Intelligent farmland monitoring data management method based on artificial intelligence

Abstract

The invention relates to the technical field of big data management, in particular to a smart farmland monitoring data management method based on artificial intelligence, and the method comprises the steps: obtaining various types of environment monitoring data of each preset farmland region in a smart farmland, and various types of feature monitoring data of each position in each farmland region; obtaining an upper-layer graph structure and a lower-layer graph structure of the double-layer graph structure of the farmland monitoring data; feature vectors of all farmland areas and feature vectors of all kinds of feature monitoring data in all the farmland areas are obtained; iterative optimization is carried out on the obtained feature vectors through a genetic algorithm, and important feature vectors are selected from the feature vectors in the farmland areas in the iteration process; obtaining the fitness value of a single individual during each iteration; updating the crossover rate and the mutation rate of each individual during iteration; and optimizing all the obtained feature vectors, and managing the farmland. The invention aims to improve the accuracy and efficiency of farmland monitoring data management.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of big data management, in particular to a smart farmland monitoring data management method based on artificial intelligence. BACKGROUND

[0002] With the global population growth and the increasingly scarce arable land resources, agricultural production is developing towards high efficiency, precision and intelligence. As a new form of the combination of traditional agriculture and modern information technology, smart agriculture is gradually changing the agricultural production mode. Among them, smart farmland is an important part of smart agriculture, which realizes real-time sensing and intelligent control of farmland environment and crop state through the integration of sensor networks, remote sensing technology, Internet of Things, satellite positioning system and edge computing and other means.

[0003] The smart farmland system usually deploys a large number of monitoring devices such as sensors, weather stations, unmanned aerial vehicles, image acquisition terminals, etc., which can continuously collect multi-source heterogeneous data such as temperature, humidity, soil nutrients, light, images, etc. In the face of multi-source data management, graph neural networks can integrate multi-source data into feature vector graphs for data management, but the multi-source heterogeneous monitoring data of smart farmland has different feature dimensions and distribution characteristics, which leads to inaccurate feature representation of monitoring data by feature vector graphs and easy falling into local optimal feature representation. SUMMARY

[0004] In view of the above, it is necessary to provide a smart farmland monitoring data management method based on artificial intelligence, which improves the accuracy and efficiency of farmland monitoring data management compared with traditional smart farmland monitoring data management methods.

[0005] The smart farmland monitoring data management method based on artificial intelligence provided by the application adopts the following technical scheme: One embodiment of the application provides a smart farmland monitoring data management method based on artificial intelligence, which comprises the following steps: Respectively acquiring various types of environment monitoring data of each farmland region in the smart farmland, and various types of feature monitoring data of each position in each farmland region; Obtaining an upper layer graph structure of a double-layer graph structure of farmland monitoring data through the positional relationship between different farmland regions and the difference in meteorological monitoring data; obtaining a lower layer graph structure of the double-layer graph structure of farmland monitoring data through the similarity between different types of feature monitoring data of each position in each farmland region; Respectively extracting feature vectors of the upper layer graph structure and the lower layer graph structure by using graph neural networks to obtain feature vectors of each farmland region and feature vectors of each type of feature monitoring data in each farmland region; The obtained feature vectors are iteratively optimized by the genetic algorithm, in the iteration process, important feature vectors are selected from the feature vectors in each farmland region through the feature vectors of various feature monitoring data; the individuals in the genetic algorithm are encoded according to the feature vectors of each farmland region and the selected important feature vectors; the fitness value of each individual in each iteration is obtained by combining the similarity between the feature vectors of different farmland regions of each individual in each iteration, the edge weight values between different farmland regions in the upper layer graph structure and the selected important feature vectors in each farmland region; the crossover rate and the mutation rate of each individual in the next iteration are updated by comparing the fitness value of each individual in each iteration with the average level of the fitness values of all individuals, and combining the crossover rate and the mutation rate of each individual in each iteration. The obtained all feature vectors are optimized by the optimized genetic algorithm, and the farmland is managed by the optimized feature vectors.

[0006] In one of the embodiments, the upper layer graph structure of the double-layer graph structure for obtaining farmland monitoring data comprises: The nodes in the upper layer graph structure are single farmland regions, and the calculation method of the edge weight value between any two farmland regions is: all the meteorological monitoring data of each farmland region is composed into a meteorological monitoring data vector, the Euclidean distance between the two farmland regions is calculated, the 2-norm of the difference between the meteorological monitoring data vectors of the two farmland regions is calculated, and the sum of the Euclidean distance and the 2-norm is taken as the edge weight value between the two farmland regions.

[0007] In one of the embodiments, the lower layer graph structure of the double-layer graph structure for obtaining farmland monitoring data comprises: The nodes in the lower layer graph structure are single-class feature monitoring data in the farmland regions corresponding to the nodes in the upper layer graph structure, and the calculation method of the edge weight value between any two types of feature monitoring data is: all the values of each type of feature monitoring data collected in each farmland region are arranged in order according to the position to form a feature monitoring data vector, and the similarity between the feature monitoring data vectors of the two types of feature monitoring data is taken as the edge weight value between the two types of feature monitoring data.

[0008] In one of the embodiments, the selection process of the important feature vectors is: The 2-norm of each feature vector in each farmland region is calculated, all the feature vectors in each farmland region are arranged in descending order according to the 2-norm, and the feature vectors arranged in the first preset number of positions are taken as the important feature vectors.

[0009] In one of the embodiments, the encoding of the individuals in the genetic algorithm comprises: The encoding of each individual C is divided into two parts and is represented as: ;in, The encoding part that represents the upper-level graph structure; The encoding part that represents the underlying graph structure; Specifically: ;in, 、 、 Represent the feature vectors of the first, second, and Nth farmland areas respectively; φ( ) represents the dimensionality reduction function, which is constructed using the fully connected layer; the encoding part of the nth farmland area Specifically: ,in, represents the kth eigenvector in the farmland of the nth region; Represents a set of important feature vectors in the nth farmland area.

[0010] In one embodiment, obtaining the fitness value of a single individual at each iteration includes: For a single individual, the similarity of the feature vectors between two randomly selected farmland areas under the single individual is recorded as feature similarity, and the difference between the edge weights between the two randomly selected farmland areas and the feature similarity is calculated; the mean of the difference between all randomly selected farmland areas under the single individual is calculated; Calculate the average value of the modulus of all important eigenvectors in all farmland areas under a single individual; The fitness value of a single individual can be further obtained through the mean and the average value.

[0011] In one embodiment, the fitness value of the single individual is a weighted sum of the mean and the average value, and the sum of the weight of the mean and the weight of the average value is 1.

[0012] In one embodiment, the update formula of the crossover rate is: ;in, represents the crossover rate of the i-th individual at the t+1-th iteration; represents the crossover rate of the i-th individual at the t-th iteration; represents the smoothing factor of the adaptive crossover rate; represents the fitness value of the i-th individual at the t-th iteration; Represents the mean fitness value of all individuals at the tth iteration; Represents an exponential function with a natural constant as its base.

[0013] In one embodiment, the update formula of the mutation rate is: ; represents the mutation rate of the i-th individual at the t+1-th iteration; represents the mutation rate of the i-th individual at the t-th iteration; Smoothing factor for the adaptive mutation rate.

[0014] In one embodiment, managing the farmland using the optimized feature vector includes: The optimized feature vector is input into the pre-trained SVM classifier, the feature vector is classified and managed, and the status management label of the farmland is output.

[0015] This application has at least the following beneficial effects: In this application, the upper-level graph structure uses farmland areas as nodes, taking into account the geographic spatial relationships between farmland areas and the differences in meteorological monitoring data. This can reflect the overall layout of farmland and the mutual influence between regions from a macro perspective. The lower-level graph structure uses the characteristic monitoring data within each farmland area as nodes, focusing on the relationship between different characteristic data within the area, achieving an organic combination of global and local information, which helps to more accurately understand and analyze farmland monitoring data. Furthermore, graph neural networks can fully utilize the topological information and node feature information in the two-layer graph structure, extract and fuse features from graph structures at different levels, and generate feature vectors with stronger expressive power. Furthermore, selecting important eigenvectors during the iteration process can remove redundant and unimportant feature information, reduce the dimension of the eigenvectors, and reduce the complexity of subsequent calculations, while retaining key features, improving the efficiency and performance of the model, and avoiding the problems of long encoding and optimization difficulties caused by too many eigenvectors; through the improvement of adaptive crossover rate and mutation rate, the genetic algorithm can dynamically adjust the crossover rate and mutation rate according to the individual fitness value, so that the algorithm can maintain a strong exploration ability during the search process, and effectively avoid premature convergence, thereby improving the global search ability and optimization accuracy of the algorithm, and further improving the optimization effect of the eigenvector; the optimized eigenvector can more efficiently reflect the actual situation of the farmland, and improve the accuracy and efficiency of farmland monitoring data management. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the following is a brief introduction to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0017] Figure 1 A flowchart of the steps of a smart farmland monitoring data management method based on artificial intelligence provided in this application; Figure 2 Optimization flowchart of feature vector. DETAILED DESCRIPTION

[0018] In the description of the embodiments of the application, the words "exemplary", "or", "for example" are used to mean as an example, instance, or illustration. Any embodiment or design presented as "exemplary" or "for example" in the embodiments of the application should not be interpreted as being more preferred or advantageous than other embodiments or design solutions. In fact, the use of "exemplary", "or", "for example" is intended to present concepts in a specific way. The use of "exemplary", "or", "for example" is intended to present concepts in a specific way.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the application is for the purpose of describing specific embodiments only and is not intended to be limiting of the application. It should be understood that the use of "or" in the application means either or both. Herein, "comprising" means "including" or "consisting of".

[0020] In addition, it should be pointed out that the terms "first", "second" in the application are used to distinguish similar objects, and are not used to describe a specific order or sequence.

[0021] The specific scheme of the intelligent farmland monitoring data management method based on artificial intelligence provided by the application will be specifically described below in combination with the drawings.

[0022] An intelligent farmland monitoring data management method based on artificial intelligence is provided in an embodiment of the application. Specifically, an intelligent farmland monitoring data management method based on artificial intelligence is provided as follows. Please refer to Figure 1 The method comprises the following steps: Step 1, respectively acquiring various types of environment monitoring data of each preset farmland area in the intelligent farmland, and various types of feature monitoring data of each position in each farmland area.

[0023] The intelligent farmland is evenly divided to obtain each farmland area. Various types of environment monitoring data are acquired through the devices installed at each preset position in each farmland area. Specifically, the environment monitoring data is acquired through the Internet of Things sensor devices in the intelligent farmland, including temperature, humidity, light intensity, carbon dioxide concentration, and soil moisture content.

[0024] Various types of weather monitoring data are acquired through a weather station, including temperature, rainfall, and wind speed.

[0025] The RGB image data of each farmland area is acquired through the unmanned aerial vehicle carrying a high-definition camera. The environment monitoring data and the RGB image data are collectively referred to as feature monitoring data.

[0026] Obtain geographical information (GIS) data of each farmland area through satellite remote sensing technology.

[0027] Data cleaning is performed on the environmental monitoring data and the meteorological monitoring data, and an outlier is identified and removed using a standard deviation method. The standard deviation method is a known technique, and will not be described herein again. After that, the cleaned data is normalized to avoid the influence of the dimension.

[0028] In this embodiment, the Min-Max normalization method is used to normalize various environmental monitoring data and various meteorological monitoring data. The Min-Max normalization method is a known technique, and will not be described herein again.

[0029] Radiometric correction and geometric correction are performed on the obtained RGB image to improve the quality of the image. The radiometric correction and geometric correction of the RGB image are known techniques, and will not be described herein again.

[0030] The preprocessing of the GIS data mainly includes coordinate system unification, attribute cleaning and topology repair, etc. The purpose is to ensure spatial consistency, boundary information closure and avoid data record duplication. The preprocessing process of the GIS data is a known technique, and will not be described herein again.

[0031] Step 2, a double-layer graph structure of farmland monitoring data is constructed, and a graph neural network is used to extract feature vectors of the upper-layer graph structure and the lower-layer graph structure, respectively. Genetic algorithm is used to iteratively optimize the obtained feature vectors.

[0032] Step 2.1, the upper-layer graph structure of the double-layer graph structure of farmland monitoring data is obtained through the positional relationship between different farmland areas and the difference in meteorological monitoring data. The lower-layer graph structure of the double-layer graph structure of farmland monitoring data is obtained through the similarity between different feature monitoring data of each position in each farmland area.

[0033] In the management of the monitoring data of the smart farmland, it is necessary to obtain various monitoring data of the whole farmland for farmland state analysis, and then implement management. However, the current smart farmland is large in area, and the monitoring equipment used in the farmland can only monitor the state change of a part of the area in real time. Using a graph neural network (GNN) based on artificial intelligence to model the monitoring data in a certain area and representing each item of monitoring data in the area as a feature vector can realize the management of the monitoring data in the area. In order to more efficiently manage the overall monitoring data of the smart farmland, it is necessary to integrate the monitoring data of each area and construct a comprehensive farmland monitoring data relationship graph. Specifically, a double-layer graph structure can be used to organize these data, and through the comprehensive global information, efficient management and analysis of the farmland monitoring data can be realized.

[0034] Specifically, taking a single farmland area as a node, taking the geographical spatial relationship between farmland areas and the difference of meteorological monitoring data as an edge weight value, an upper-layer graph structure of the entire farmland is constructed. Taking each farmland area as a unit, a lower-layer graph structure is constructed. In the lower-layer graph structure, each type of feature monitoring data is taken as a node, and the relationship between the feature monitoring data is taken as an edge weight value.

[0035] However, the overall graph structure of farmland monitoring data has complex information, contains a large amount of monitoring data of farmland areas, and the data has the characteristics of multi-source heterogeneity. The environmental monitoring data and various meteorological monitoring data are low-dimensional numerical data, and the RGB image is high-dimensional image data. These data formats differ greatly, and it is necessary to map these different modal data to the same graph space for more accurate feature representation, improve the expression ability of the double-layer graph structure to multi-source data, and thus improve the efficient management of the monitoring data.

[0036] To solve the above problems, the genetic algorithm improved by the adaptive crossover rate and mutation rate is used to optimize the farmland monitoring data feature vector output by the GNN, improve the representation ability of the model to the overall farmland monitoring data, and facilitate the model to perform efficient classification management through the optimized farmland monitoring data feature vector.

[0037] The upper-layer graph structure of the entire farmland can be obtained by the construction strategy of the double-layer graph structure, which is represented as: wherein, represents the nth farmland area; represents the edge weight value between the nth and the mth farmland area, and the specific calculation formula is: ; in the formula, represents the Euclidean distance between the nth and the mth farmland area; , represents the meteorological monitoring data vector composed of all meteorological monitoring data of the nth and the mth farmland area, respectively; represents the calculation of the 2-norm operation.

[0038] Each node in the upper-layer graph contains its own lower-layer graph structure. The lower-layer graph structure is a graph structure composed of various types of feature monitoring data and their relationships in the farmland area corresponding to each node. The lower-layer graph structure of the nth area is represented as wherein, represents the ith type of feature monitoring data in the nth farmland area; The edge weight value between the i-th and j-th feature monitoring data in the n-th farmland area is represented, and the specific calculation method is as follows: taking temperature and humidity as low-dimensional numerical data, when the temperature in a farmland area is too high, it will cause the evaporation of water in the area, directly leading to the decrease of humidity. All the temperature and humidity data collected in the n-th farmland area are arranged in order according to the acquisition position to form a temperature data vector and a humidity data vector. The cosine similarity between the temperature data vector and the humidity data vector is taken as the edge weight value between the temperature and the humidity in the n-th farmland area. Taking humidity and RGB image as an example, when the humidity is too low for a long time, it will cause the color of the soil and crops in the image to change, such as dry soil and yellowing crops showing higher pixel values. According to the number of humidity monitoring devices in the n-th farmland area, the RGB image of the n-th farmland area is divided into small blocks, and the humidity in each small block is obtained by the same humidity monitoring device. The number of small blocks obtained by dividing the RGB image of the n-th farmland area is equal to the number of humidity monitoring devices in the n-th farmland area. The average pixel value of all pixel points in each small block of the RGB image of the n-th farmland area in R, G and B channels is calculated. The normalized value of the average pixel value of all small blocks of the RGB image of the n-th farmland area and the humidity are arranged in order according to the acquisition position to form an image data vector and a humidity vector of the n-th farmland area. The cosine similarity between the image data vector and the humidity vector is taken as the edge weight value between the RGB image and the humidity in the n-th farmland area. The temperature data vector, the humidity data vector, the image data vector and the humidity vector are collectively referred to as feature monitoring data vectors.

[0039] In this embodiment, the Min-Max normalization method is used to obtain the normalized value of the average pixel value.

[0040] Step 2.2, the graph neural network is used to extract the feature vectors of the upper layer graph structure and the lower layer graph structure respectively, and the feature vectors of each farmland area and the feature vectors of each type of feature monitoring data in each farmland area are obtained.

[0041] Further, for the constructed double-layer graph structure of farmland monitoring data, GNN is used to extract the feature vectors of the upper layer graph structure and the lower layer graph structure respectively. The feature vector of the upper layer graph structure can be represented as wherein, , , respectively represent the feature vectors of the first, second and N-th farmland areas. The feature vector of the n-th farmland area can be represented as wherein, , , respectively represent the feature vectors of the first type, the second type, and the I type feature monitoring data. Among them, the feature vectors extracted by GNN are known technologies, and this application will not be repeated.

[0042] It should be noted that: the monitoring data contains low-dimensional numerical type data and high-dimensional image type data, different feature extraction methods are adopted for different types of data, low-dimensional numerical data can be mapped by a fully connected layer network structure, and high-dimensional image data can be extracted by a convolutional neural network. Among them, the fully connected layer and the convolutional neural network are existing known technologies, and the specific feature extraction process will not be repeated.

[0043] Step 2.3, the obtained feature vectors are iteratively optimized by a genetic algorithm, in the iteration process, the individuals in the genetic algorithm are coded; the fitness value of each individual at each iteration is obtained; the crossover rate and mutation rate of each individual at each iteration are updated; and the obtained all feature vectors are optimized by the optimized genetic algorithm.

[0044] The double-layer graph structure of the constructed farmland monitoring data contains a large amount of multi-source monitoring data of multiple farmland areas, these data have large differences in format, and need to map these different modal data to the same graph space, GNN may rely too much on local features when facing such a large amount of multi-source data features, ignoring the correlation between global features, making the model fall into local optimum, especially when constructing a deep graph structure such as a double-layer graph structure, which further leads to the problem of inaccurate feature vector information representation.

[0045] In view of the above problems, the genetic algorithm with adaptive crossover rate and mutation rate is used to optimize the farmland monitoring data feature vectors output by GNN, the corresponding hierarchical coding strategy and fitness function are designed according to the double-layer graph structure, and then the crossover rate and mutation rate of the genetic algorithm are adaptively adjusted to improve the search ability of the genetic algorithm to find the optimal solution, and then improve the feature representation ability of the model to the overall farmland monitoring data.

[0046] The genetic algorithm needs to determine the coding method to encode the data into the initial population, simulate the phenomena of reproduction, gene crossing and gene mutation in natural selection and genetics to find the population with the highest fitness, and then keep it, that is, find the optimal solution.

[0047] (1) Select important feature vectors from the feature vectors in each farmland area through the feature vectors of various feature monitoring data; encode the individuals in the genetic algorithm according to the feature vectors of each farmland area and the selected important feature vectors.

[0048] According to the double-layer graph structure of the farmland monitoring data constructed by the application, the feature vectors are also divided into the feature vectors of the upper-layer graph structure and the lower-layer graph structure. The two-layer feature vectors use the same encoding mode which is not suitable for the distribution of the data features. In view of this problem, the feature vectors of the data are encoded by the hierarchical encoding mode, so that the encoding mode of the feature vectors of the double-layer graph structure is obtained, and the graph structure constraint between the upper layer and the lower layer is maximized.

[0049] Specifically, the hierarchical encoding scheme divides the encoding of each individual C into two parts, which are represented as follows: , wherein, represents the encoding part of the upper-layer graph structure; represents the encoding part of the lower-layer graph structure; Specifically, ; wherein, , , represent the feature vectors of the first, second and Nth farmland areas respectively; φ() represents a dimension reduction function, which is constructed by using a full connection layer, and it is assumed that the dimension of the feature is k, and the purpose of φ() is to compress the k-dimensional feature of to , so as to reduce the search space of the algorithm and improve the search efficiency. The encoding part of the nth farmland area is specifically: , wherein, represents the kth feature vector in the nth farmland area; the 2-norm of each feature vector in the nth farmland area is calculated, and all the feature vectors in the nth farmland area are arranged in descending order according to the 2-norm, and the feature vectors arranged in the first preset number of positions are taken as important feature vectors, represents a set composed of the important feature vectors in the nth farmland area, and the K most important feature vectors are selected from all the feature vectors in the nth farmland area, so that the redundant encoding of the feature vectors in the same farmland area can be avoided. In the embodiment, the value of K is 4, and the value of K is preset by a person, and the implementer can set it according to the actual scene, and the application does not make special limitation.

[0050] By the hierarchical encoding mode, the upper-layer and lower-layer feature vectors in the double-layer graph structure of the farmland monitoring data are hierarchically encoded, the graph structure constraint between the upper layer and the lower layer can be preserved, and the difficulty that the algorithm is difficult to optimize due to the too long encoding caused by too many feature vectors can be avoided.

[0051] (2) The similarity between the feature vectors of different farmland areas of a single individual at each iteration, and the edge weight between different farmland areas in the upper-layer graph structure, and the important feature vectors selected in each farmland area are combined to obtain the fitness value of the single individual at each iteration.

[0052] The fitness function can evaluate each individual in the population, and the fitness value of each individual can be calculated by the fitness function to measure the quality of the individual. According to the form of the feature vector of the double-layer graph structure of the farmland monitoring data, the design of the fitness function needs to include: the consistency between the farmland areas at the upper layer of the graph structure, and the stability of the monitoring data within the farmland area at the lower layer of the graph structure. Based on the above analysis, the fitness value of a single individual at each iteration is obtained by the similarity between the feature vectors of different farmland areas of the single individual at each iteration, and the edge weight value between different farmland areas in the upper layer of the graph structure, combined with the selected important feature vectors within each farmland area, and the expression is: ; In the formula, the fitness value of a single individual; α and β respectively represent the preset weight greater than 0, and the sum of α and β is 1; , respectively represent the feature vectors of the nth and mth farmland areas of a single individual; N represents the number of farmland areas divided by a single individual; cosine( ) represents the cosine similarity function; represents the edge weight value between the nth and mth farmland areas of a single individual; represents the absolute value operation; K represents the number of important feature vectors within a single farmland area of a single individual; represents the kth important feature vector within the Lth farmland area of a single individual.

[0053] In this embodiment, the values of α and β are 0.4 and 0.6 respectively, and the values of α and β are preset by humans, and the implementer can set them according to the actual situation, and the present application does not make special limitations.

[0054] It should be noted that: the fitness value designed above can better optimize the feature vectors output by the GNN, and the feature representation of the farmland monitoring data is optimized by combining the feature vectors of the upper layer graph structure and the lower layer graph structure, and the optimal solution of the feature representation of the farmland monitoring data is obtained. The continuity of adjacent farmland areas in meteorological data can be protected by the deviation of the farmland area feature similarity and the edge weight; The most important monitoring data in the farmland area can be calculated to represent the physical law reflected by the environmental factors in the farmland area.

[0055] The GNN feature vector before optimization is encoded to obtain an individual, and the fitness value of the individual is calculated after decoding the individual into a usable feature vector.

[0056] (3) updating the crossover rate and the mutation rate of each individual in the next iteration by the fitness value of each individual in each iteration compared with the average level of the fitness values of all individuals, respectively combining the crossover rate and the mutation rate of each individual in each iteration.

[0057] The crossover operation in the genetic algorithm is to cross the selected individuals to generate new individuals, simulate gene recombination, and generate better solutions. The mutation operation is to randomly mutate the new individuals to introduce diversity and prevent early convergence. The mutation operation changes part of the genes of the individual to increase the ability to explore the solution space. In the traditional genetic algorithm, the crossover rate and the mutation rate of each individual are fixed values, which increases the risk of local optimal solution of the genetic algorithm in the iteration process, and in high-dimensional space, frequent crossover and mutation may cause the semantic information of the feature vector to be destroyed, affecting the analysis of the subsequent model.

[0058] In the initial search, a higher crossover rate and mutation rate are needed to increase the exploration range, and in the later stage close to the optimal solution, a lower mutation rate is needed for local optimization. Therefore, the core idea of adaptive crossover rate and mutation rate is to dynamically adjust the crossover rate and mutation rate according to the fitness value of the population. With the number of iterations, the update formula of the crossover rate and the mutation rate can be expressed as: ; wherein, represents the crossover rate of the i-th individual in the t+1-th iteration; represents the crossover rate of the i-th individual in the t-th iteration; represents the smoothing factor of the adaptive crossover rate; represents the fitness value of the i-th individual in the t-th iteration; represents the average of the fitness values of all individuals in the t-th iteration; represents the mutation rate of the i-th individual in the t+1-th iteration; represents the mutation rate of the i-th individual in the t-th iteration; represents the smoothing factor of the adaptive mutation rate; represents the exponential function with the natural constant as the base number.

[0059] In this embodiment, the values of the smoothing factor of the adaptive crossover rate and the smoothing factor of the adaptive mutation rate are both 0.5, and the values of the smoothing factor of the adaptive crossover rate and the smoothing factor of the adaptive mutation rate are preset by the person, and the implementer can set them according to the actual situation. This application does not make special restrictions.

[0060] In this embodiment, the initial values of the crossover rate and the mutation rate are set to 0.5 and 0.1, respectively, and the crossover rate and mutation rate The initial values of the crossover rate and the mutation rate can be set by the implementer according to the implementation scenario, and the present application does not make special limitations.

[0061] In order to better optimize the feature vector, the maximum value and the minimum value of the crossover rate are set to 0.9 and 0.1 respectively. In order to better optimize the feature vector, the maximum value and the minimum value of the mutation rate are set to 0.2 and 0.001 respectively. The size of the feature vector can be set by the implementer according to the implementation scenario, and the present application does not make special limitations. It should be noted that the adaptive crossover rate and the mutation rate can be adjusted according to the fitness value of the individual. If the fitness of the individual is lower, a higher crossover rate and mutation rate are used to increase the exploration range. If the fitness of the individual is higher, the crossover rate and mutation rate are reduced to preserve good individuals, thereby avoiding premature convergence and improving the local search ability of the algorithm.

[0062] It should be noted that the adaptive crossover rate and the mutation rate can be adjusted according to the fitness value of the individual. If the fitness of the individual is lower, a higher crossover rate and mutation rate are used to increase the exploration range. If the fitness of the individual is higher, the crossover rate and mutation rate are reduced to preserve good individuals, thereby avoiding premature convergence and improving the local search ability of the algorithm.

[0063] The selection strategy of the genetic algorithm uses the roulette selection strategy to select appropriate individuals for reproduction according to the fitness value. The crossover method of the genetic algorithm uses the multi-point crossover method. The mutation operation of the genetic algorithm selects the Gaussian mutation. The newly generated individuals replace part of the individuals in the population to form a new population, which enters the next generation. The replacement operator uses the fitness replacement strategy, which is positively correlated with the fitness of the individual, and improves the probability of selecting individuals with high fitness. The number of iterations of the genetic algorithm is 100, which can be limited by the implementer according to the actual situation. The optimization process of the genetic algorithm is a prior art, and the specific process will not be described here.

[0064] By optimizing the feature vector of the farmland monitoring data using the optimized genetic algorithm, the representation of the feature vector by the GNN can be avoided to fall into a local optimal feature representation, the feature representation ability of the feature vector graph for the complex double-layer graph structure of the farmland can be improved, and the feature vector graph can be used to realize more efficient data management using the optimized feature vector. The optimization process of the feature vector is shown in Figure 2 .

[0065] Step 3, managing the farmland by using the optimized feature vector.

[0066] ​​​​​The feature vector of the farmland monitoring data optimized by the genetic algorithm with improved adaptive crossover rate and mutation rate is input into the pre-trained SVM classifier, the feature vector is classified and managed, and the state management label of the farmland, such as normal state, water shortage state, and nutrient deficiency state, is output. Management instructions such as irrigation adjustment and fertilization suggestions are generated according to the farmland state management label, and the optimized feature vector of the farmland monitoring data is stored in the database for traceability, thereby improving the accuracy and efficiency of the intelligent farmland monitoring data management.

[0067] In this embodiment, the feature vectors of 10,000 farmland regions and the feature vectors of various types of feature monitoring data in the 10,000 farmland regions are obtained from historical data, and are divided into a training set and a test set in a ratio of 7:3 for training and testing the SVM classifier.

[0068] In summary, the upper-layer graph structure takes the farmland region as a node, considers the geographical spatial relationship between the farmland regions and the difference in meteorological monitoring data, and can reflect the overall layout of the farmland and the mutual influence between the regions from a macro perspective. The lower-layer graph structure takes the feature monitoring data in each farmland region as a node, focuses on the relationship between different feature data in the region, and realizes the organic combination of global and local information, which helps to more accurately understand and analyze the farmland monitoring data. Further, the graph neural network can fully utilize the topological information and node feature information in the double-layer graph structure to extract and fuse the features of different levels of graph structure, and generate feature vectors with stronger expression ability. Further, selecting important feature vectors in the iteration process can remove redundant and unimportant feature information, reduce the dimension of the feature vector, reduce the subsequent calculation complexity, retain key features, improve the efficiency and performance of the model, and avoid the problems of long encoding and optimization difficulty caused by too many feature vectors. Through the improvement of the adaptive crossover rate and mutation rate, the genetic algorithm can dynamically adjust the crossover rate and mutation rate according to the fitness value of the individual, so that the algorithm can maintain strong exploration ability in the search process and effectively avoid premature convergence, improve the global search ability and optimization accuracy of the algorithm, and further improve the optimization effect of the feature vector. The optimized feature vector can more efficiently reflect the actual situation of the farmland, and improve the accuracy and efficiency of the farmland monitoring data management.

[0069] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flow diagrams and / or block diagrams.

[0070] It is apparent that a person skilled in the art can make a variety of modifications to the application described above without departing from the spirit and scope of the application. Therefore, the above-described embodiments of the application are intended to be illustrative only and not restrictive.

Claims

1. A smart farmland monitoring data management method based on artificial intelligence, characterized in that: The method comprises the following steps: Obtain various environmental monitoring data for each preset farmland area in the smart farmland, as well as various feature monitoring data for each location within each farmland area; The upper layer of the double-layer graph structure of farmland monitoring data is obtained by the positional relationship between different farmland areas and the difference in meteorological monitoring data; the lower layer of the double-layer graph structure of farmland monitoring data is obtained by the similarity between different types of feature monitoring data at each location in each farmland area; The graph neural grid is used to extract the feature vectors of the upper and lower graph structures respectively, and the feature vectors of each farmland area and the feature vectors of various feature monitoring data in each farmland area are obtained; The acquired feature vectors are iteratively optimized through a genetic algorithm. During the iteration process, important feature vectors are selected from the feature vectors in each farmland area through the feature vectors of various feature monitoring data. Individuals in the genetic algorithm are encoded based on the feature vectors of each farmland area and the selected important feature vectors. The fitness value of a single individual in each iteration is obtained by combining the similarity between the feature vectors of different farmland areas under a single individual at each iteration and the edge weights between different farmland areas in the upper graph structure with the important feature vectors selected within each farmland area. The fitness value of each individual in each iteration is compared with the average fitness value of all individuals at each iteration, and the crossover rate and mutation rate of each individual at each iteration are combined to update the crossover rate and mutation rate of each individual at the next iteration. All the obtained eigenvectors are optimized by the optimized genetic algorithm, and the farmland is managed by the optimized eigenvectors.

2. The method for managing smart farmland monitoring data based on artificial intelligence according to claim 1, wherein: The upper layer graph structure of the double-layer graph structure for obtaining farmland monitoring data includes: The nodes in the upper-level graph structure are single farmland areas. The method for calculating the edge weight between any two farmland areas is as follows: all meteorological monitoring data of each farmland area are combined into a meteorological monitoring data vector, the Euclidean distance between the any two farmland areas is calculated, the 2-norm of the difference between the meteorological monitoring data vectors of the any two farmland areas is calculated, and the sum of the Euclidean distance and the 2-norm is used as the edge weight between the any two farmland areas.

3. The method for managing farmland monitoring data based on artificial intelligence according to claim 1, wherein: The lower layer graph structure of the double-layer graph structure for obtaining farmland monitoring data includes: The nodes in the lower-level graph structure are single-category feature monitoring data in the farmland area corresponding to each node in the upper-level graph structure. The method for calculating the edge weight between any two categories of feature monitoring data is as follows: all values ​​of each type of feature monitoring data collected in each farmland area are arranged in order of position to form each feature monitoring data vector, and the similarity of the feature monitoring data vectors between any two categories of feature monitoring data is used as the edge weight between the any two categories of feature monitoring data.

4. The method for managing farmland monitoring data based on artificial intelligence according to claim 1, wherein: The process of selecting the important eigenvectors is as follows: Calculate the 2-norm of each eigenvector in each farmland area, sort all eigenvectors in each farmland area in descending order according to the 2-norm, and take the eigenvectors ranked in the front by a preset number of digits as important eigenvectors.

5. The method for managing farmland monitoring data based on artificial intelligence according to claim 1, wherein: The encoding of individuals in the genetic algorithm comprises: The encoding of each individual C is divided into two parts, expressed as: ;in, The encoding part that represents the upper-level graph structure; The encoding part that represents the underlying graph structure; Specifically: ;in, 、 、 Represent the feature vectors of the first, second, and Nth farmland areas respectively; φ( ) represents the dimensionality reduction function, which is constructed using the fully connected layer; the encoding part of the nth farmland area Specifically: ,in, represents the kth eigenvector in the farmland of the nth region; Represents a set of important feature vectors in the nth farmland area.

6. The method for managing farmland monitoring data based on artificial intelligence according to claim 2, wherein: The step of obtaining the fitness value of a single individual at each iteration includes: For a single individual, the similarity of the feature vectors between two randomly selected farmland areas under the single individual is recorded as feature similarity, and the difference between the edge weights between the two randomly selected farmland areas and the feature similarity is calculated; the mean of the difference between all randomly selected farmland areas under the single individual is calculated; Calculate the average value of the modulus of all important eigenvectors in all farmland areas under a single individual; The fitness value of a single individual can be further obtained through the mean and the average value.

7. The method for managing farmland monitoring data based on artificial intelligence according to claim 6, wherein: The fitness value of the single individual is a weighted sum of the mean and the average value, and the sum of the weight of the mean and the weight of the average value is 1.

8. The method for managing smart farmland monitoring data based on artificial intelligence according to claim 1, wherein: The update formula of the crossover rate is: ;in, represents the crossover rate of the i-th individual at the t+1-th iteration; represents the crossover rate of the i-th individual at the t-th iteration; represents the smoothing factor of the adaptive crossover rate; represents the fitness value of the i-th individual at the t-th iteration; Represents the mean fitness value of all individuals at the tth iteration; Represents an exponential function with a natural constant as its base.

9. The method for managing smart farmland monitoring data based on artificial intelligence according to claim 8, wherein: The update formula of the mutation rate is: ; represents the mutation rate of the i-th individual at the t+1-th iteration; represents the mutation rate of the i-th individual at the t-th iteration; Smoothing factor for the adaptive mutation rate.

10. The method for managing smart farmland monitoring data based on artificial intelligence according to claim 1, wherein: The farmland management by using the optimized feature vector includes: The optimized feature vector is input into the pre-trained SVM classifier, the feature vector is classified and managed, and the status management label of the farmland is output.

Images