Red soil slope crop management method

Abstract

The invention relates to the technical field of crop management, in particular to a red soil slope crop management method. Acquiring crop data, surrounding environment data and soil data corresponding to crops planted in the target red soil slope area; identifying the crop data and the soil data, and determining a current drought level corresponding to the crops; identifying the crop data, and determining a current insect disease type and a current insect disease grade corresponding to the crops; determining a drought grade reason and an insect disease grade reason corresponding to the crops based on the surrounding environment data and the soil data; outputting an irrigation strategy based on the current drought level and the drought level reason; based on the current pest disease type, the current pest disease grade and the current pest disease grade reason, outputting pest control measures corresponding to the crops. The method achieves the comprehensive, scientific and accurate management of the crops in the red soil slope, improves the efficiency, and saves the time cost and manpower cost.

Description

Technical Field

[0001] The present invention relates to the technical field of crop management, and particularly to a method for managing crops on red soil slopes. Background Art

[0002] Red soil slopes are widely distributed in the southern regions of China, accounting for about 23% of the total national land area, and are important land resources for southern agricultural production. In terms of crop planting, red soil slopes mainly grow food and cash crops such as rice, corn, and sugarcane. However, due to soil and terrain conditions, the growth and development of crops are significantly affected, and it is difficult to improve the yield and quality. At present, a series of technical measures have been adopted for crop management on red soil slopes. In terms of soil improvement, the soil pH value is adjusted by applying lime, and the soil fertility is enhanced by increasing organic fertilizers and biochar to improve the soil structure; in terms of soil and water conservation, measures such as contour planting, terrace construction, and planting slope protection plants are implemented, which have reduced soil and water loss to a certain extent.

[0003] However, there are still obvious deficiencies in the existing technologies. The traditional method of applying lime is prone to cause excessive fluctuations in the soil pH value, affecting the activity of soil microorganisms and the absorption of other nutrients by crops; the application rates and application times of organic fertilizers and biochar lack precise regulation, making it difficult to fully exert their soil improvement and fertility enhancement effects. In terms of soil and water conservation, the costs of contour planting and terrace construction are relatively high, and it is difficult to promote them on a large scale in some areas due to terrain and financial constraints; the selection and configuration of slope protection plants are not scientific enough, and the protection effect is limited. In addition, most of the existing management technologies focus on solving single problems and lack comprehensive consideration of the growth environment of crops on red soil slopes, and it is impossible to achieve the coordinated development of soil improvement, soil and water conservation, and high-yield and high-quality crops.

[0004] Therefore, how to comprehensively manage the crops on red soil slopes has become an urgent problem to be solved. Summary of the Invention

[0005] In view of this, the present invention provides a method for managing crops on red soil slopes to solve the problem of how to comprehensively manage the crops on red soil slopes.

[0006] In the first aspect, the present invention provides a method for managing crops on red soil slopes, the method comprising: obtaining crop data, surrounding environment data, and soil data corresponding to the crops planted in the target red soil slope area; the crop data includes crop image data, crop spectral data, and crop thermal infrared imaging data; the surrounding environment data includes at least one of light intensity, humidity, and wind speed; the soil data includes at least one of soil physical property data, soil chemical property data, soil biological property data, and water property data; Identify crop data and soil data to determine the current drought level corresponding to the crops. Identify crop data to determine the current pest and disease type and the current pest and disease level corresponding to the crops. Based on the surrounding environment data and soil data, determine the reasons for the drought level and the pest and disease level corresponding to the crops. Based on the current drought level and the reasons for the drought level, output an irrigation strategy; the irrigation strategy includes an intermittent supplementary irrigation mode and a continuous irrigation mode. Based on the current pest and disease type, the current pest and disease level, and the reasons for the current pest and disease level, output the pest and disease control measures corresponding to the crops.

[0007] The red soil sloping land crop management method provided by the embodiments of the present application obtains the crop data, the surrounding environment data, and the soil data corresponding to the crops planted in the target red soil sloping land area; identify the crop data and the soil data to determine the current drought level corresponding to the crops. This ensures the accuracy of the determined current drought level. Compared with single-index evaluation, it can more accurately reflect the actual drought degree faced by the crops, which helps to take corresponding drought resistance measures in a timely manner. Identify the crop data to determine the current pest and disease type and the current pest and disease level corresponding to the crops, which ensures the accuracy of the determined current pest and disease type and the current pest and disease level, thus providing a strong basis for formulating targeted pest and disease control strategies and avoiding inappropriate control measures caused by misdiagnosis or misjudgment. Based on the surrounding environment data and soil data, determine the reasons for the drought level and the pest and disease level corresponding to the crops. It can deeply explore various environmental and soil factors affecting crop growth, clarify the action mechanisms of each factor on the occurrence and development of drought and pests and diseases, thus providing a direction for fundamentally solving problems and helping to take more targeted and effective prevention and control measures. Based on the current drought level and the reasons for the drought level, output an irrigation strategy. This ensures the accuracy of the output irrigation strategy. Based on the current pest and disease type, the current pest and disease level, and the reasons for the current pest and disease level, output the pest and disease control measures corresponding to the crops, which ensures the accuracy of the output pest and disease control measures. These management measures can accurately target the specific problems faced by the crops and their roots, effectively improve the efficiency and quality of agricultural production, reduce the impact of drought and pests and diseases on crop yield and quality, and achieve the sustainable development of agriculture. It realizes the comprehensive, scientific and accurate management of red soil sloping land crops, improves efficiency, and saves time and labor costs.

[0008] In an alternative embodiment, the soil data includes soil physical property data, soil chemical property data, and moisture property data. Identifying the crop data and the soil data to determine the current drought level corresponding to the crops includes: Input crop image data, crop spectral data, crop thermal infrared imaging data, soil physical property data, soil chemical property data, and moisture property data into a preset drought level determination model; The preset drought level determination model extracts features from the input data and performs feature fusion to obtain a fused feature vector; Extract features from the fused feature vector to generate a target feature vector; Based on the target feature vector, output the current drought level corresponding to the crop.

[0009] In an optional implementation, extracting features from the fused feature vector to generate a target feature vector includes: Map the fused feature vector to a mean vector and a variance vector in the latent space based on a variational autoencoder; Perform sampling processing based on the mean vector and the variance vector to obtain at least one sampling point; the sampling points contain the mixed coding information of multimodal data in the latent space; the mixed coding information includes multiple sub-features; Based on a preset decoder, transform and combine each sampling point to generate a target feature vector with the same dimension as the input data.

[0010] In an optional implementation, based on a preset decoder, transforming and combining each sampling point to generate a target feature vector with the same dimension as the input data includes: For each layer in the preset decoder, calculate the similarity between each sub-feature in the sampling point and a preset drought key feature template; Determine the weight information of each sub-feature according to the calculated similarity; Based on the weight information corresponding to each sub-feature, perform weighted processing on each sub-feature to obtain a weighted vector; The weighted vector undergoes a convolution operation through a convolutional layer in the preset decoder to generate a target feature vector.

[0011] In an optional implementation, identifying crop data to determine the current pest and disease type and the current pest and disease level corresponding to the crop includes: Extract features from the crop image data, crop spectral data, and crop thermal infrared imaging data, and fuse the extracted features to generate an initial fused feature; Extract features from the initial fused feature based on a spatio-temporal attention mechanism to generate a target fused feature; Input the target fused feature into a preset pest and disease level determination model, and output the current pest and disease type and the current pest and disease level corresponding to the crop.

[0012] In an alternative embodiment, feature extraction is performed on crop image data, crop spectral data, and crop thermal infrared imaging data, and the extracted features are fused to generate initial fused features, including: Feature extraction is performed on the crop image data to obtain target image feature values; The crop spectral data is encoded using a preset encoder to obtain a low-dimensional sparse representation; the low-dimensional sparse representation highlights specific spectral band features related to pests and diseases; The crop thermal infrared imaging data is arranged in chronological order, and each pixel point in each thermal infrared image in the crop thermal infrared imaging data is used as a node; Each node corresponds to a specific spatio-temporal position, and the attribute of the node is the temperature value at the corresponding moment of the spatio-temporal position; Edges between two nodes are determined according to the temperature change relationship between two nodes at adjacent time points and adjacent spatial positions; The weight of the edge is determined according to the absolute value of the temperature difference between the two nodes, and a spatio-temporal graph is generated; The target image feature values, the low-dimensional sparse representation, and the spatio-temporal graph are fused to generate initial fused features.

[0013] In an alternative embodiment, feature extraction is performed on the crop image data to obtain target image feature values, including: Feature extraction is performed on the crop image data to generate initial color features, initial texture features, and initial shape features; The initial color features, initial texture features, and initial shape features are mapped to a discrete value range to create an image feature search space; Coding rules are formulated according to the discrete value range of the image feature search space; An initial state is randomly assigned to each initial qubit; An initial quantum population is initialized; the initial quantum population includes multiple initial quantum states, and each initial quantum state is represented by a group of initial qubits, representing a possible combination of image features; According to the fitness function, the fitness evaluation results of the image feature combinations corresponding to the initial quantum states in the initial quantum population are calculated; The rotation angles of each initial qubit are calculated according to the fitness evaluation results; the magnitude and direction of the rotation angle depend on the fitness evaluation results of the current quantum state and the target state; Quantum rotation gate operations are performed on each initial qubit according to the calculated rotation angles to obtain each updated qubit and the updated quantum states composed of the updated qubits; Each updated quantum state is measured, and the updated quantum state is converted into classical information to obtain a classical state; Decode the classical state into the initial image eigenvalue according to the coding rule; Filter the initial image eigenvalue to obtain the target image eigenvalue.

[0014] In an alternative embodiment, input the target fusion feature into a preset pest and disease level determination model, and output the current pest and disease type and the current pest and disease level corresponding to the crop, including: Obtain the target knowledge graph; the target knowledge graph is composed of crop variety characteristics, pest and disease transmission laws, and red soil slope environment data; Input the target fusion feature into a preset pest and disease determination model, and the preset pest and disease determination model outputs an initial determination result; the initial determination result includes the initial pest and disease type and the initial pest and disease level corresponding to the crop; Construct a query statement with the initial pest and disease type as the core, combining the variety, growth stage, surrounding environment data, and soil data of the crop; Query in the target knowledge graph based on the query statement to obtain a query result; the query result includes the occurrence probability and severity level of the initial pest and disease type; Fuse the initial determination result and the query result to obtain the current pest and disease and the current pest and disease level.

[0015] In an alternative embodiment, based on the surrounding environment data and soil data, determine the reasons for the drought level and the pest and disease level corresponding to the crop, including: Use the current drought level, the current pest and disease level, the surrounding environment data, and the soil data as nodes, and connect the corresponding nodes with directed edges according to the relationships between the current drought level, the current pest and disease level, the surrounding environment data, and the soil data to generate a directed acyclic graph; Input the directed acyclic graph into the feature extraction network in the preset causality model; The feature extraction network extracts features from the directed acyclic graph to obtain the node features, edge features, and graph structure features corresponding to the directed acyclic graph, and inputs the node features, edge features, and graph structure features into the decision network in the preset causality model; The decision network determines the reasons for the drought level and the current pest and disease level based on the node features, edge features, and graph structure features.

[0016] In an alternative embodiment, the decision network determines the reasons for the drought level and the current pest and disease level based on the node features, edge features, and graph structure features, including: Construct an initial state corresponding to the state space in the decision network according to the node features, edge features, and graph structure features; Based on the initial state, an initial action is output from the action space corresponding to the decision-making network; the initial action represents an initial estimated value of the influence amount of each factor on the current drought level and the current pest and disease level. Starting from the nodes that have an impact on the current drought level and the current pest and disease level, the decision-making network derives the causal transmission path along the directed edges in the graph. According to the intensity of the edge features and the position information of each node in the causal chain, key nodes are determined from each node. Based on each key node, the initial estimated value of the influence amount of each factor on the current drought level and the pest and disease level in the initial action is adjusted to obtain the target estimated value of the influence amount of each factor on the current drought level and the pest and disease level. According to each target estimated value of the influence amount, target factors with a target estimated value of the influence amount greater than a preset threshold are determined from each factor. Based on the target factors, the cause of the drought level and the cause of the current pest and disease level are determined. Description of the Drawings

[0017] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for use in the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0018] Figure 1 It is a schematic flowchart of a method for managing crops on red soil slopes according to an embodiment of the present invention. Figure 2 It is a schematic flowchart of another method for managing crops on red soil slopes according to an embodiment of the present invention. Detailed Embodiments

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0020] According to an embodiment of the present invention, an embodiment of a method for managing crops on red soil slopes is provided. It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer-executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0021] In this embodiment, a method for managing crops on red soil slopes is provided, which can be used in the above-mentioned electronic device. Figure 1 It is a flowchart of the method for managing crops on red soil slopes according to an embodiment of the present invention, as Figure 1 shown. The process includes the following steps: Step S101, obtain crop data, surrounding environment data, and soil data corresponding to the crops planted in the target red soil slope area.

[0022] Among them, the crop data includes crop image data, crop spectral data, and crop thermal infrared imaging data. The surrounding environment data includes at least one of light intensity, humidity, and wind speed; the soil data includes at least one of soil physical property data, soil chemical property data, soil biological property data, and water property data.

[0023] Specifically, the electronic device can collect crop data and surrounding environment data based on the intelligent wearable collection device equipped for farmers. Among them, the intelligent wearable collection device can integrate a micro image collector, a spectral sensor, a thermal infrared detector, and an environment monitoring module. The integrated micro image collector can collect crop image data (such as leaf morphology and color change images at different growth stages); the spectral sensor can collect crop spectral data (spectral characteristics reflecting the physiological state of the crop); the infrared detector can collect crop thermal infrared imaging data (monitor the crop temperature and judge the water stress situation). At the same time, the environment monitoring module can collect surrounding environment data, including light intensity (light amount at different time periods and different slopes), humidity (dynamic changes in air humidity and soil humidity), wind speed (affecting crop transpiration and pest and disease transmission), and soil data (soil texture, nutrient content, pH, etc.).

[0024] Step S102, identify the crop data and the soil data to determine the current drought level corresponding to the crops.

[0025] Specifically, the electronic device can input the crop data and the soil data into a preset drought level determination model to output the current drought level corresponding to the crops.

[0026] This step will be introduced in detail below.

[0027] Step S103, identify the crop data to determine the current pest and disease type and the current pest and disease level corresponding to the crops.

[0028] Specifically, the electronic device can identify crop image data, crop spectral data, and crop thermal infrared imaging data respectively. Then, based on the identification results, determine the current pest and disease type and the current pest and disease level corresponding to the crop. For example, if there are yellow spots and curling on the crop leaves, it may be infected with a virus disease; if there are holes and notches on the leaves, it may be caused by chewing mouthpart pests such as the diamondback moth.

[0029] Evaluate the pest and disease level based on indicators such as the damage degree of pests and diseases to crops, the incidence range, and the number of pests. Generally, it can be divided into three levels: mild, moderate, and severe. For example, mild diseases may only show symptoms on a few plants and have little impact on the overall growth of crops; while severe diseases may cause large areas of plants to be damaged, seriously affecting the yield and quality of crops.

[0030] This step will be introduced in detail below.

[0031] Step S104, based on the surrounding environment data and soil data, determine the reasons for the drought level and the reasons for the pest and disease level corresponding to the crop.

[0032] Specifically, the electronic device can determine the reasons for the drought level and the reasons for the pest and disease level corresponding to the crop according to the relationship between the surrounding environment data and soil data and the current drought level, the current pest and disease type, and the current pest and disease level.

[0033] This step will be introduced in detail below.

[0034] Step S105, based on the current drought level and the reasons for the drought level, output an irrigation strategy.

[0035] Among them, the irrigation strategy includes an intermittent supplementary irrigation mode and a continuous irrigation mode.

[0036] Specifically, the electronic device can clarify the current drought level where the crop is located. The higher the drought level, the more serious the water stress faced by the crop and the more urgent the need for irrigation. For example, a high drought level may mean that the soil moisture is severely insufficient and the growth of the crop is significantly inhibited. In addition, the electronic device determines the reasons for the drought level, such as too fast soil water infiltration, insufficient irrigation, dry climate, etc. This helps to formulate an irrigation strategy targeted. For example, if the drought is caused by too fast soil water infiltration rate, it is necessary to consider how to slow down the water infiltration or increase the irrigation frequency when formulating the irrigation strategy.

[0037] Then, the electronic device obtains the water requirement characteristics of the crop variety, the current growth stage, and the real-time soil water content data. Among them, different crop varieties have different water requirement characteristics. For example, rice is a water-loving crop and requires more water during the growth process, while corn is relatively drought-tolerant and has relatively less water requirement. In addition, the water requirements of crops are also different at different growth stages. For example, at the seedling stage, the roots of crops are not yet well-developed and the water requirement is relatively less, but at the flowering and fruiting stage, the water requirement is usually large. The real-time soil water content data is obtained through devices such as soil moisture sensors, and based on this, the amount of water that still needs to be supplemented in the soil to meet the growth requirements of the crop at the current growth stage is calculated, that is, the theoretical water requirement.

[0038] The electronic device calculates the initial irrigation supplement amount according to the water requirement characteristics of the crop variety, the current growth stage, and the real-time soil water content data. The specific formula is as follows: Q = S × H × (Wtarget - Wcurrent). Where Q represents the initial irrigation supplement amount (unit: cubic meters or liters); S is the irrigation area (unit: square meters); H is the depth of the main soil layer where the crop roots are distributed (unit: meters); Wtarget is the suitable soil water content of the crop at the current growth stage (expressed as a volume percentage); Wcurrent is the real-time monitored soil water content (expressed as a volume percentage).

[0039] Then, the electronic device corrects the theoretical water requirement in combination with the water infiltration rate of the red soil slope to obtain the target irrigation supplement amount. The water infiltration rate of the red soil slope is relatively fast, which will cause more water loss. Therefore, it is necessary to appropriately increase the irrigation amount to make up for this part of the loss. According to the specific value of the water infiltration rate and the theoretical water requirement, the actual amount of water that needs to be supplemented considering the soil infiltration factor is calculated, that is, the target irrigation supplement amount.

[0040] When the current drought level is less than the preset drought level, or the reason for the drought level is mainly due to insufficient water in the short term and the soil water retention capacity is relatively good, the intermittent irrigation supplement mode can be considered. For example, the crop is in a mild drought level and the soil texture is relatively heavy, and the water infiltration is relatively slow.

[0041] Specific implementation: Divide the target irrigation supplement amount into several smaller shares and conduct multiple irrigations at certain time intervals. This can avoid excessive water loss caused by one-time irrigation and at the same time allow the crop to have enough time to absorb and utilize the water from each irrigation. For example, divide the target irrigation supplement amount into 3 - 5 times and conduct irrigation once every 1 - 2 days.

[0042] When the current drought level is greater than or equal to the preset drought level, the crops are severely water stressed, or the reason for the drought level is long-term water shortage and extremely fast soil water infiltration rate, etc., a continuous irrigation mode may be more appropriate. For example, the crops are in a severe drought level, and the water infiltration rate of the red soil slope is very fast, and the soil water retention capacity is poor.

[0043] Specific implementation: Continuously irrigate for a period of time to keep the soil at a relatively high water content to relieve the drought situation of the crops as soon as possible. However, attention should be paid to controlling the irrigation speed to avoid problems such as waterlogging and soil erosion. Drip irrigation, sprinkler irrigation and other methods can be used to irrigate with a relatively uniform and continuous water flow until the target supplementary irrigation amount is reached or the soil water content reaches the appropriate range.

[0044] Step S106, based on the current pest and disease type, the current pest and disease level, and the reason for the current pest and disease level, output the corresponding pest and disease control measures for the crops.

[0045] Specifically, the electronic device can output one of the corresponding agricultural control measures, physical control measures, biological control measures, and chemical control measures for the crops based on the current pest and disease type, the current pest and disease level, and the reason for the current pest and disease level. Among them, the agricultural control measures include: reasonable crop rotation, cleaning the fields, reasonable close planting, scientific fertilization and other control measures; the physical control measures include: color board trapping, light trapping, high-temperature greenhouse steaming and other control measures; the biological control measures include: using natural enemies, using biological agents and other control measures; chemical control measures: when the effects of other control measures are not good or the pest and disease occur severely, chemical pesticides can be selected for control. However, attention should be paid to selecting pesticides with high efficiency, low toxicity and low residue, and strictly operating in accordance with the pesticide use instructions to ensure the quality safety of agricultural products and environmental safety.

[0046] The crop management method for red soil slopes provided by the embodiments of the present application obtains the crop data, surrounding environment data, and soil data corresponding to the crops planted in the target red soil slope area; identifies the crop data and soil data to determine the current drought level corresponding to the crops. This ensures the accuracy of the determined current drought level. Compared with single-index evaluation, it can more accurately reflect the actual drought degree faced by the crops, which helps to take corresponding drought resistance measures in a timely manner. Identifies the crop data to determine the current pest and disease type and the current pest and disease level corresponding to the crops, ensuring the accuracy of the determined current pest and disease type and the current pest and disease level, thereby providing a strong basis for formulating targeted pest control strategies and avoiding inappropriate control measures caused by misdiagnosis or misjudgment. Based on the surrounding environment data and soil data, determines the reasons for the drought level and the reasons for the pest and disease level corresponding to the crops. It can deeply explore various environmental and soil factors affecting crop growth, clarify the mechanism of action of each factor on the occurrence and development of drought and pests and diseases, thereby providing a direction for fundamentally solving the problem and helping to take more targeted and effective prevention and control measures. Based on the current drought level and the reasons for the drought level, outputs an irrigation strategy. This ensures the accuracy of the output irrigation strategy. Based on the current pest and disease type, the current pest and disease level, and the reasons for the current pest and disease level, outputs the pest control measures corresponding to the crops, ensuring the accuracy of the output pest control measures. These management measures can accurately target the specific problems faced by the crops and their roots, effectively improve the efficiency and quality of agricultural production, reduce the impact of drought and pests and diseases on crop yield and quality, and achieve the sustainable development of agriculture. It realizes the comprehensive, scientific, and accurate management of red soil slope crops, improves efficiency, and saves time and labor costs.

[0047] In this embodiment, a crop management method for red soil slopes is provided, which can be used in the above-mentioned electronic device. Figure 2 It is a flowchart of the crop management method for red soil slopes according to the embodiments of the present invention, as Figure 2 shown. The process includes the following steps: Step S201: Obtain the crop data, surrounding environment data, and soil data corresponding to the crops planted in the target red soil slope area.

[0048] For this step, please refer to the introduction of step S101 above and will not be elaborated here.

[0049] Step S202: Identify the crop data and soil data to determine the current drought level corresponding to the crops.

[0050] Specifically, the soil data includes soil physical property data, soil chemical property data, and moisture property data. The above step S202 may include the following steps: Step S2021: Input the crop image data, crop spectral data, crop thermal infrared imaging data, soil physical property data, soil chemical property data, and moisture property data into a preset drought level determination model.

[0051] Specifically, the electronic device can input the crop image data, crop spectral data, crop thermal infrared imaging data, soil physical property data, soil chemical property data, and moisture property data into a preset drought level determination model.

[0052] Step S2022: The preset drought level determination model extracts features from the input data and performs feature fusion to obtain a fused feature vector.

[0053] Specifically, the preset drought level determination model can use a pre-trained ResNet model to extract the texture features, shape features, and color features of the crop leaves in the crop image data, which can reflect the health status and growth state of the crops. For the crop spectral data, the preset drought level determination model uses the principal component analysis (PCA) method to extract the main features of the spectral curve, reduce the data dimension, and retain important spectral information at the same time. For the soil data, features such as soil porosity, pH value, and organic matter content are extracted according to its physical, chemical, and moisture properties.

[0054] Then, based on the multi-modal fusion network in the preset drought level determination model, different types of features are fused to obtain an initial feature vector. The multi-modal fusion network can adopt a deep neural network structure, such as a multi-layer perceptron (MLP) or a long short-term memory network (LSTM). During the fusion process, different weights are assigned to different types of features and are adaptively adjusted according to their importance for drought level judgment.

[0055] Then, feature compression and dimensionality reduction techniques, such as an autoencoder (Autoencoder) or linear discriminant analysis (LDA), are used to compress the high-dimensional initial feature vector into a low-dimensional fused feature vector.

[0056] Step S2023: Extract features from the fused feature vector to generate a target feature vector.

[0057] Specifically, the above Step S2023 may include the following steps: Step a1: Map the fused feature vector to the mean vector and variance vector in the latent space based on the variational autoencoder.

[0058] Specifically, the fused feature vector is input into a variational encoder network. The variational encoder network processes the input fused feature vector and finally outputs two vectors, a mean vector and a log variance vector, through a series of linear transformations and non-linear activation functions.

[0059] Step a2, perform sampling processing based on the mean vector and the variance vector to obtain at least one sampling point.

[0060] Among them, the sampling points contain the mixed coding information of the multi-modal data in the latent space; the mixed coding information includes multiple sub-features.

[0061] Specifically, the mean vector and the variance vector obtained by the variational autoencoder define a Gaussian distribution in the latent space, that is, z~N(μ,σ 2 I), where z represents the sample in the latent space and I is the identity matrix. This distribution characteristic is the basis for subsequent sampling, indicating that the sampling points will be generated from the Gaussian distribution that conforms to this mean and variance.

[0062] Since the variance vector σ 2 represents the variance in each dimension, in order to perform sampling, the standard deviation vector σ needs to be calculated. By taking the square root of the variance in each dimension, that is , the standard deviation vector is obtained. Each element of this vector corresponds to the standard deviation in one dimension of the latent space, which determines the fluctuation range of sampling in that dimension.

[0063] Then, use a random number generator to generate random numbers of a standard Gaussian distribution with the same dimension as the latent space , that is ~N(0,I). The standard Gaussian distribution is a normal distribution with a mean of 0 and a variance of 1, and these random numbers will be used to introduce randomness in the sampling process.

[0064] According to the reparameterization trick, the sampling point z is sampled from the defined Gaussian distribution, and the calculation formula is z = μ + σ⊙ , where ⊙ represents element-wise multiplication. Through this formula, the mean vector μ, the standard deviation vector σ, and the random numbers of the standard Gaussian distribution are combined to obtain the sampling point z. Such a calculation method not only uses the mean and variance to determine the central position and fluctuation range of the sampling point, but also introduces randomness through random numbers, making the results obtained by each sampling different.

[0065] The obtained sampling point z is located in the latent space, which contains the mixed coding information of multimodal data in the latent space. This mixed coding information is an abstract representation of the original multimodal data (such as crop image data, spectral data, thermal infrared imaging data, soil data, etc.) after being encoded by a variational autoencoder. Each dimension in the latent space corresponds to one or more factors affecting the crop state. Therefore, the different element values of the sampling point z constitute multiple sub-features. These sub-features reflect information such as the growth status of crops, environmental factors, and pest and disease risks from different perspectives, and are a high-level generalization and refinement of the original multimodal data.

[0066] Finally, according to actual needs, the above sampling steps can be repeated multiple times to generate multiple sampling points. Different sampling points represent different states in the latent space. Through these sampling points, the latent space can be explored more comprehensively, and the information contained in the multimodal data can be mined, providing rich data support for subsequent data analysis and decision-making.

[0067] Step a3: Based on a preset decoder, transform and combine each sampling point to generate a target feature vector with the same dimension as the input data.

[0068] Specifically, the above step a3 may include the following steps: Step a31: In each layer of the preset decoder, calculate the similarity between each sub-feature in the sampling point and a preset drought key feature template.

[0069] Specifically, starting from the input layer of the preset decoder, using the sampling point as the input, a set of sub-features is obtained after being processed by the first layer of the network, and then the similarity between these sub-features and the drought key feature template is calculated based on a preset similarity algorithm. Among them, the preset similarity algorithm can be any one of similarity algorithms such as the Euclidean distance algorithm, Manhattan distance algorithm, cosine similarity algorithm, etc.

[0070] Step a32: Determine the weight information of each sub-feature according to the calculated similarity.

[0071] Specifically, the electronic device assigns weight information to each sub-feature according to the magnitude of the calculated similarity.

[0072] Step a33: Based on the weight information corresponding to each sub-feature, perform a weighting process on each sub-feature to obtain a weighted vector.

[0073] Specifically, the electronic device performs a weighting process on each sub-feature based on the weight information corresponding to each sub-feature to obtain a weighted vector.

[0074] Step a34: The weighted vector undergoes a convolution operation through the convolution layer in the preset decoder to generate a target feature vector.

[0075] Specifically, the convolutional layer performs a convolution operation on the weighted vector (which appears as a feature map in data such as images). The convolution kernel slides over the weighted vector, extracts local features through the convolution operation, and combines these local features to generate a target feature vector. The convolutional layer can effectively capture the local structure and spatial relationships of features, and is very effective in processing data with spatial structures (such as features related to crop images). At the same time, the convolutional layer can also further reduce the dimension of features through operations such as pooling, reducing the computational amount.

[0076] Step S2024, based on the target feature vector, output the current drought level corresponding to the crop.

[0077] Specifically, map the target feature vector to a probability distribution space, and output the probability distribution of the crop being in different drought levels (such as mild drought, moderate drought, severe drought, etc.). Then, determine the drought level with the maximum probability value as the current drought level according to the probability distribution. This probability output method can not only give the judgment of the drought level, but also reflect the uncertainty of the judgment, providing more abundant information for agricultural decision-making.

[0078] Step S203, identify the crop data to determine the current pest and disease type and the current pest and disease level corresponding to the crop.

[0079] Specifically, the above step S203 may include the following steps: Step S2031, extract features from the crop image data, crop spectral data, and crop thermal infrared imaging data, and fuse the extracted features to generate an initial fused feature.

[0080] Specifically, the above step S2031 may include the following steps: Step b1, extract features from the crop image data to obtain target image feature values.

[0081] Specifically, the above step b1 may include the following steps: Step b11, extract features from the crop image data to generate initial color features, initial texture features, and initial shape features.

[0082] Specifically, the electronic device can convert the crop image data from the common RGB color space to a preset color space to obtain the initial color features. The preset color space can be HSV (hue, saturation, value), Lab (luminance, green-red axis, blue-yellow axis), etc. By calculating some feature values (such as contrast, correlation, energy, entropy, etc.) of the GLCM for the crop image data, the electronic device can obtain the initial texture features of the image. Among them, the contrast reflects the severity of the gray-scale changes in the image, the correlation represents the directionality of the texture, the energy represents the uniformity of the texture, and the entropy represents the complexity of the texture. The electronic device uses an edge detection algorithm to detect the edges of the crop image, and then obtains the contour curve of the crop through a contour tracking algorithm to get the initial shape features. The contour curve can intuitively represent the shape boundary of the crop and provide a basis for subsequent shape analysis.

[0083] Step b12: Map the initial color features, initial texture features, and initial shape features to a discrete value range to create an image feature search space.

[0084] Specifically, the electronic device maps the initial color features, initial texture features, and initial shape features to a discrete value range to construct an image feature search space, thereby converting the continuous feature values into discrete states, which is convenient for subsequent encoding and processing using qubits. Taking the color features as an example, the value range of each channel is quantized into several levels, so that the color information can be classified, making it possible to more clearly distinguish and process different color features in the subsequent search process.

[0085] Step b13: Formulate an encoding rule according to the discrete value range of the image feature search space.

[0086] Specifically, the electronic device can use qubits to encode each feature dimension in the image feature search space. A single qubit can be in the state of 0, 1, or a superposition of both. Through the combination of multiple qubits, a high-dimensional feature space can be represented.

[0087] Then, the electronic device assigns corresponding encoding rules to each qubit according to the discrete value range of the image feature search space. For example, for a binary feature, a single qubit can be used to represent it, where the 0 state represents a feature value of 0 and the 1 state represents a feature value of 1; for a multi-valued feature, multiple qubits can be used for encoding.

[0088] Step b14: Randomly assign an initial state to each initial qubit.

[0089] Specifically, an initial state is randomly assigned to each initial qubit. Usually, it is put in a superposition state of 0 and 1. This is done because there is no prior information on which feature combination is optimal before starting to search for feature combinations, and random initialization allows the algorithm to explore widely in the entire feature space. Using quantum gate operations such as Hadamard gates to implement the initialization of the superposition state of qubits makes use of the characteristics of quantum gate operations to accurately control the state of qubits.

[0090] Step b15, initialize an initial quantum population.

[0091] Among them, the initial quantum population includes multiple initial quantum states. Among them, each initial quantum state is represented by a group of initial qubits and represents a possible image feature combination.

[0092] Specifically, the electronic device can initialize an initial quantum population according to the initial state randomly assigned to each initial qubit.

[0093] Step b16, calculate the fitness evaluation results of the image feature combinations corresponding to the initial quantum states in the initial quantum population according to the fitness function.

[0094] Specifically, the electronic device can generate a fitness function based on the recognition accuracy of the lesions, and then, based on the fitness function, calculate the fitness evaluation results of the image feature combinations corresponding to the initial quantum states in the initial quantum population. By evaluating the recognition accuracy of the feature combinations corresponding to different quantum states for the lesions, it is possible to know which feature combinations are more conducive to the recognition of lesions.

[0095] Step b17, calculate the rotation angle of each initial qubit according to the fitness evaluation result.

[0096] Among them, the magnitude and direction of the rotation angle depend on the fitness evaluation result of the current quantum state and the fitness evaluation result of the target state.

[0097] Specifically, the electronic device can calculate the difference between the fitness evaluation result and the target fitness. Specifically, the difference ΔF = Ftarget - Fcurrent. Where Ftarget is the target fitness and Fcurrent is the fitness evaluation result. Then, determine the direction and magnitude of the rotation angle θ according to the positive or negative and magnitude of the difference ΔF. For example: If ΔF>0, it means that there is still room for improvement in the current quantum state. In order to approach the target state, the rotation angle is determined according to a certain proportional relationship. A proportionality coefficient k (0 < k < 1) can be set. And determine the rotation direction according to the characteristic differences between the current quantum state and the target state (for example, by comparing the values of each characteristic in the characteristic combination corresponding to the quantum state). If ΔF = 0, it means that the current quantum state has reached the target state or is in a local optimum. At this time, a smaller random rotation angle can be selected to avoid being trapped in the local optimum and continue to explore other possible characteristic combinations. If ΔF < 0, it indicates that the current quantum state may deviate from the target state and needs to be adjusted in the opposite direction. Similarly, determine the magnitude of the rotation angle according to |ΔF| and the proportionality coefficient k, and determine the appropriate rotation direction.

[0098] Step b18, perform a quantum rotation gate operation on each initial qubit according to the calculated rotation angle to obtain each updated qubit and the updated quantum state composed of the updated qubits.

[0099] Specifically, for each initial qubit , multiply it by the quantum rotation gate matrix R(θ), where θ is the rotation angle, to obtain the updated qubit . The specific calculation process is as follows:

[0100] Through the above calculation, the updated qubit is obtained, and its coefficients and reflect the change of the qubit state caused by the rotation operation.

[0101] Step b19, measure each updated quantum state to convert the updated quantum state into classical information and obtain the classical state.

[0102] Specifically, the electronic device can measure the updated quantum state based on a dedicated quantum measurement device (such as the measurement module in a quantum computer). These devices interact with the qubits to achieve the measurement of the quantum state. The measurement device will output the corresponding measurement result according to the state of the qubit.

[0103] The measurement result is the eigenstate (|0〉 or |1〉) of the qubit, and these eigenstates can directly correspond to the binary bits (0 or 1) in the classical information. For the measurement results of multiple qubits, combine the measurement results of each qubit in sequence to obtain a classical binary string, that is, the classical state. For example, when measuring an updated quantum state containing n qubits, the measurement result is |b1b2…bn〉 (bi ∈ {0, 1}, i = 1, 2, …, n), and converting it into classical information is the binary string b1b2…bn. This binary string can be further mapped to specific image feature combination information according to the pre-set coding rules.

[0104] Through the above measurement and transformation operations on the updated quantum state, the quantum state information obtained from quantum computing is converted into classical information, realizing the conversion from the quantum state to the classical state, which lays a foundation for further using this information to solve practical problems.

[0105] Step b110: Decode the classical state into the initial image feature values according to the encoding rule.

[0106] Specifically, the encoding rule is formulated to map various values of image features into a form that can be processed by quantum computing (such as the state of quantum bits). The encoding rules include binary encoding, numerical encoding, one-hot encoding, etc. The electronic device needs to determine the image features corresponding to each part in the classical state according to the encoding rule. For example, if the encoding rule is to encode the color feature, texture feature, and shape feature with n quantum bits respectively, then the binary string of the classical state can be divided according to a certain number of bits, corresponding to different features respectively. Suppose the color feature is encoded with the first m1 bits, the texture feature is encoded with the next m2 bits, and the shape feature is encoded with the last m3 bits (m1 + m2 + m3 = n), we can split the classical state string into corresponding parts.

[0107] If binary encoding is adopted, for the encoding part of the color feature, convert the binary number into a decimal number, and then determine the corresponding color feature values such as hue, saturation, or brightness according to the previously divided discrete intervals. For example, if the encoding of the hue value is represented by an 8-bit binary number with a range of [0, 255], corresponding to the actual hue value range of [0, 360], after converting the binary number into a decimal number x, the actual hue value y = 255 / 360 * x.

[0108] For the features encoded with numerical encoding (such as shape features), directly determine the feature values according to the corresponding relationship between the numerical values and the feature categories in the encoding rule. If it is one-hot encoding, find the position with a value of 1 in the encoding vector, and determine the feature values according to the corresponding relationship between this position and the feature values. For example, for the one-hot encoding vector [0, 1, 0] of the color feature, it represents the color value corresponding to the second position, and determine the specific color feature value (such as green) according to the encoding rule.

[0109] Finally, integrate the feature values decoded from each part in the classical state to form a complete set of initial image feature values. This set contains various image feature information such as color feature values, texture feature values, and shape feature values, which can reflect the basic features of the crop image.

[0110] Step b111: Screen the initial image feature values to obtain the target image feature values.

[0111] Specifically, the electronic device can check whether there is redundancy among the initial image feature values. If there is a strong linear correlation (correlation coefficient close to 1 or -1) between two or more features, it indicates that they may contain similar information and there is redundancy. For example, in color features, two components in certain color spaces may be highly correlated, and only one of them needs to be retained.

[0112] Then, the electronic device uses dimensionality reduction techniques such as principal component analysis to remove redundant features and obtain the target image feature values. PCA can transform multiple correlated features into a few uncorrelated principal components, which can retain most of the information of the original features. By selecting an appropriate number of principal components, the purpose of removing redundancy and simplifying the feature space can be achieved.

[0113] Step b2: Use a preset encoder to encode the crop spectral data to obtain a low-dimensional sparse representation.

[0114] Among them, the low-dimensional sparse representation highlights the specific spectral band features related to pests and diseases.

[0115] Specifically, the electronic device inputs the crop spectral data into the preset encoder. During the forward propagation of the preset encoder, the crop spectral data is processed through a series of neuron layers. Each layer will transform the crop spectral data and gradually extract higher-level features. For the crop spectral data, the preset encoder will learn the correlation and potential structure between different spectral bands. Through operations such as convolutional layers, the information of spectrally adjacent bands can be integrated to extract more representative features. At the same time, in order to achieve a low-dimensional sparse representation, some regularization methods may be adopted, such as L1 regularization, which can make most elements in the feature representation learned by the preset encoder zero, thus achieving sparsity and highlighting the key features related to pests and diseases.

[0116] After multiple layers of processing by the preset encoder, a low-dimensional representation of the crop spectral data is finally obtained. This low-dimensional representation is the mapping of the original high-dimensional spectral data in the low-dimensional space. It retains the most important information in the original data and, through the sparsity constraint, highlights the specific spectral band features related to pests and diseases. For example, in the low-dimensional representation, the feature values corresponding to the spectral bands sensitive to pests and diseases may be relatively large, while the feature values corresponding to other bands unrelated to pests and diseases may be suppressed to values close to zero.

[0117] Step b3: Arrange the crop thermal infrared imaging data in chronological order, and take each pixel point in each thermal infrared image in the crop thermal infrared imaging data as a node.

[0118] Specifically, the electronic device can arrange the crop thermal infrared imaging data in chronological order. Then, each thermal infrared image is parsed into a two-dimensional array, and each array element represents a pixel point. Each pixel point in the thermal infrared image is regarded as a node.

[0119] Step b4, each node corresponds to a specific spatio-temporal position, and the attribute of the node is the temperature value at the corresponding moment of the spatio-temporal position.

[0120] Specifically, each node corresponds to a specific spatio-temporal position, which is jointly determined by the spatial coordinates (row and column indices) of the pixel point in the image and the shooting time of the image. Then, the electronic device converts the pixel value into the actual temperature value according to the calibration parameters of the thermal infrared camera. Different thermal infrared cameras may have different conversion formulas. The converted temperature value is used as the attribute of the node. A dictionary or a custom data structure can be used to store the information of the node, where the key is the identifier of the node (such as the combination of spatio-temporal positions), and the value is the corresponding temperature value.

[0121] Step b5, determine the edge between two nodes according to the temperature change relationship between two nodes at adjacent time points and adjacent spatial positions.

[0122] Specifically, for each node, it is necessary to find other nodes at adjacent time points and adjacent spatial positions. Spatially, adjacent positions usually refer to the four adjacent pixels above, below, left, and right of the pixel point; temporally, adjacent time points refer to the images taken before and after. If two nodes meet the adjacent conditions, an edge is established between them, indicating that there is a temperature change relationship between them.

[0123] Step b6, determine the weight of the edge according to the absolute value of the temperature difference between two nodes, and generate a spatio-temporal graph.

[0124] Specifically, for the two nodes connected by the determined edge, calculate the absolute value of their temperature difference, and use the calculated absolute value of the temperature difference as the weight of the edge to generate a spatio-temporal graph.

[0125] Step b7, perform feature fusion on the target image feature value, low-dimensional sparse representation, and spatio-temporal graph to generate an initial fusion feature.

[0126] Specifically, the electronic device can encode the feature values of the target image, low-dimensional sparse representation, and spatio-temporal graph respectively into a form of "genes" that is convenient for operation. For the numerical target image feature value and low-dimensional sparse representation, binary encoding can be used. For the spatio-temporal graph feature, since it contains information of nodes and edges, information such as node attributes and edge weights can be serially encoded.

[0127] Randomly combine the above encoded characteristic genes to generate a certain number of individuals, forming an initial population. Each individual represents a possible feature fusion method. Select individuals for crossover operation from the current population according to a certain selection strategy. For the selected individuals, randomly determine the crossover points and exchange some of their gene segments to generate two new individuals. Then, perform mutation operation on randomly selected individuals from the population with a certain mutation probability (usually a relatively small value, such as 0.01 - 0.1). For the selected individuals, randomly determine the mutation positions in the gene sequence. At the determined mutation positions, randomly change the genes. The mutation operation can introduce random changes into the population, prevent the algorithm from falling into local optimum, discover some unexpected feature patterns, and increase the diversity of feature combinations.

[0128] Design a fitness function according to task objectives such as the accuracy of crop pest and disease diagnosis and the reliability of growth status assessment. For example, input the fused features into a trained pest and disease diagnosis model, and use the diagnosis accuracy rate of the model for test samples as the fitness value. Calculate the fitness values of each individual in the population, and sort the individuals from high to low according to the fitness. According to the set screening ratio, eliminate the individuals with lower fitness and retain the individuals with higher fitness to enter the next generation population. Through this screening mechanism, retain and strengthen the feature combinations that have a positive effect on the task, gradually eliminate the feature combinations that are not conducive to the task objectives, and promote the continuous evolution of the fused features.

[0129] Repeat the crossover, mutation, and screening operations to continuously generate new populations, enabling the feature combinations to continuously evolve in the iterative process. Each generation of population inherits the excellent features of the previous generation and introduces new feature patterns through crossover and mutation, gradually developing towards a better feature fusion method. Set iteration termination conditions, such as reaching the preset maximum number of iterations, the fitness value not significantly improving for multiple consecutive generations, etc. When the termination conditions are met, select the individual with the highest fitness from the last generation population, decode the corresponding feature combination, and output it as the final initial fused feature.

[0130] Step S2032, perform feature extraction on the initial fused feature based on the spatio-temporal attention mechanism to generate the target fused feature.

[0131] Specifically, the electronic device can arrange the initial fusion features in chronological order to construct a feature sequence. Then, the electronic device can use a Recurrent Neural Network (RNN) and its variants to model the feature sequence. After being processed by networks such as RNN, the feature representation at each time step is obtained. Then, a temporal attention module is used to calculate the attention weights in the time dimension. The calculation process is as follows: Let the feature sequence after being processed by RNN be H = [h1, h2, …, hT], where ht represents the feature vector at the t-th time step, and T is the total number of time steps. Input ht into the MLP to obtain the attention score et = wTtanh(Wht + b), where W and w are weight matrices and vectors, and b is the bias vector. Normalize the attention scores to obtain the attention weights in the time dimension .

[0132] Then, the electronic device weights and sums the features at each time step according to the calculated temporal attention weights to obtain the aggregated features in the time dimension . This step enables the model to focus on the features at the moments that are more important for the task in the time series

[0133] In addition, for the initial fusion features containing spatial information (such as the spatio-temporal map features corresponding to thermal infrared imaging data), the electronic device can divide them according to spatial positions. For the features of each spatial unit, a spatial attention module is used to calculate its importance. Let the feature of a certain spatial unit be x i , and obtain the feature map f i = σ(W conv* x i + b conv ) through a convolutional layer and an activation function, where W conv is the convolutional kernel, b conv is the bias, and σ is the activation function. Input the feature map f i into a global average pooling layer to obtain a scalar value g i , representing the comprehensive feature intensity of this spatial unit. Then, calculate the spatial attention weights through a fully connected layer and an activation function, where N is the total number of spatial units. According to the spatial attention weights, weight and sum the features of each spatial unit to obtain the aggregated features in the spatial dimension . This can highlight the features in the regions that are crucial for the task in space

[0134] Finally, the electronic device fuses the aggregated features h time in the time dimension and the aggregated features h space in the spatial dimension. For the fused feature h fusionPerform further processing, such as performing feature transformation through a fully connected layer to obtain the final target fusion feature.

[0135] Step S2033: Input the target fusion feature into a preset pest and disease level determination model, and output the current pest and disease type and the current pest and disease level corresponding to the crop.

[0136] Specifically, the above step S2033 may include the following steps: Step c1: Obtain the target knowledge graph.

[0137] Among them, the target knowledge graph is composed of crop variety characteristics, pest and disease transmission laws, and red soil slope environment data.

[0138] Specifically, the electronic device can widely collect various data related to crop health. Then, use natural language processing technology and data mining algorithms to extract knowledge from the collected data. Extract entities from text data, such as crop varieties, pest and disease names, environmental factors, etc.; identify the relationships between entities, such as "a certain crop variety is susceptible to a certain pest and disease", "the transmission speed of a certain pest and disease increases at a specific temperature", etc.; extract attribute information, such as the damage symptoms of pests and diseases, the occurrence probability, and the associated values of environmental factors. Finally, the electronic device fuses the extracted knowledge, eliminates duplicate and contradictory information. Represent the knowledge in a standard format such as the Resource Description Framework (RDF), and store it in a graph database (such as Neo4j) to construct a crop health knowledge graph. Such a graph structure can intuitively display the relationships between various types of knowledge, facilitating subsequent query and analysis.

[0139] Step c2: Input the target fusion feature into a preset pest and disease determination model, and the preset pest and disease determination model outputs an initial determination result.

[0140] Among them, the initial determination result includes the initial pest and disease type and the initial pest and disease level corresponding to the crop.

[0141] Specifically, the electronic device inputs the target fusion feature into a preset pest and disease determination model. First, it passes through the input layer of the preset pest and disease determination model, and then feature extraction and abstraction are performed in the hidden layer. Taking CNN as an example, the convolution kernel of the convolutional layer slides on the feature map to extract local features related to pests and diseases, such as the edge texture and color patches of disease spots; the pooling layer reduces the dimension of the extracted features, reducing the computational amount while retaining important features. For RNN and its variants, through the cyclic connection between neurons, the time series features are analyzed to capture the dynamic trends of pests and diseases over time, such as the spread speed of the disease and the change in the reproduction cycle of pests. As the number of network layers increases, the model continuously abstracts and combines features to form more advanced and discriminative feature representations.

[0142] After multiple layers of processing, features of different types and levels are fused in the preset pest and disease determination model. These fused features contain multi-dimensional information such as the spatial, temporal, and spectral information of crop pests and diseases. The preset pest and disease determination model makes comprehensive decisions based on these features. The fused features are mapped to the output layer through a fully connected layer, and the number of neurons in the output layer is set according to the types and grades of pest and disease types. After the result of the output layer is processed by an activation function (such as the Softmax function, which converts the output of the neuron into a probability distribution), the probability values of the crop belonging to different pest and disease types are obtained. The preset pest and disease determination model selects the category with the highest probability value as the initially determined pest and disease type.

[0143] Step c3: Taking the initial pest and disease type as the core, construct a query statement by combining the crop variety, growth stage, surrounding environment data, and soil data.

[0144] Specifically, the electronic device can clarify the specific query target according to the initial pest and disease type. For example, if it is wheat rust, the query target may be to obtain the cause of the disease, symptoms, control methods, and precautions for control under the current crop variety, growth stage, and environmental conditions.

[0145] Then, the electronic device reasonably incorporates the crop variety, growth stage, surrounding environment data, and soil data into the query statement to construct the query statement. These data elements can be arranged in a certain logical order to make the query statement clearer and more organized. For example: "For [specific crop variety], at the [growth stage], with the surrounding environment temperature of [X] °C, humidity of [X]%, light duration of [X] hours, wind force of [X] level, and [surrounding vegetation situation] around the farmland, with the soil pH of [X], fertility status of [X], and air permeability of [X], for the [initial pest and disease type], please query its cause of disease, symptoms, and effective control methods.

[0146] Step c4: Query in the target knowledge graph based on the query statement to obtain the query result.

[0147] Among them, the query result includes the occurrence probability and severity level of the initial pest and disease type.

[0148] Specifically, the electronic device converts a query statement in natural language form into a graph query language supported by the target knowledge graph. During the conversion process, according to the structure of the knowledge graph and the relationships between nodes and edges, a corresponding query expression is constructed. The converted graph query language is input into the graph database to perform a query operation. The graph database traverses the knowledge graph according to the query statement, searches for nodes and edges that meet the conditions, and extracts relevant information. During the query process, the integration of information of multiple nodes and edges may be involved. Through the association relationships of the target knowledge graph, information from different sources is associated and analyzed to obtain more comprehensive knowledge.

[0149] Extract information related to the occurrence probability and severity level of the initial pest and disease types from the query results. This information may be stored in the knowledge graph in different forms, such as the attribute values of nodes, the weights of edges, etc. Finally, further processing is performed on the extracted results, such as data formatting, adjustment of probability values (calibration according to actual situations), clarification of severity levels, etc. Finally, the processed results are presented in a suitable form, such as presenting information such as the initial pest and disease types, occurrence probabilities, and severity levels in a table form for the convenience of users to understand and use.

[0150] Step c5, fuse the initial determination result and the query result to obtain the current pest and disease and the current pest and disease level.

[0151] Specifically, the electronic device can assign different weights to the initial determination result and the query result according to the evaluation of the initial determination result and the query result. Then, according to the weights assigned to the initial determination result and the query result, the initial determination result and the query result are fused to obtain the current pest and disease and the current pest and disease level.

[0152] For example, if the preset pest and disease determination model has been verified through a large number of experiments and has a high accuracy, a higher weight can be assigned to the initial determination result; and if the data of the target knowledge graph is rich and authoritative, an appropriate weight can also be assigned to the query result accordingly.

[0153] Step S204, based on the surrounding environment data and soil data, determine the reasons for the drought level and the reasons for the pest and disease level corresponding to the crops.

[0154] Specifically, the above step S204 may include the following steps: Step S2041, using the current drought level, the current pest and disease level, the surrounding environment data, and the soil data as nodes, and according to the relationships between the current drought level, the current pest and disease level, the surrounding environment data, and the soil data, connect the corresponding nodes with directed edges to generate a directed acyclic graph.

[0155] Specifically, the electronic device can use the drought level, the current pest and disease level, the surrounding environment data, and the soil data as nodes. Then, based on the relationship between the current drought level and other nodes, the electronic device connects the current drought level node and the corresponding nodes with directed edges. For example, drought may affect the soil moisture content, causing the soil to dry out. Therefore, there is a directed edge from the current drought level node to the soil moisture content node in the soil data, indicating the impact of drought on the soil moisture content.

[0156] Based on the relationship between the current pest and disease level and other nodes, the electronic device connects the current pest and disease level node and the corresponding nodes with directed edges. For example, the occurrence and development of pests and diseases may be affected by surrounding environmental factors. At the same time, pests and diseases may also have an impact on crop growth, which in turn affects soil fertility. Because diseased or pest-infested crops may not be able to fully absorb soil nutrients, resulting in changes in soil fertility, there is a directed edge from the current pest and disease level node to the soil fertility node in the soil data.

[0157] In addition, based on the relationship between the surrounding environment data and the soil data, the electronic device connects the corresponding nodes with directed edges. For example, the precipitation in the surrounding environment directly affects the soil moisture content. Therefore, there is a directed edge from the precipitation node in the surrounding environment data to the soil moisture content node in the soil data.

[0158] Finally, based on the determined nodes and directed edges, the electronic device represents each node graphically to generate a directed acyclic graph.

[0159] Step S2042: Input the directed acyclic graph into the feature extraction network in the preset causal relationship model.

[0160] Specifically, the electronic device can input the directed acyclic graph into the feature extraction network in the preset causal relationship model.

[0161] Step S2043: The feature extraction network extracts features from the directed acyclic graph to obtain the node features, edge features, and graph structure features corresponding to the directed acyclic graph, and inputs the node features, edge features, and graph structure features into the decision network in the preset causal relationship model.

[0162] Specifically, the feature extraction network extracts features from each node in the directed acyclic graph to determine the numerical features, category features corresponding to the nodes, and the association information between adjacent nodes. Then, the numerical features, category features corresponding to the nodes, and the association information between adjacent nodes are fused to generate node features. Then, the feature extraction network analyzes the attributes of the edges in the directed acyclic graph, such as the strength of the causal relationship represented by the edge, the direction of influence, etc., to obtain its initial features as an edge. The feature extraction network will consider the context information of the edge in the entire directed acyclic graph, and through interaction with adjacent edges and nodes, learn more comprehensive features of the edge, so as to dynamically adjust the weights corresponding to the initial features of the edge, thereby obtaining edge features.

[0163] The feature extraction network will analyze the topological structure of the directed acyclic graph and extract the overall structural features of the graph. For the hierarchical structure existing in the directed acyclic graph, the feature extraction network will learn the structural features of different levels in a hierarchical manner. Finally, through the analysis of the topological structure and hierarchical structure, the feature extraction network will generate a global feature vector that can represent the entire directed acyclic graph, that is, the graph structure feature. This vector contains the structural information of the graph, the relationship information between nodes, etc., and is an overall abstract representation of the graph.

[0164] Then, the extracted node features, edge features, and graph structure features are integrated to form a comprehensive feature vector. The integrated feature vector is input into the decision network in the preset causal relationship model. The decision network is usually a neural network-based structure, such as a multi-layer perceptron or a recurrent neural network, etc. The decision network will perform calculations and inferences based on the input feature vector to predict the causal relationships between different factors, the growth trends of crops, and the possible decision-making measures to be taken, etc.

[0165] Step S2044, the decision network determines the causes of the drought level and the current level of insect and disease damage based on the node features, edge features, and graph structure features.

[0166] Specifically, the above step S2044 may include the following steps: Step d1, construct the initial state corresponding to the state space in the decision network according to the node features, edge features, and graph structure features.

[0167] Specifically, the decision-making network can fuse node features, edge features, and graph structure features, and can concatenate these three vectors in a splicing manner to form a comprehensive feature vector. Then, the fused comprehensive feature vector is mapped into the state space of the decision-making network as the initial state. The state space is an abstract space in the decision-making network used to describe the system state, and the setting of the initial state determines the starting point of the decision-making process. By mapping the features of the directed acyclic graph into the state space, the decision-making network can perform subsequent reasoning and decision-making based on this information, analyze the causal relationships between different factors and their impacts on crop growth.

[0168] Step d2: Based on the initial state, output an initial action from the action space corresponding to the decision-making network.

[0169] Among them, the initial action represents an initial estimated value of the influence amount of each factor on the current drought level and the current pest and disease level.

[0170] Specifically, when the decision-making network receives the initial state information, the initial state information is transmitted from the input layer to the hidden layer. In the hidden layer, the neurons perform weighted summation on the input initial state and perform non-linear transformation through activation functions (such as ReLU, Sigmoid, etc.), so as to extract higher-level and more abstract feature representations. These feature representations contain an in-depth understanding of the current state of the crops and the relationships between various factors.

[0171] After being processed by the hidden layer, the information is transmitted to the output layer. The neurons in the output layer calculate the scores or probabilities of each action in the action space according to the output results of the hidden layer. Finally, according to the action scores or probabilities, the decision-making network selects the action with the highest score or the largest probability in the action space as the initial action output.

[0172] Step d3: The decision-making network starts from the nodes that have an impact on the current drought level and the current pest and disease level, and derives the causal transmission path along the directed edges in the graph.

[0173] Specifically, starting from the initial node, the decision-making network gradually traces the causal relationships between various factors according to the direction of the edges in the directed acyclic graph. The directed edges represent the direction of influence between factors. For example, a directed edge from the temperature node to the current pest and disease level node indicates that changes in temperature will affect the current pest and disease level. The decision-making network will analyze the effects of each node on subsequent nodes along these directed edges in sequence. During the derivation process, the decision-making network will consider the interaction effects of multiple factors. A node may be affected by multiple precursor nodes and may also affect multiple subsequent nodes. For example, the soil fertility node may be affected by multiple factors such as fertilization amount and rainfall, and it will in turn affect the growth status of crops, thereby affecting the current pest and disease level. The decision-making network will comprehensively analyze these complex interaction relationships and accurately derive the causal transmission path. The decision-making network will perform multi-level derivations, gradually expanding from directly affected nodes to indirectly affected nodes. For example, changes in temperature may first affect the physiological state of crops (such as affecting the photosynthesis rate), then affect the resistance of crops, and ultimately affect the current pest and disease level. The decision-making network will dig deep into the potential causal relationships between various factors along such multi-level paths. During the derivation process, the decision-making network will record the nodes and edges passed through to form a complete causal transmission path. These paths can be represented in the form of a graph or stored using data structures (such as linked lists, trees, etc.) for subsequent analysis and processing.

[0174] Step d4, determine the key nodes from each node according to the strength of the edge features and the position information of each node in the causal chain.

[0175] Specifically, the strength of the edge features reflects the tightness or influence magnitude of the causal relationship between two nodes. The position information of a node in the causal chain describes the role and status of the node in the entire causal relationship network. Evaluate and sort the strengths of all edge features. Higher-strength edges indicate a stronger causal relationship between the two nodes they connect, and the nodes connected by these edges are more likely to be key nodes. The directionality of the edges is also important because it determines the direction of causal relationship transmission. For an edge pointing from the initial node to other nodes, its strength reflects the promoting effect of the initial node on subsequent nodes; while for an edge pointing to the target node, its strength reflects the influence degree of various factors on the target node.

[0176] The starting node is usually the origin of the causal relationship and plays an important driving role in the entire causal chain. If a starting node is connected to multiple subsequent nodes by high-strength edges, then it is very likely to be a key node. Intermediate nodes play a connecting role in the causal chain. If an intermediate node is connected to multiple high-strength edges and the causal relationship directions of these edges are different, that is, it receives influences from multiple upstream nodes and transmits influences to multiple downstream nodes, then this intermediate node is often a key node. The end node is the ultimate direction of the causal relationship. Although they do not directly affect other nodes themselves, they are the goals of the entire causal chain. The intensity of the edge features connected to the end node reflects the contribution degree of each factor to the final result. If the intensity of the edge features between a certain node and the end node is particularly high, it indicates that the node has a great influence on the final result and is a key node.

[0177] Comprehensively consider the edge feature intensity and node position information, and determine the key nodes according to the preset evaluation criteria. The preset evaluation criteria can be a quantitative index. For example, set a threshold for the edge feature intensity. When the number of high-strength edges among the edges connecting a node to other nodes exceeds a certain proportion, and the position of the node in the causal chain conforms to the characteristics of a key node (such as a starting node, an important intermediate node, or a node with a strong connection to the end node), then it is determined as a key node.

[0178] Step d5, according to each key node, adjust the initial estimated values of the influence amounts of each factor in the initial action on the current drought level and pest and disease level, and obtain the target estimated values of the influence amounts of each factor on the current drought level and pest and disease level.

[0179] Specifically, for the key nodes related to the current drought level, according to the influence of the key nodes on the current drought level, adjust the initial estimated values of the influence amounts of each factor corresponding to the key nodes in the initial action on the current drought level.

[0180] For the key nodes related to the current pest and disease level, according to the influence of the key nodes on the current pest and disease level, adjust the initial estimated values of the influence amounts of each factor corresponding to the key nodes in the initial action on the current pest and disease level. In addition, multiple key nodes will simultaneously affect the current drought level and the current pest and disease level, and there are also interactions between them. Therefore, when adjusting the estimated values of the influence amounts, these synergistic effects should be comprehensively considered. When evaluating the influence amount of the irrigation volume on the current drought level, not only the direct effect of irrigation on soil moisture and the drought condition of crops should be considered, but also the indirect effect that the change in the irrigation volume may have on the current pest and disease level by affecting the growth condition of crops should be considered.

[0181] Step d6: Based on the estimated target values of each influencing quantity, determine the target factors among all factors for which the estimated target value of the influencing quantity is greater than the preset threshold.

[0182] Specifically, compare the estimated target values of the influencing quantities corresponding to each factor with the preset threshold. When the estimated target value of the influencing quantity of a certain factor on the current drought level and / or pest and disease level is greater than the preset threshold, this factor is identified as the target factor.

[0183] Step d7: Based on the target factors, determine the reasons for the drought level and the reasons for the current pest and disease level.

[0184] Specifically, comprehensively consider all target factors related to the drought level and their influencing mechanisms, and comprehensively summarize the reasons for the increase or decrease of the drought level. For example, the increase in the drought level may be the result of the combined action of multiple factors such as scarce precipitation, insufficient irrigation, loose soil texture, and high temperature leading to rapid water evaporation. For the pest and disease level, by comprehensively analyzing all relevant target factors, clarify the reasons for the change in the pest and disease level, such as high pest density, unreasonable pesticide use, poor crop disease resistance, and high field humidity conducive to the growth of pathogens, which jointly affect the occurrence and development degree of pests and diseases.

[0185] Step S205: Based on the current drought level and the reasons for the drought level, output an irrigation strategy; the irrigation strategy includes an intermittent supplementary irrigation mode and a continuous irrigation mode.

[0186] For this step, please refer to the above introduction to step S105 and will not be elaborated here.

[0187] Step S206: Based on the current pest and disease type, the current pest and disease level, and the reasons for the current pest and disease level, output the corresponding pest and disease control measures for the crops.

[0188] For this step, please refer to the above introduction to step S106 and will not be elaborated here.

[0189] The crop management method for red soil slopes provided by the embodiments of the present application inputs crop image data, crop spectral data, crop thermal infrared imaging data, soil physical property data, soil chemical property data, and moisture property data into a preset drought level determination model. This avoids the limitation of relying solely on a single data type for judgment, thereby more accurately evaluating the drought level of crops. The preset drought level determination model extracts features from the input data and performs feature fusion to obtain a fused feature vector. This makes the data easier to process and analyze. At the same time, it fully explores the potential relationships between various data, further improving the representativeness and effectiveness of the data. Based on the variational autoencoder, the fused feature vector is mapped to the mean vector and variance vector in the latent space. This realizes the dimensionality reduction processing and feature compression of the high-dimensional fused feature vector. At the same time, the variational autoencoder can also learn the probability distribution of the data, providing a probability basis for subsequent sampling operations. Sampling processing is performed based on the mean vector and variance vector to obtain at least one sampling point. Different feature combinations can be obtained from the latent space, avoiding being trapped in local optimal solutions and improving the generalization ability of the model. In each layer of the preset decoder, the similarity between each sub-feature in the sampling point and a preset drought key feature template is calculated, which can accurately measure the degree of association between each sub-feature and the drought key feature. According to the calculated similarity, the weight information of each sub-feature is determined. Thus, the importance of each sub-feature in the feature representation can be reasonably allocated according to the degree of association between the sub-feature and the drought key feature. Based on the weight information corresponding to each sub-feature, each sub-feature is weighted to obtain a weighted vector. Through the weighting process, the sub-features that have an important impact on the drought level judgment are enhanced in the weighted vector. The weighted vector undergoes a convolution operation through the convolutional layer in the preset decoder to generate a target feature vector. The convolutional layer can effectively extract the local information and spatial structure of the features. Through these operations, the generated target feature vector can comprehensively consider the weight information of each sub-feature and their mutual relationships, and more comprehensively and accurately represent the drought status of the crops. Finally, the current drought level output based on such a target feature vector will be more accurate and reliable, providing strong support for decision-making in agricultural production. Based on the target feature vector, the current drought level corresponding to the crop is output. It can directly provide a clear decision-making basis for agricultural production. Accurate drought level assessment can help farmers timely understand the moisture status of crops, take corresponding irrigation or other drought resistance measures, contribute to the rational use of water resources, improve the yield and quality of crops, and reduce losses caused by drought.

[0190] In addition, feature extraction is performed on crop image data to generate initial color features, initial texture features, and initial shape features, providing a rich information basis for subsequent accurate identification of pests and diseases. The initial color features, initial texture features, and initial shape features are mapped to a discrete value range to create an image feature search space. This makes the representation of features more standardized and orderly, while also restricting the value range of features and reducing the uncertainty of features. Encoding rules are formulated according to the discrete value range of the image feature search space; this provides a clear basis for the encoding and decoding of quantum bits. An initial state is randomly assigned to each initial quantum bit, which allows the algorithm to start searching from multiple different starting points in the search space, increasing the randomness and comprehensiveness of the search. An initial quantum population is initialized. According to the fitness function, the fitness evaluation results of the image feature combinations corresponding to the initial quantum states in the initial quantum population are calculated. The fitness function can quantitatively evaluate the effectiveness of each image feature combination for identifying crop pests and diseases. The rotation angles of each initial quantum bit are calculated based on the fitness evaluation results. It can adaptively adjust the search direction according to the current search situation and evolve towards a direction with higher fitness, so as to find the optimal image feature combination more quickly. Quantum rotation gate operations are performed on each initial quantum bit according to the calculated rotation angles to obtain each updated quantum bit and the updated quantum state composed of the updated quantum bits, thus realizing an efficient exploration of the search space. Each updated quantum state is measured to convert the updated quantum state into classical information, obtaining a classical state. According to the encoding rules, the classical state is decoded into the initial image feature values. The conversion from quantum information to image features is realized. The initial image feature values are screened to obtain the target image feature values. Through screening, some features that contribute less to pest and disease identification can be removed, further optimizing the image features and improving the quality and representativeness of the features. Then, the preset encoder is used to encode the crop spectral data to obtain a low-dimensional sparse representation, enabling the preset pest and disease level determination model to focus more on the spectral regions sensitive to pests and diseases. The crop thermal infrared imaging data is arranged in chronological order, and each pixel point in each thermal infrared image in the crop thermal infrared imaging data is used as a node; each node corresponds to a specific spatio-temporal position, and the attribute of the node is the temperature value at the corresponding moment of the spatio-temporal position; according to the temperature change relationship between two nodes at adjacent time points and adjacent spatial positions, the edges between the two nodes are determined; according to the absolute value of the temperature difference between the two nodes, the weights of the edges are determined to generate a spatio-temporal graph. The continuity and dynamics of the data in time and space are fully considered. It can clearly depict the change trend of the crop surface temperature over time and space. The target image feature values, low-dimensional sparse representation, and spatio-temporal graph are fused in features to generate initial fused features. The one-sidedness and uncertainty that may be brought by a single data source are reduced. Then, based on the spatio-temporal attention mechanism, feature extraction is performed on the initial fused features to generate the target fused features.Thus, the quality and effectiveness of the features are improved, ensuring the accuracy of the obtained target fusion features. Obtain the target knowledge graph. Input the target fusion features into a preset pest and disease determination model, and the preset pest and disease determination model outputs an initial determination result, ensuring the accuracy of the output initial determination result, providing a basic result for subsequent in-depth analysis, saving the time and effort of manual analysis, and the model judgment has a certain objectivity and accuracy. Taking the initial pest and disease type as the core, construct a query statement by combining the variety, growth stage, surrounding environment data, and soil data of the crop. Query in the target knowledge graph based on the query statement to obtain a query result. The query result further enriches the understanding of the pest and disease situation, considering the occurrence possibility and severity of pests and diseases under specific environmental and crop conditions, providing more basis for accurately evaluating the pest and disease situation. Fuse the initial determination result and the query result to obtain the current pest and disease and the current pest and disease level. It can fully combine the preliminary judgment of the model and the rich domain knowledge in the knowledge graph, complement and verify each other, so as to obtain a more accurate and practical current pest and disease and the current pest and disease level, improving the accuracy and reliability of pest and disease diagnosis, and providing a more accurate basis for subsequent effective management measures.

[0191] Finally, taking the current drought level, the current pest and disease level, the surrounding environment data, and the soil data as nodes, according to the relationships among the current drought level, the current pest and disease level, the surrounding environment data, and the soil data, use directed edges to connect the corresponding nodes to generate a directed acyclic graph. Thus, the relationships among complex multi-source data can be represented in an intuitive graphical structure. Inputting the directed acyclic graph into the feature extraction network in the preset causal relationship model can automatically extract the key features in the graph data by the network. The feature extraction network extracts features from the directed acyclic graph to obtain the node features, edge features, and graph structure features corresponding to the directed acyclic graph, and inputs the node features, edge features, and graph structure features into the decision-making network in the preset causal relationship model. It can enable the decision-making network to comprehensively consider all aspects of the data and more accurately mine the causal relationships in the data. According to the node features, edge features, and graph structure features, construct the initial state corresponding to the state space in the decision-making network to provide a basis for the subsequent decision-making process. Based on the initial state, output the initial action from the action space corresponding to the decision-making network to provide a starting point for more accurate subsequent analysis and adjustment, which helps to initially judge which factors may have a greater impact on the drought and pest and disease levels. The decision-making network starts from the nodes that affect the current drought level and the current pest and disease level, and follows the directed edges in the graph to deduce the causal transmission path. It can deeply understand the causal mechanism of the entire system, discover potential causal relationship paths, and provide strong support for accurately determining the causes of the drought and pest and disease levels. According to the intensity of the edge features and the position information of each node in the causal chain, determine the key nodes from each node. It can accurately identify the factors that play a key role in the entire causal relationship. According to each key node, adjust the initial estimated values of the influence amounts of each factor on the current drought level and the pest and disease level in the initial action to obtain the target estimated values of the influence amounts of each factor on the current drought level and the pest and disease level. It can correct the biases in the initial estimates, improve the quantization accuracy of the influence degrees of each factor, and provide more reliable data support for subsequent determination of the target factors and accurate analysis of the causes of the drought and pest and disease levels. According to each target estimated value of the influence amount, thus determine the target factors whose target estimated values of the influence amount are greater than the preset threshold from each factor. It can screen out the factors that have a significant impact on the current drought level and the pest and disease level. According to the target factors, determine the cause of the drought level and the cause of the current pest and disease level, ensuring the accuracy of the determined cause of the drought level and the cause of the current pest and disease level.

[0192] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A method for managing crops on red soil slopes, characterized in that, The method includes: Obtaining crop data, surrounding environment data, and soil data corresponding to the crops planted in the target red soil slope area; the crop data includes crop image data, crop spectral data, and crop thermal infrared imaging data; the surrounding environment data includes at least one of light intensity, humidity, and wind speed; the soil data includes at least one of soil physical property data, soil chemical property data, soil biological property data, and water property data; Identifying the crop data and the soil data to determine the current drought level corresponding to the crop; Identifying the crop data to determine the current pest and disease type and the current pest and disease level corresponding to the crop; Based on the surrounding environment data and the soil data, determining the reasons for the drought level and the reasons for the pest and disease level corresponding to the crop; Based on the current drought level and the reasons for the drought level, outputting an irrigation strategy; the irrigation strategy includes an intermittent supplementary irrigation mode and a continuous irrigation mode; Based on the current pest and disease type, the current pest and disease level, and the reasons for the current pest and disease level, outputting pest and disease control measures corresponding to the crop.

2. The method according to claim 1, characterized in that, The soil data includes soil physical property data, soil chemical property data, and water property data. The identifying the crop data and the soil data to determine the current drought level corresponding to the crop includes: Inputting the crop image data, the crop spectral data, the crop thermal infrared imaging data, the soil physical property data, the soil chemical property data, and the water property data into a preset drought level determination model; The preset drought level determination model extracts features from the input data and performs feature fusion to obtain a fused feature vector; Extracting features from the fused feature vector to generate a target feature vector; Based on the target feature vector, outputting the current drought level corresponding to the crop.

3. The method according to claim 2, characterized in that, The extracting features from the fused feature vector to generate a target feature vector includes: Mapping the fused feature vector to a mean vector and a variance vector in the latent space based on a variational autoencoder; Performing sampling processing based on the mean vector and the variance vector to obtain at least one sampling point; the sampling point contains the mixed coding information of multi-modal data in the latent space; the mixed coding information includes multiple sub-features; Based on a preset decoder, converting and combining each of the sampling points to generate the target feature vector having the same dimension as the input data.

4. The method according to claim 3, characterized in that The converting and combining each of the sampling points based on a preset decoder to generate the target feature vector having the same dimension as the input data includes: For each layer in the preset decoder, calculating the similarity between each sub-feature in the sampling point and a preset drought key feature template; Based on the calculated similarity, determining the weight information of each sub-feature; Based on the weight information corresponding to each sub-feature, performing weighted processing on each of the sub-features to obtain a weighted vector; The weighted vector undergoes a convolution operation through a convolution layer in the preset decoder to generate the target feature vector.

5. The method according to claim 1, characterized in that, The identification of the crop data to determine the current pest and disease type and the current pest and disease level corresponding to the crop includes: Performing feature extraction on the crop image data, the crop spectral data, and the crop thermal infrared imaging data, and fusing the extracted features to generate an initial fused feature; Performing feature extraction on the initial fused feature based on a spatio-temporal attention mechanism to generate a target fused feature; Inputting the target fused feature into a preset pest and disease level determination model to output the current pest and disease type and the current pest and disease level corresponding to the crop.

6. The method according to claim 5, wherein The performing feature extraction on the crop image data, the crop spectral data, and the crop thermal infrared imaging data, and fusing the extracted features to generate an initial fused feature includes: Performing feature extraction on the crop image data to obtain a target image feature value; Encoding the crop spectral data using a preset encoder to obtain a low-dimensional sparse representation; the low-dimensional sparse representation highlights specific spectral band features related to pests and diseases; Arranging the crop thermal infrared imaging data in chronological order, and taking each pixel point in each thermal infrared image in the crop thermal infrared imaging data as a node; Each node corresponds to a specific spatio-temporal position, and the attribute of the node is the temperature value at the corresponding moment of the spatio-temporal position; Determining the edges between two nodes according to the temperature change relationship between two nodes at adjacent time points and adjacent spatial positions; Determining the weight of the edge according to the absolute value of the temperature difference between the two nodes to generate a spatio-temporal graph; Performing feature fusion on the target image feature value, the low-dimensional sparse representation, and the spatio-temporal graph to generate the initial fused feature.

7. The method according to claim 6, wherein The performing feature extraction on the crop image data to obtain a target image feature value includes: Performing feature extraction on the crop image data to generate an initial color feature, an initial texture feature, and an initial shape feature; Mapping the initial color feature, the initial texture feature, and the initial shape feature to a discrete value range to create an image feature search space; Formulating an encoding rule according to the discrete value range of the image feature search space; Randomly assigning an initial state to each initial qubit; Initializing an initial quantum population; the initial quantum population includes a plurality of initial quantum states, and each initial quantum state is represented by a group of initial qubits, representing a possible combination of image features; Calculating the fitness evaluation result of the image feature combination corresponding to each initial quantum state in the initial quantum population according to a fitness function; Calculating the rotation angle of each initial qubit according to the fitness evaluation result; the magnitude and direction of the rotation angle depend on the fitness evaluation result of the current quantum state and the fitness evaluation result of the target state; Perform a quantum rotation gate operation on each of the initial qubits according to the calculated rotation angle to obtain updated qubits and an updated quantum state composed of the updated qubits; Measure each of the updated quantum states to convert the updated quantum states into classical information and obtain a classical state; Decode the classical state into initial image eigenvalues according to the encoding rule; Screen the initial image eigenvalues to obtain target image eigenvalues.

8. The method according to claim 5, wherein The step of inputting the target fusion feature into a preset pest and disease level determination model and outputting the current pest and disease type and the current pest and disease level corresponding to the crop includes: Obtain a target knowledge graph; the target knowledge graph is composed of crop variety characteristics, pest and disease transmission laws, and red soil slope environment data; Input the target fusion feature into a preset pest and disease determination model, and the preset pest and disease determination model outputs an initial determination result; the initial determination result includes the initial pest and disease type and the initial pest and disease level corresponding to the crop; Construct a query statement with the initial pest and disease type as the core, combining the variety, growth stage of the crop, the surrounding environment data, and the soil data; Query in the target knowledge graph based on the query statement to obtain a query result; the query result includes the occurrence probability and severity level of the initial pest and disease type; Fuse the initial determination result and the query result to obtain the current pest and disease and the current pest and disease level.

9. The method according to claim 1, characterized in that, The step of determining the reasons for the drought level and the pest and disease level corresponding to the crop based on the surrounding environment data and the soil data includes: Using the current drought level, the current pest and disease level, the surrounding environment data, and the soil data as nodes, and connecting the corresponding nodes with directed edges according to the relationships between the current drought level, the current pest and disease level, the surrounding environment data, and the soil data to generate a directed acyclic graph; Input the directed acyclic graph into a feature extraction network in a preset causal relationship model; The feature extraction network extracts features from the directed acyclic graph to obtain node features, edge features, and graph structure features corresponding to the directed acyclic graph, and inputs the node features, the edge features, and the graph structure features into a decision-making network in the preset causal relationship model; The decision-making network determines the reasons for the drought level and the current pest and disease level based on the node features, the edge features, and the graph structure features.

10. The method according to claim 9, wherein The decision-making network determines the reasons for the drought level and the current pest and disease level based on the node features, the edge features, and the graph structure features, including: Construct an initial state corresponding to the state space in the decision-making network according to the node features, the edge features, and the graph structure features; Based on the initial state, output an initial action from the action space corresponding to the decision-making network; the initial action represents an initial estimated value of the influence amount of each factor on the current drought level and the current pest and disease level; The decision-making network starts from the nodes that affect the current drought level and the current pest and disease level, and derives the causal transmission path along the directed edges in the graph; According to the intensity of the edge features and the position information of each node in the causal chain, key nodes are determined from each of the nodes; According to each of the key nodes, the initial estimated values of the influence amounts of the factors in the initial actions on the current drought level and the pest and disease level are adjusted to obtain the target estimated values of the influence amounts of the factors on the current drought level and the pest and disease level; According to each of the target estimated values of the influence amounts, target factors with target estimated values of the influence amounts greater than a preset threshold are determined from among the factors; According to the target factors, the causes of the drought level and the causes of the current pest and disease level are determined.

Images