[0139] Data acquisition module 1 is used to acquire environmental data of Isatis indigotica cultivation, and to preprocess the environmental data to extract environmental feature data.

[0140] Anomaly analysis module 2 is used to traverse all environmental feature data, find neighboring environmental feature data of the target environmental feature data, compare the difference between the target environmental feature data and the neighboring environmental feature data, and generate new target environmental feature data as the key factors affecting the growth of Isatis indigotica based on the comparison results.

[0141] The model prediction and regulation module 3 is used to construct a neural network prediction model using key factors, and to use the neural network prediction model to predict the growth requirements of Isatis indigotica at future times, generate irrigation and fertilization regulation schemes, and realize environmental monitoring of Isatis indigotica planting.

[0142] To facilitate understanding of the above technical solutions of the present invention, the following provides a detailed description of the monitoring of the planting environment of Isatis indigotica based on neural networks in actual practice.

[0143] I. Environmental data is acquired through IoT sensors, remote sensing and drone monitoring, and meteorological station data, and preprocessed to extract environmental characteristic data that affect the growth of Isatis indigotica.

[0144] 1) IoT sensor monitoring:

[0145] Temperature sensor: measures the temperature of soil and air.

[0146] Humidity sensor: measures the moisture content of the soil.

[0147] Light sensor: measures light intensity per hour.

[0148] pH sensor: measures the acidity or alkalinity of soil.

[0149] 2) Remote sensing and UAV monitoring: Information such as vegetation index (NDVI) and vegetation coverage is obtained through remote sensing technology and UAV photography to analyze the growth status of Isatis indigotica.

[0150] 3) Meteorological station data: macro-meteorological data such as temperature, humidity, precipitation, and wind speed.

[0151] 4) Specific examples of environmental data are shown in Table 1:

[0152] Table 1. Specific examples of environmental data

[0153] 5) Environmental feature data extraction:

[0154] Temperature characteristics: The daily average temperature and diurnal temperature range are calculated by collecting air and soil temperature data.

[0155] Moisture characteristics: Based on soil moisture, precipitation, and irrigation, data such as soil moisture content and evaporation are extracted.

[0156] Nutrient characteristics: The content of nitrogen, phosphorus, potassium and trace elements was extracted through soil analysis.

[0157] Illumination characteristics: Calculate the hourly effective radiant radiation (PAR) value using illuminance data.

[0158] Second, using anomaly analysis algorithms, the differences between target environmental feature data and neighboring environmental feature data are compared, and new target environmental feature data are generated through weighting coefficients to ultimately determine the key factors affecting the growth of Isatis indigotica.

[0159] 1) Target environmental characteristic data includes:

[0160] Temperature: For example, 22°C.

[0161] Humidity and moisture: For example, soil moisture is 25%.

[0162] Light intensity: for example, 200 μmol / m² / s.

[0163] Soil characteristics: For example, the soil pH value is 6.5.

[0164] Nutrient supply: For example, the soil nitrogen content is 30 mg / kg.

[0165] Air quality and CO2 concentration: For example, the carbon dioxide concentration is 380 ppm.

[0166] Atmospheric pressure and wind speed: for example, wind speed is 3.2 m / s.

[0167] Irrigation amount: For example, the irrigation amount is 10mm.

[0168] 2) Traverse all environmental feature data and find records that are similar to the target environmental feature data. For example, if environmental feature data such as temperature, humidity, and soil pH are selected as the target, search for nearby records and calculate the differences between them and the target data.

[0169] For example, the target data is:

[0170] Temperature: 22°C.

[0171] Humidity: 25%.

[0172] Light intensity: 200 μmol / m² / s.

[0173] Neighboring data 1:

[0174] Temperature: 23°C.

[0175] Humidity: 26%.

[0176] Light intensity: 210 μmol / m² / s.

[0177] The differences are:

[0178] Temperature difference = |22-23| = 1°C.

[0179] Humidity difference = |25-26| = 1%.

[0180] Difference in light intensity = |200-210| = 10 μmol / m² / s.

[0181] Neighboring data 2:

[0182] Temperature: 20°C.

[0183] Humidity: 24%.

[0184] Light intensity: 195 μmol / m² / s.

[0185] The differences are:

[0186] Temperature difference = |22-20| = 2°C.

[0187] Humidity difference = |25-24| = 1%.

[0188] Difference in light intensity = |200-195| = ​​5 μmol / m² / s.

[0189] 3) By calculating these differences, weighting coefficients are applied to update the target data. For example, large temperature differences may affect growth, while small light differences have a smaller impact. Taking all differences into account, new target environmental characteristic data are generated.

[0190] Third, use neural networks to predict the future growth needs of Isatis indigotica, and then formulate reasonable irrigation and fertilization plans.

[0191] 1) Constructing a neural network model:

[0192] Input data: Environmental characteristics (e.g., temperature, humidity, light intensity, soil nutrients, etc.)

[0193] Input features, for example:

[0194] Temperature = 22°C.

[0195] Humidity = 25%.

[0196] Light intensity = 200 μmol / m² / s.

[0197] Soil pH=6.5.

[0198] Irrigation amount = 10mm.

[0199] Neural network architecture:

[0200] Input layer: Receives the above environmental feature data.

[0201] Hidden layer: Contains multiple neurons for data processing and feature extraction.

[0202] Output layer: Outputs the predicted irrigation and fertilization requirements.

[0203] 2) Use historical data (such as environmental data and actual fertilization and irrigation amounts over the past three months) to train the neural network, optimize model parameters, and ensure accurate prediction results. Employ algorithms such as backpropagation for weight adjustment and model training.

[0204] 3) Generate irrigation and fertilization control plans:

[0205] Prediction results: The neural network model predicts the growth requirements of Isatis indigotica in the coming week, recommending an irrigation amount of 15 mm and a fertilizer application amount of 20 g / m².

[0206] Regulation plan:

[0207] Irrigation control: Based on the predicted water demand, develop an irrigation plan and irrigate every two days to ensure that the soil moisture is maintained within the optimal range (20%-30%).

[0208] Fertilizer regulation: Based on the predicted soil nutrient levels, fertilize once every two weeks using a compound fertilizer containing nitrogen, phosphorus, and potassium at a rate of 20g / m².

[0209] Furthermore, in another specific embodiment, after the execution of the generated irrigation and fertilization control scheme, the environmental response data is sampled in stages based on the set initial sampling window and delayed sampling window. Specifically, the first environmental response dataset is collected in the initial sampling window after the execution of the irrigation and fertilization control scheme, and the first environmental response state parameters are analyzed. The second environmental response dataset is collected in the delayed sampling window, and the second environmental response state parameters are analyzed.

[0210] The environmental response dataset includes soil moisture, leaf photosynthetic reflectance, and soil electrical conductivity at each sampling point within the sampling window. Environmental response state parameters are calculated, specifically including mean soil moisture, standard deviation of soil moisture, mean leaf photosynthetic reflectance, standard deviation of leaf photosynthetic reflectance, mean soil electrical conductivity, and standard deviation of soil electrical conductivity.

[0211] Soil moisture, leaf photosynthetic reflectance, and soil electrical conductivity are combined as environmental response parameters.

[0212] It should be explained that soil moisture is used to characterize the change in soil water content after the implementation of irrigation and fertilization control programs. It is defined as the proportion of water in a unit volume of soil, expressed as volumetric water content (%). The system collects data in real time using soil moisture sensors deployed at different depths in the planting area. Leaf photosynthetic reflectance is used to characterize the photosynthetic intensity and water and nitrogen status of plant leaves after the implementation of irrigation and fertilization control programs. It is defined as the ratio of reflected light intensity to incident light intensity in a specific wavelength band (visible and near-infrared bands). The system acquires spectral reflectance signals through multispectral reflectance sensors deployed above the canopy, calculates the average reflectance and its standard deviation after band normalization, and uses this to reflect the trend of plant photosynthetic activity and leaf area changes. Soil electrical conductivity is used to reflect the ion concentration and fertilizer dissolution state in the soil solution after the implementation of irrigation and fertilization control programs. It is defined as the current conduction capacity of the soil solution per unit distance, expressed as millisiemens per centimeter (mS / cm). The system measures the electrical conductivity of the soil solution using a conductivity probe installed in the root zone, and combines the mean and standard deviation from multiple sampling points to characterize changes in nutrient concentration and fertilization effectiveness.

[0213] It should be noted that the first and second environmental response datasets involve the same collected data, differing only in the sampling time.

[0214] The environmental response state stability is calculated based on the first environmental response state parameter and the second environmental response state parameter, and is specifically defined as follows: , S represents the environmental response state stability. X For the stability of the Xth environmental response object, Let X be the mean of the first environmental response object. The mean of the Xth second environmental response object. Let X be the standard deviation of the first environmental response object. Let X be the standard deviation of the Xth second environmental response object, where X = 1, 2, 3, X = 1 represents soil moisture, X = 2 represents leaf photosynthetic reflectance, and X = 3 represents soil electrical conductivity.

[0215] It should be explained that the difference between the mean of the second environmental response state parameter and the mean of the first environmental response state parameter is used to characterize the overall trend of change of the corresponding parameter. When the difference in mean is large, it indicates that the environment is still in the transition stage after regulation. The difference in standard deviation is used to characterize the change in the degree of data fluctuation. When the fluctuation amplitude is still large, it indicates that the system has not yet reached a steady state. Therefore, by normalizing the combination of the difference in mean and the difference in standard deviation, the drift trend and stability of the environmental response can be captured simultaneously, thus avoiding misjudgment based solely on a single sample value or instantaneous change. In other words, the design of the stability formula for the environmental response state parameter enables the system to comprehensively judge whether the response after regulation has tended to stabilize. It can reflect both the parameter change trend caused by irrigation or fertilization and the fluctuation convergence characteristics of each parameter within the delay window, thus providing a reliable basis for judgment in subsequent data analysis and adjustment stages.

[0216] If the stability of the environmental response state is less than or equal to the preset environmental response state stability threshold in the database, a stable signal is generated.

[0217] If the environmental response stability is greater than the preset environmental response stability threshold in the database, an extended sampling strategy is executed. Specifically, the environmental response stability threshold is subtracted from the environmental response stability to obtain the environmental response stability deviation. A pre-established mapping table or index between the environmental response stability deviation and the extended sampling time value is used to quickly match and extract the extended sampling time value. Based on the extended sampling time value, the environmental response parameters are re-acquired and designated as the third environmental response parameters. The environmental response stability between the second and third environmental response parameters is then recalculated, and a comparison is made between the environmental response stability and the environmental response stability threshold. If the environmental response stability is still greater than the preset environmental response stability threshold in the database, the extended sampling strategy is re-executed until the calculated environmental response stability is less than the environmental response stability threshold, generating a stable signal.

[0218] It should also be noted that the larger the environmental response state stability deviation value, the more significant the difference between the current environmental response state and the system's preset stable state, indicating that a stable state has not yet been reached. To ensure that subsequent control or analysis is based on reliable environmental response data, the system needs to extend the sampling time to allow the environmental response state to stabilize. Correspondingly, the sampling time extension value extracted from the database will increase with the increase of the environmental response state stability deviation value, allowing the strategy of extending the sampling time to be adaptively adjusted according to the magnitude of the deviation, thereby effectively improving the accuracy and reliability of the environmental response data.

[0219] The preset environmental response stability threshold is determined based on statistical analysis of historical environmental response data and crop growth experience. Specifically, the system normalizes and calculates the stability of soil moisture, leaf photosynthetic reflectance, and soil electrical conductivity collected over multiple growth cycles to obtain the typical fluctuation range of each parameter under steady-state conditions. These steady-state fluctuation ranges are then weighted and aggregated to generate a reference environmental response stability value. A safety margin is then set based on this value to form the preset environmental response stability threshold in the database.

[0220] The preset principle of the following response deviation threshold is the same as that of the preset environmental response state stability threshold, so it will not be repeated here.

[0221] After receiving a stable signal, environmental response parameters are re-collected at a preset sampling period and recorded as stable environmental response parameters, including mean soil moisture, mean leaf photosynthetic reflectance, and mean soil electrical conductivity. The target average environmental response parameters corresponding to the irrigation and fertilization control schemes stored in the database are extracted, including the target mean soil moisture, target mean leaf photosynthetic reflectance, and target mean soil electrical conductivity. Based on these parameters, the comprehensive response deviation factor is calculated, defined as follows: , R C To determine the overall response deviation factor, D X w is the deviation factor for the Xth stable environment response object. X The weight of the Xth stable environment response object. Let X be the mean value of the response object in the stable environment. Let X be the mean value of the b-th target environmental response object, where X=1, 2, 3, X=1 represents soil moisture, X=2 represents leaf photosynthetic reflectance, and X=3 represents soil electrical conductivity.

[0222] If the overall response deviation factor is less than the preset response deviation threshold in the database, it means that the irrigation and fertilization control scheme has achieved the expected results. No further processing is required, and the above calculated data is uploaded.

[0223] If the overall response deviation factor is greater than or equal to the preset response deviation threshold in the database, feedback optimization of the neural network prediction model is performed to obtain an activation sensitivity parameter set. This activation sensitivity parameter set consists of three independent activation sensitivity factors (soil moisture activation sensitivity factor, leaf photosynthetic reflectance activation sensitivity factor, and soil electrical conductivity activation sensitivity factor), used to control the response amplitude of the corresponding feature subspaces. Specifically, taking the acquisition of the soil moisture activation sensitivity factor as an example, based on the soil moisture deviation factor in the deviation factors of the calculated stable environmental response state parameters, it is mapped to the activation sensitivity adjustment value corresponding to the moisture response cluster to obtain the current soil moisture execution activation sensitivity factor. The activation sensitivity adjustment value is added to the current soil moisture execution activation sensitivity factor to obtain the soil moisture activation sensitivity factor. The above steps are repeated to obtain the leaf photosynthetic reflectance activation sensitivity factor and the soil electrical conductivity activation sensitivity factor, respectively.

[0224] It should be noted that after calculating the soil moisture deviation factor, since it is obtained by subtracting the target value (i.e., the target soil moisture mean) from the measured value in the steady state (i.e., the mean of the steady soil moisture), its sign reflects the direction of the humidity response deviation: when the soil moisture deviation factor is positive, it indicates that the currently collected steady soil moisture is higher than the target moisture, meaning the system's actual humidity response is too strong; when the soil moisture deviation factor is negative, it indicates that the current steady soil moisture is lower than the target moisture, meaning the system's actual humidity response is too weak. Therefore, when mapping the deviation factor to the activation sensitivity adjustment value corresponding to the humidity response cluster, the system follows the following directional rule: if the soil moisture deviation factor is positive, it indicates that the model's response to humidity changes is too strong, and the corresponding activation sensitivity needs to be appropriately reduced to decrease the model's amplification of the humidity input, thereby suppressing the tendency of the output prediction to be overly sensitive to humidity; if the soil moisture deviation factor is negative, it indicates that the model's response to humidity changes is insufficient, and the corresponding activation sensitivity needs to be appropriately increased to improve the model's response amplitude to the humidity input, making the model output more sensitive to humidity changes. Through this directional mapping rule, the system can automatically select the increasing or decreasing trend of activation sensitivity according to the actual deviation direction of soil moisture, so as to realize the adaptive adjustment of the model's response to humidity input: suppress the response when the humidity is too high and enhance the response when the humidity is too low, so that the model's predicted output gradually approaches the target reference state.

[0225] It's important to clarify that when the environment is stable but prediction bias still exists, the problem is often that the neural network prediction model's response to certain input features is incorrect (too strong or too weak), rather than the feature combination relationship being completely wrong. Adjusting the activation sensitivity can amplify or suppress the model's influence on feature inputs, thereby adjusting the output response without changing the higher-order feature representations already learned by the network.

[0226] It should also be noted that the above embodiments use soil moisture, leaf photosynthetic reflectance, and soil electrical conductivity as analysis data. Therefore, the corresponding adjustment is to the activation sensitivity factors of the three feature subspaces of soil moisture, leaf photosynthetic reflectance, and soil electrical conductivity. In actual implementation, the analysis data can be flexibly adjusted based on actual needs. This invention does not impose any specific limitations.

[0227] In addition, this application also provides an electronic device. For example... Figure 3 The diagram illustrates the hardware operating environment of an electronic device, which may include: a processor (e.g., CPU), memory, a user interface, a network interface, and a communication bus. The communication bus is used to enable communication between components. The user interface may include a display screen and an input unit such as a keyboard; optionally, the user interface may also include a standard wired interface or a wireless interface. The network interface may optionally include a standard wired interface or a wireless interface. The memory may be high-speed RAM or stable non-volatile memory, such as disk storage. Alternatively, the memory may be a storage device independent of the aforementioned processor.

[0228] Those skilled in the art will understand that Figure 3 The electronic devices shown do not constitute a limitation on electronic devices and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0229] like Figure 3 As shown, a memory, as a type of computer storage medium, may include an operating system, a network communication module, a user interface module, and device management programs. The operating system is a program that manages and controls the hardware and software resources of electronic devices, supporting the operation of electronic devices and other software or programs. Figure 3 In the electronic device shown, the user interface is mainly used to connect to the terminal and communicate with the terminal, such as receiving user signaling data sent by the terminal; the network interface is mainly used to communicate with the backend server; the processor can be used to call the program stored in the memory and execute the steps of the method or system described above.

[0230] Furthermore, embodiments of this application also propose a computer-readable storage medium storing a device management program, which, when executed by a processor, implements the steps of the method or system described above.

[0231] The specific implementation of the computer-readable storage medium in this application is basically the same as the embodiments of the methods or systems described above, and will not be repeated here. Furthermore, to achieve the above objectives, this application also provides a computer program product, comprising: a computer program, which, when executed by a processor, implements the steps of the methods or systems described above.

[0232] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for monitoring the planting environment of Isatis indigotica based on neural networks, characterized in that, include: Environmental data on Isatis indigotica cultivation was obtained, and the environmental data was preprocessed to extract environmental feature data. Traverse all environmental feature data, find neighboring environmental feature data of the target environmental feature data, compare the difference between the target environmental feature data and the neighboring environmental feature data, and based on the comparison results, generate new target environmental feature data as the key factor affecting the growth of Isatis indigotica. A neural network prediction model was constructed using key factors, and this model was used to predict the future growth needs of Isatis indigotica, generating irrigation and fertilization control schemes to achieve environmental monitoring of Isatis indigotica cultivation.

2. The method for monitoring the planting environment of Isatis indigotica based on neural networks according to claim 1, characterized in that, The process of traversing all environmental feature data, finding neighboring environmental feature data of the target environmental feature data, comparing the differences between the target environmental feature data and the neighboring environmental feature data, and generating new target environmental feature data based on the comparison results as key factors affecting the growth of Isatis indigotica includes: Initialize the parameters and maximum number of iterations of the anomaly analysis algorithm, and set the starting point of the anomaly analysis as the initial environmental characteristics of Isatis indigotica growth, with the goal of identifying key factors affecting growth. The search algorithm is used to traverse all environmental feature data, find the neighboring environmental feature data of the target environmental feature data, and check whether the difference between the neighboring environmental feature data and the target environmental feature data exceeds the preset value. If the difference between the neighboring environmental feature data and the target environmental feature data is greater than the preset value, a weighting coefficient is introduced to generate new target environmental feature data; otherwise, the preset value is adjusted and new target environmental feature data is generated by combining the new weighting coefficient. Check for anomalies in the correlation between the new target environmental feature data and the neighboring environmental feature data. If no anomalies are found, add the new target environmental feature data to the analysis model. If anomalies are found, readjust the preset values ​​and weight coefficients to update the target environmental feature data. Determine if the maximum number of iterations has been reached. If it has, output the optimal target environment characteristic data as the key factor affecting the growth of Isatis indigotica; otherwise, continue iterative analysis.

3. The method for monitoring the planting environment of Isatis indigotica based on neural networks according to claim 2, characterized in that, The step of using a search algorithm to traverse all environmental feature data, finding neighboring environmental feature data of the target environmental feature data, and checking whether the difference between the neighboring environmental feature data and the target environmental feature data exceeds a preset value includes: Initialize the parameters of the search algorithm and the maximum number of iterations; Starting with the target environment feature data, construct the environment search structure and initialize the expandable state set, and begin to filter and expand the paths that match the target environment feature data from the candidate environment feature set; Using a screening algorithm, environmental feature data that best matches the combination of target environmental features is selected from the current candidate environmental feature set for expansion, and the optimal expansion path matching the target environmental feature data is gradually established. Recursively generate other candidate environmental feature data for the current state and extend them to the remaining branches of the previous state to form a diverse neighboring environmental feature data space. The differences between the environmental feature data in each extended path and the target environmental feature data are evaluated. If the difference exceeds a preset threshold, conflict information is generated for further screening. Based on the conflict information, environmental feature data that does not meet the preset conditions are eliminated, the scalable candidate environmental feature data set is updated, and the search continues until the best neighboring environmental feature data is found.

4. The method for monitoring the planting environment of Isatis indigotica based on neural networks according to claim 3, characterized in that, The step of using a screening algorithm to select environmental feature data that best matches the target environmental feature combination from the current candidate environmental feature set for expansion, and gradually establishing the optimal expansion path for matching the target environmental feature data, includes: Set the parameters and maximum number of iterations for the sieving algorithm, and initialize the search state set; Based on the differences between candidate environmental feature data and target environmental feature data, the fitness value of each candidate environmental feature data is calculated as the basis for selecting expansion paths; Calculate the search space size of the current candidate environment feature data according to the preset formula, and determine the scalable data volume; All candidate environmental feature data are sorted in a non-dominated manner, and then divided into primary and secondary data groups. The primary data group with the smallest difference from the target environmental feature data is selected for expansion. Calculate the fitness value of the newly generated candidate environment feature data and compare it with the parent data to select the optimal expansion path. Update the scalable candidate environment feature data set. If the maximum number of iterations is reached, terminate the search and output the optimal expansion path; otherwise, continue iterating.

5. The method for monitoring the planting environment of Isatis indigotica based on neural networks according to claim 4, characterized in that, The step of performing non-dominated sorting on all candidate environmental feature data and dividing all candidate environmental feature data into primary and secondary data groups, and prioritizing the expansion of the primary data group with the smallest difference from the target environmental feature data, includes: Fitness assessment is performed on all candidate environmental feature data, and the difference between each candidate environmental feature data and the target environmental feature data is calculated as the basis for ranking and grouping. Based on the fitness values, the candidate environmental feature data are non-dominatedly sorted and divided into primary data group and secondary data group. Select candidate environmental feature data that has the smallest difference from the target environmental feature data from the main data set, and use it as the starting point for the expansion path to ensure the matching of the expansion with the target. The selected main data sets are expanded to generate new candidate environmental feature data, and the search continues to gradually approach the target environmental feature data.

6. The method for monitoring the planting environment of Isatis indigotica based on neural networks according to claim 1, characterized in that, The process of constructing a neural network prediction model using key factors and using this model to predict the future growth needs of Isatis indigotica, generating irrigation and fertilization control plans, and achieving environmental monitoring for Isatis indigotica cultivation includes: The key factors were divided into training and testing sets. Principal component analysis was used to decompose the key factors, extract multi-dimensional growth characteristics of Isatis indigotica, and establish multiple neural network prediction models based on each type of Isatis indigotica growth characteristics. The parameters of multiple neural network prediction models are optimized using a parameter optimization algorithm to obtain multiple optimized neural network prediction models; Multiple optimized neural network prediction models are used to predict the growth requirements of Isatis indigotica at future times. The prediction results of each neural network prediction model are superimposed and reconstructed to obtain a comprehensive prediction result. Based on the comprehensive prediction result, irrigation and fertilization control schemes are formulated to realize environmental monitoring of Isatis indigotica planting.

7. The method for monitoring the planting environment of Isatis indigotica based on neural networks according to claim 6, characterized in that, The process of optimizing the parameters of multiple neural network prediction models using a parameter optimization algorithm to obtain multiple optimized neural network prediction models includes: Set the parameters and maximum number of iterations for the parameter optimization algorithm, and determine the optimization objective function and its constraints for multiple neural network prediction models; Initialize multiple individual neural network prediction models, each representing a set of parameter configurations for a neural network prediction model, and generate the initial neural network prediction model structure; Based on the parameter configuration of the current neural network prediction model, train each neural network prediction model and evaluate its performance, record the current global best neural network prediction model and its corresponding parameter configuration, and update the global best neural network prediction model through random search. If the fitness of the current neural network prediction model is greater than that of the best neural network prediction model in the previous iteration, the parameter configuration is updated by a forward operation; otherwise, a new parameter space is explored by a roll operation. If the maximum number of iterations is reached, a breeding operation is performed to generate a new generation of neural network prediction model individuals and update the parameter configuration. The parameter configuration of the optimal neural network prediction model is output, and the next round of optimization is carried out. After the optimization process is completed, multiple optimized neural network prediction models and their optimal parameter configurations are output.

8. A neural network-based monitoring system for the planting environment of Isatis indigotica, used to implement the neural network-based monitoring method for the planting environment of Isatis indigotica according to any one of claims 1-7, characterized in that, The system includes: The data acquisition module is used to acquire environmental data of Isatis indigotica cultivation, and to preprocess the environmental data to extract environmental feature data. The anomaly analysis module is used to traverse all environmental feature data, find neighboring environmental feature data of the target environmental feature data, compare the difference between the target environmental feature data and the neighboring environmental feature data, and generate new target environmental feature data as the key factors affecting the growth of Isatis indigotica based on the comparison results. The model prediction and control module is used to build a neural network prediction model using key factors, and to use the neural network prediction model to predict the growth needs of Isatis indigotica in the future, generate irrigation and fertilization control plans, and realize environmental monitoring of Isatis indigotica planting.

9. An electronic device, characterized in that, The electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When executed by the processor, the computer program implements the steps of the neural network-based monitoring method for Isatis indigotica planting environment as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the neural network-based method for monitoring the planting environment of Isatis indigotica as described in any one of claims 1 to 7.

Images