A fertilization decision method and system based on soil microbial diversity analysis

Abstract

The application relates to the technical field of agricultural intelligent fertilization, and discloses a fertilization decision-making method and system based on soil microbial diversity analysis, which comprises the following steps: obtaining soil samples, environmental parameters and crop growth data, analyzing the soil samples to obtain microbial community composition data and soil physicochemical parameters; determining the functional structure characteristics of the microbial community based on the microbial community composition data; generating soil parameter distribution characteristics based on the soil physicochemical parameters; performing weighted fusion on the functional structure characteristics and the soil parameter distribution characteristics based on a double-flow graph convolution-attention model to obtain soil microbial fusion characteristics; generating the microbial metabolic rate and the crop nutrient demand increment based on the functional structure characteristics of the microbial community, the environmental parameters, the soil parameter distribution characteristics and the crop growth data; and inputting the microbial metabolic rate, the crop nutrient demand increment, the fusion characteristics, the environmental parameters and the soil parameter distribution characteristics into a reinforcement learning model to obtain a target fertilization decision.

Description

Technical Field

[0001] This application relates to the field of intelligent agricultural fertilization technology, and includes, but is not limited to, a fertilization decision-making method and system based on soil microbial diversity analysis. Background Technology

[0002] Precision fertilization, as one of the core technologies of modern smart agriculture, has long been based on the measurement of soil physicochemical properties (such as pH value, electrical conductivity, organic matter content, available nitrogen, phosphorus and potassium, etc.), and combined with crop growth models, nutrient balance equations (such as nitrogen budget models) or empirically recommended fertilization formulas (such as soil testing and formula fertilization technology) to make decisions.

[0003] However, this paradigm has a fundamental limitation: it treats soil as a physicochemical reaction vessel, neglecting the core driving role of soil microbial communities as living engines in nutrient mineralization, nitrogen fixation and phosphorus solubilization, organic matter turnover, and rhizosphere interactions. Because microbial activity is highly dependent on environmental conditions and exhibits modal heterogeneity, fertilization plans based solely on static physicochemical indicators are insufficient to dynamically match the actual needs of soil biological processes, often leading to over- or under-fertilization. This not only reduces nitrogen, phosphorus, and potassium use efficiency (typically less than 40%) but also exacerbates environmental risks such as nitrate leaching, greenhouse gas emissions, and soil acidification. Summary of the Invention

[0004] In view of this, embodiments of this application provide a fertilization decision-making method and system based on soil microbial diversity analysis, which at least solves the problem of inaccurate fertilization amount caused by ignoring the core driving role of soil microbial communities.

[0005] The technical solution of this application embodiment is implemented as follows: In a first aspect, embodiments of this application provide a fertilization decision-making method based on soil microbial diversity analysis, applied to a fertilization decision-making system based on soil microbial diversity analysis. The system includes a data acquisition module, a data processing module, and a fertilization decision-making module. The method includes: The data acquisition module is used to acquire soil samples, environmental parameters and crop growth data of the target planting area. The soil samples are then analyzed to obtain microbial community composition data and soil physicochemical parameters. The data processing module uses the microbial community composition data to determine the functional structure characteristics of the microbial community; it generates soil parameter distribution characteristics based on the soil physicochemical parameters; and it uses a dual-flow graph convolution-attention model to perform weighted fusion of the functional structure characteristics and the soil parameter distribution characteristics to obtain soil microbial fusion characteristics. Using the fertilization decision module, based on the functional structure characteristics of the microbial community, the environmental parameters, the distribution characteristics of the soil parameters, and the crop growth data, microbial metabolic rate and crop nutrient demand increment are generated; the microbial metabolic rate, crop nutrient demand increment, soil microbial fusion characteristics, environmental parameters, and soil parameter distribution characteristics are input into a reinforcement learning model to obtain the target fertilization decision output by the reinforcement learning model.

[0006] Secondly, embodiments of this application provide a fertilization decision-making system based on soil microbial diversity analysis. The system includes: a data acquisition module, a data processing module, and a fertilization decision-making module, wherein: The data acquisition module is used to acquire soil samples, environmental parameters and crop growth data of the target planting area, and to analyze the soil samples to obtain microbial community composition data and soil physicochemical parameters. The data processing module is used to determine the functional structure characteristics of the microbial community based on the microbial community composition data; generate soil parameter distribution characteristics based on the soil physicochemical parameters; and perform weighted fusion of the functional structure characteristics and the soil parameter distribution characteristics based on the dual-flow graph convolution-attention model to obtain soil microbial fusion characteristics. The fertilization decision module is used to generate microbial metabolic rate and crop nutrient demand increment based on the functional structure characteristics of the microbial community, the environmental parameters, the soil parameter distribution characteristics, and the crop growth data; and input the microbial metabolic rate, crop nutrient demand increment, soil microbial fusion characteristics, the environmental parameters, and the soil parameter distribution characteristics into a reinforcement learning model to obtain the target fertilization decision output by the reinforcement learning model.

[0007] The beneficial effects of the technical solutions provided in this application include at least the following: By using the functional structure characteristics of microbial communities and soil physicochemical characteristics as the basis for fertilization decisions, this method solves the problem of inaccurate fertilization rates caused by neglecting the driving role of microorganisms in traditional technologies. It achieves dual-dimensional decision-making based on soil physicochemical parameters and microbial community data, improving the accuracy and scientific nature of fertilization decisions. A dual-flow graph convolution-attention model is used to accurately fuse the functional structure characteristics of microbial communities and the distribution characteristics of soil parameters, effectively mining potential correlations between data, avoiding decision-making biases based on a single data dimension, and improving the comprehensiveness and accuracy of feature extraction. Through a reinforcement learning model, the system autonomously learns the synergistic relationships between soil, microorganisms, and crops, and the output fertilization decisions can be adapted to the real-time environment and crop growth status of the target planting area, improving decision-making accuracy and flexibility. Attached Figure Description

[0008] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 A flowchart illustrating a fertilization decision-making method based on soil microbial diversity analysis, provided for an embodiment of this application; Figure 2 This is a schematic diagram of the composition structure of a fertilization decision system based on soil microbial diversity analysis, provided as an embodiment of this application. Detailed Implementation

[0009] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0010] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0011] It should be noted that the terms "first, second, and third" used in the embodiments of this application are merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0012] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of this application pertain. It should also be understood that terms such as those defined in general dictionaries should be understood to have a meaning consistent with their meaning in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0013] This application provides a fertilization decision-making method based on soil microbial diversity analysis, applied to electronic devices. The electronic devices include, but are not limited to, mobile phones, laptops, tablets, handheld internet devices, multimedia devices, streaming media devices, mobile internet devices, wearable devices, or other types of electronic devices. The functions implemented by this method can be achieved by a processor in the electronic device calling program code. The program code can be stored in a computer storage medium; therefore, the electronic device includes at least a processor and a storage medium. The processor can be used to process the fertilization decision-making process based on soil microbial diversity analysis, and the memory can be used to store the data required and generated during the fertilization decision-making process based on soil microbial diversity analysis.

[0014] Figure 1 This application provides a flowchart illustrating a fertilization decision-making method based on soil microbial diversity analysis, applicable to a fertilization decision-making system based on soil microbial diversity analysis. The system includes a data acquisition module, a data processing module, and a fertilization decision-making module. Figure 1 As shown, the method includes at least the following steps: Step S110: Use the data acquisition module to acquire soil samples, environmental parameters and crop growth data of the target planting area, analyze the soil samples to obtain microbial community composition data and soil physicochemical parameters; The data acquisition module, as the starting point of the fertilization decision-making system, has the core objective of acquiring multi-source heterogeneous data to provide basic input support for subsequent analysis. Its data sources are clearly defined, including soil samples collected from the target planting area in the field, covering data related to microbial communities and physicochemical properties; environmental parameters obtained through specific monitoring methods; and crop growth data reflecting the crop growth status. The environmental parameters include temperature, humidity, light, and precipitation, while the crop growth data includes the Normalized Difference Vegetation Index (NDVI), biomass, and yield.

[0015] Targeted collection and analysis techniques can be employed for different types of data. For soil samples, 16S rRNA or ITS (Internal Transcribed Spacer) high-throughput sequencing technology can be used to analyze the microbial community structure in the soil samples. The microbial community structure includes an abundance table of Operational Taxonomic Units (OTUs) and an abundance table of Amplicon Sequence Variants (ASVs). The carbon source metabolism capacity of the microbial community, characterized by the Average Well Color Development (AWCD) value, can also be determined for the soil samples using the Biolog Eco microplate method. The AWCD is used to characterize the carbon source utilization activity of the microbial community.

[0016] It can also simultaneously measure key physicochemical properties of the soil, such as pH, moisture, and electrical conductivity, through the soil samples to obtain soil physicochemical parameters and comprehensively acquire basic soil-related data; it can also rely on the deployed Internet of Things sensor network to realize real-time acquisition of environmental parameters such as temperature and precipitation, ensuring the timeliness and continuity of environmental data; it can also use multispectral cameras to collect the NDVI values ​​of crops and combine them with ground sensors to record crop biomass data in real time, accurately capture crop growth dynamics, and complete the collection of crop growth data.

[0017] Step S120: Using the data processing module, determine the functional structure characteristics of the microbial community based on the microbial community composition data; generate soil parameter distribution characteristics based on the soil physicochemical parameters; and perform weighted fusion of the functional structure characteristics and the soil parameter distribution characteristics based on the dual-flow graph convolution-attention model to obtain soil microbial fusion characteristics. The data processing module acquires microbial community composition data (e.g., OTU / ASV tables, AWCD values) and soil physicochemical parameters (e.g., pH, moisture, electrical conductivity). During processing, it constructs a microbial co-occurrence network based on the microbial community composition data and calculates the functional structure characteristics of the microbial community. These functional structure parameters can be characterized by microbial network topology parameters. The soil physicochemical parameters can be normalized and spatially interpolated to generate a soil parameter distribution map. Analyzing the soil parameter distribution map yields the soil parameter distribution characteristics. Simultaneously, a Bi-directional Graph Convolutional Network with Spatio-Temporal Attention (Bi-GCN-STA) model can be used to dynamically weight and fuse the microbial network topology parameters and the soil physicochemical parameters to generate a multidimensional feature vector (i.e., the soil microbial fusion feature).

[0018] Step S130: Using the fertilization decision module, based on the functional structure characteristics of the microbial community, the environmental parameters, the soil parameter distribution characteristics, and the crop growth data, generate the microbial metabolic rate and the incremental crop nutrient demand; input the microbial metabolic rate, the incremental crop nutrient demand, the soil microbial fusion characteristics, the environmental parameters, and the soil parameter distribution characteristics into the reinforcement learning model to obtain the target fertilization decision output by the reinforcement learning model.

[0019] The fertilization decision module acquires the functional structure characteristics of the microbial community, the soil microbial fusion characteristics, the environmental parameters, the soil parameter distribution characteristics, and the crop growth data. During processing, the functional structure characteristics can be calculated using Metabolic Theory of Ecology (MTE) and Particle Swarm Optimization (PSO) algorithms to determine the microbial metabolic rate and the incremental crop nutrient demand. The microbial metabolic rate, the incremental crop nutrient demand, the soil microbial fusion characteristics, the environmental parameters, and the soil parameter distribution characteristics are then input into a reinforcement learning model to generate a target fertilization decision. This reinforcement learning model can be a Deep Q-Network (DQN) or a Soft Actor-Critic (SAC) algorithm.

[0020] In the above embodiments, by using the functional structure characteristics of the microbial community and the physicochemical characteristics of the soil as the basis for fertilization decisions, the problem of inaccurate fertilization caused by neglecting the driving role of microorganisms in traditional technologies is solved. This achieves dual-dimensional decision-making based on soil physicochemical parameters and microbial community data, improving the accuracy and scientific nature of fertilization decisions. A dual-flow graph convolution-attention model is used to accurately integrate the functional structure characteristics of the microbial community and the distribution characteristics of soil parameters, effectively mining the potential correlations between data, avoiding decision-making biases based on a single data dimension, and improving the comprehensiveness and accuracy of feature extraction. Through reinforcement learning models, the collaborative relationships between soil, microorganisms, and crops are autonomously learned, and the output fertilization decisions can be adapted to the real-time environment and crop growth status of the target planting area, improving decision-making accuracy and flexibility.

[0021] In some embodiments, the microbial community composition data includes an operational taxonomic unit (OTU) abundance table, an amplicon sequence variant (ASV) abundance table, and an average color change rate (AWCD), wherein the AWCD is used to characterize the carbon source utilization activity of the microbial community; the functional structural features include a modularity index and node centrality. Step S120, "using the data processing module to determine the functional structural characteristics of the microbial community based on the microbial community composition data," includes the following steps: Step S1201: Based on the OTU abundance table and the ASV abundance table, determine the co-occurrence association between species through the Jaccard similarity coefficient; if the Jaccard similarity coefficient between species is greater than a preset association threshold, determine that there is a co-occurrence relationship between the corresponding species; use species as nodes, add edges between species with co-occurrence relationships, and construct a co-occurrence network of microbial communities. The co-occurrence associations between species can be determined using the following formula (1): Formula (1); in, This represents the set of all samples containing species i. The set of all samples of species j The Jaccard similarity coefficient between species i and species j is represented. The preset association threshold can be 0.3, that is, if the Jaccard similarity coefficient between species i and species j is greater than 0.3, it is determined that species i and species j have a co-occurrence association. At this time, an edge can be added between species i and species j. Each microbial group (i.e., species) can be used as a node in the co-occurrence network. An adjacency matrix can be constructed based on the judgment result of the co-occurrence association between species. The non-zero values ​​in the adjacency matrix correspond to the edges of the co-occurrence network.

[0022] Step S1202: Configure an initial feature vector for each species node in the co-occurrence network. The initial feature vector includes the average abundance of the corresponding species in each soil sample, the carbon source utilization capacity score of the corresponding species determined based on the AWCD, and the functional label annotation. The initial feature vector of each species node includes the average abundance of the species in each soil sample, the carbon source utilization capacity score of the corresponding species determined based on the AWCD, and the functional label annotation. The carbon source utilization capacity quantification score of a single microbial species can be calculated based on the AWCD, which characterizes the overall carbon source utilization activity of the microbial community, combined with the species' weight or contribution in the community. The functional label annotation can include nitrogen cycle-related genes. Self-loops can be added to the adjacency matrix to allow each species node to retain its own information when updating its feature vector. Symmetrical normalization of the adjacency matrix after adding self-loops can balance the feature contribution weights of nodes with different connectivity. Furthermore, Z-score normalization or Min-Max normalization can be used to standardize the feature vectors of nodes to eliminate dimensional differences between different feature dimensions.

[0023] Step S1203: The Louvain algorithm is used to iteratively optimize the community division of the microbial community to obtain the functional partitions of the microbial community; the modularity index of the functional partitions and the node centrality of species nodes in the co-occurrence network are determined. The modularity index is used to quantify the functional partition strength of the microbial community, and the node centrality is used to locate the contribution of key species in the microbial community.

[0024] The Louvain algorithm was then used to iteratively optimize the community partitioning of the microbial community. The maximum number of iterations in the Louvain algorithm is 1000, and the resolution parameter is 1.0. The functional partitioning strength and the contribution of key species in the microbial community are quantified by calculating two key parameters: the modularity index Q and node centrality. The modularity index Q is used to identify the functional partitioning characteristics of the microbial community; a higher Q value indicates more significant functional partitioning within the community. Node centrality is used to locate key species that have a significant impact on ecological functions; a higher node centrality indicates a greater impact of the species on the interactions and ecological functions of the entire community.

[0025] The modularity index Q can be calculated using the following formula (2): Formula (2); Where L is the actual number of edges, E is the expected number of edges, and M is the total number of edges (the maximum number of edges that may exist in the network).

[0026] The node centrality can be the node betweenness centrality. It can be calculated using the following formula (3): Formula (3); in, This represents the number of shortest paths from species s to species t. This represents the number of shortest paths from species s to species t via species v. For example, in a farmland sample, Q is 0.48 (significant functional zoning) and node centrality is 0.85 (key species contribute 82.3%). In this case, the modularity index and node centrality can be directly used in the fertilization decision module to calculate the microbial metabolic rate.

[0027] In the above embodiments, carbon source utilization activity is characterized by AWCD, which makes up for the shortcomings of traditional diversity indicators that only reflect species richness and cannot reflect functional potential, and realizes the quantitative assessment of microorganisms from their presence or absence to their functional strength. By using the modular index and node centrality of co-occurrence networks, the functional zoning strength and core species contribution of microbial communities are clarified, providing accurate biological parameter support for subsequent calculation of microbial metabolic rates and avoiding ambiguous judgments on the function of microbial communities.

[0028] In some embodiments, step S120, "generating soil parameter distribution characteristics based on the soil physicochemical parameters," includes the following steps: Step S1204: Normalize the soil physicochemical parameters to obtain normalized physicochemical parameters; The soil physicochemical parameters can be normalized using the following formula (4) to eliminate dimensional differences between different parameters: Formula (4); in, Represents the original soil physicochemical parameters. This represents the normalized soil physicochemical parameters (i.e., normalized physicochemical parameters). This represents the maximum value of soil physicochemical parameter X across all soil samples. This represents the minimum value of soil physicochemical parameter X among all soil samples.

[0029] Step S1205: Using the Kriging space interpolation algorithm, the normalized physicochemical parameters are interpolated to generate a spatial distribution map of soil parameters in the target planting area. The spatial distribution map of soil parameters is then analyzed to obtain the distribution characteristics of soil parameters.

[0030] Kriging spatial interpolation algorithm can also be used to interpolate the normalized physicochemical parameters to generate a spatial distribution map of soil parameters. The interpolation radius in the Kriging spatial interpolation algorithm can be 50 meters, and the variogram can be an exponential model. The 50-meter interpolation radius is suitable for the grid density of farmland soil sampling, and the exponential model can accurately fit the spatial gradient characteristics of soil parameters, ensuring the spatial continuity of the interpolation results.

[0031] Building upon this foundation, the Bi-GCN-STA (Bi-GCN-STA) model achieves dynamic fusion through two independent flows. For the microbial flow, using a microbial co-occurrence network as the basic carrier, the functional structural features are embedded into the network node features of the co-occurrence network. Utilizing the neighborhood aggregation capability of the graph convolutional layer of the Bi-GCN-STA model, the basic node features are deeply coupled with network topological features and functional structural features to extract high-order topological functional features reflecting species interaction patterns, functional group synergy, and competitive relationships. For the soil flow, using a soil parameter distribution map as input, based on the soil parameter distribution features, the spatial feature capture capability of the convolutional neural network of the Bi-GCN-STA model is utilized to calculate high-order spatial modal features of soil parameters through convolutional kernel sliding. The high-order topological functional features and the high-order spatial modal features are adaptively weighted and fused through an attention mechanism. Based on the ecological association between microbial functional structural features and soil parameter distribution, the weight allocation is dynamically adjusted to ultimately obtain soil microbial fusion features that take into account both the functional characteristics of the microbial community and the distribution characteristics of the soil environment.

[0032] It should be noted that the core function of the data processing module is to construct a microbial co-occurrence network and achieve deep fusion of multimodal data, transforming the original microbial community data and soil physicochemical parameters into high-dimensional features that can drive precise decision-making, providing processed feature data for the decision-making process of the fertilization decision module.

[0033] Through Q value ( It can accurately quantify the intensity of functional zones, breaking through the limitations of traditional α diversity indicators (such as the Shannon index, which only reflects species richness and cannot reveal functional zones); in the actual measurement, the nitrogen efficiency of high Q region (Q>0.45) is significantly higher than that of low Q region (Q<0.4).

[0034] In the above embodiments, the normalization process unifies the dimensions of different physicochemical parameters, avoiding the imbalance of feature weights caused by differences in parameter numerical ranges and improving data comparability. The Kriging interpolation algorithm can generate a continuous spatial distribution map of soil parameters based on discrete sample points, accurately reflecting the differences in physicochemical parameters of different sub-regions within the target planting area. This solves the problem of regional decision-making bias caused by traditional point sampling to area promotion, and provides a high-precision spatial basis for subsequent zonal variable fertilization.

[0035] In some embodiments, the environmental parameters include ambient temperature, the soil parameter distribution characteristics include soil pH, and the crop growth data includes crop functional gene abundance and soil nutrient supply. Step S130, "using the fertilization decision module, based on the functional structure characteristics of the microbial community, the environmental parameters, the soil parameter distribution characteristics, and the crop growth data, to generate microbial metabolic rate and incremental crop nutrient demand," includes the following steps: Step S1301: Determine the microbial metabolic rate based on the functional structure characteristics of the microbial community, the ambient temperature, the soil pH, the target temperature coefficient, the target pH coefficient, and the preset basal metabolic rate constant. The metabolic rate of microorganisms can be calculated using the temperature-pH two-factor metabolic formula, as shown in formula (5) below: Formula (5); in, This represents the preset basal metabolic rate constant. Indicates the modularity index. The weighting coefficients of the modular index are represented. Indicates node centrality. The weight coefficients representing node centrality This indicates the ambient temperature. This indicates the soil pH. Indicates the target temperature coefficient. Indicates the target pH coefficient. This indicates the rate of microbial metabolism.

[0036] In one embodiment, , (The target temperature coefficient can be dynamically optimized using the particle swarm optimization algorithm.) (The target pH coefficient can be dynamically optimized using the particle swarm optimization algorithm.)

[0037] Step S1302: Calculate the first change between the microbial metabolic rate and the basal metabolic rate constant, and calculate the difference between the first change and the second change in soil nutrient supply. Wherein, the first change is the change in metabolic rate, which can be expressed as: The second change is the change in soil nutrient supply, which can be expressed as: .

[0038] Step S1303: Determine the incremental crop nutrient demand based on the target microbial metabolic rate when the difference is minimal, the abundance of crop functional genes, and the preset nutrient conversion coefficient.

[0039] Wherein, the abundance of crop functional genes can be represented as F, and the preset nutrient conversion coefficient can be represented as The incremental crop nutrient requirement can be expressed as The incremental nutrient requirement of the crop can be calculated using the following formula (6): Formula (6); Among them, the difference mode can be minimized. The objective function is to ultimately output the increment of crop nutrient requirements. .

[0040] In the above embodiments, environmental factors such as temperature and pH are correlated with the functional and structural characteristics of microorganisms to achieve real-time quantification of metabolic rate. This solves the problem that traditional technologies cannot dynamically assess the nutrient conversion capacity of microorganisms and accurately reflects the actual nutrient supply potential of the soil. By analyzing the difference between metabolic rate and soil nutrient supply, and combining the abundance of crop functional genes to calculate the incremental demand, it ensures that the amount of fertilizer applied meets the needs of crop growth without exceeding the conversion capacity of soil microorganisms, avoiding resource waste and environmental risks caused by excessive fertilizer application, and improving fertilizer utilization efficiency.

[0041] In some embodiments, the method further includes: Step S1304: Based on the target temperature coefficient, temperature sensitivity coefficient, ambient temperature, and target microbial metabolic rate of the current iteration, update the target temperature coefficient for the next iteration. The target temperature coefficient can be updated using the following formula (7): Formula (7); in, This represents the target temperature coefficient for the current iteration round. Indicates the temperature sensitivity coefficient. This indicates the ambient temperature. This indicates the metabolic rate of the target microorganism. This indicates the target temperature coefficient for the next round; in one embodiment, Possible values .

[0042] Step S1305: Based on the target pH coefficient, pH sensitivity coefficient, soil pH, and target microbial metabolic rate of the current iteration, update the target pH coefficient for the next iteration.

[0043] The target pH coefficient can be updated using the following formula (8): Formula (8); in, This represents the target pH coefficient for the current iteration round. Indicates pH sensitivity coefficient, This indicates the soil pH. This indicates the metabolic rate of the target microorganism. Indicates the target pH coefficient for the next round; in one embodiment, Possible values .

[0044] The updated version will follow. Feedback is sent to step S1301 to form a parameter calibration closed loop.

[0045] In the above embodiments, by iteratively updating the target temperature coefficient and the target pH coefficient, the microbial metabolic rate calculation model can adapt to the real-time changes in environmental parameters, avoid the calculation deviation of fixed coefficients under different environmental conditions, and improve the robustness and adaptability of the model. The iterative optimization mechanism allows the model to continuously learn the correlation between environmental factors and microbial metabolism in long-term applications, and the decision accuracy gradually improves with the application cycle. Compared with fixed parameter models, it is more suitable for fertilization decision-making scenarios across seasons and regions.

[0046] In some embodiments, the crop growth data includes the normalized vegetation index and the current soil nutrient content, and the reward function of the reinforcement learning model is represented by the following formula (9): Formula (9); in, It is the change in the normalized vegetation index. yes The corresponding weights It is the cost of candidate fertilization decisions. yes The corresponding weights This is the increase in the crop's nutrient requirements. This refers to the current nutrient content of the soil. It is a maximum value function. yes The corresponding weights It is the reward value for candidate fertilization decisions; The reinforcement learning model is used to determine the candidate fertilization decision with the highest reward value as the target fertilization decision.

[0047] The state space of the reinforcement learning model can be defined as [R, ΔN, soil microbial fusion features, T, ... ], ΔN can be directly used to calculate the nutrient deficiency term of the reward function ( In one embodiment, , , The DQN model optimization design experience replay pool capacity is 10. 6 The batch size is 64; the target network parameters are synchronized every 100 steps; the final output is the target fertilization decision, which may include a real-time fertilization prescription map (spatial distribution of fertilization amount) and dynamic fertilization amount suggestions (adjusted according to the crop's fertilization requirements).

[0048] In the above embodiments, the reward function takes into account crop growth benefits (vegetation index changes), economic costs (fertilization costs), and ecological benefits (nutrient supply and demand matching), avoiding the problem of excessive fertilization in pursuit of yield, and achieving synergistic optimization of high yield, cost saving, and environmental protection; using the maximization of reward value as the decision criterion, the reinforcement learning model can autonomously weigh multi-dimensional objectives and output the optimal fertilization plan, reducing the cost of human intervention and improving the autonomy and intelligence level of the decision-making system.

[0049] In some embodiments, the target fertilization decision includes a real-time fertilization prescription map, which contains the fertilization amount for each sub-region of the target planting area. The system further includes an execution module, and the method further includes: Step S140: Using the execution module, plan the operation path of the drone based on the fertilizer prescription map; Step S150: Using the drone, spray the corresponding amount of fertilizer onto each of the sub-areas sequentially according to the operation path.

[0050] The execution module can take real-time fertilization prescription maps and dynamic fertilization amount suggestions as inputs. During processing, it can control the intelligent fertilizer applicator to precisely adjust the fertilization amount according to the spatial distribution of the real-time fertilization prescription map via a 4G module or RS485 protocol. It can also precisely spray fertilizer according to the dynamic fertilization amount suggestions via a drone path planning system. At the same time, it records crop growth (such as NDVI changes) and soil parameter (such as nutrient content and pH) changes after fertilization through sensors and image acquisition devices to form feedback data. The output includes the precise fertilization execution results containing the actual fertilization amount and coverage area, as well as fertilization effect feedback data containing crop response data and soil change data. The precise fertilization execution results and the fertilization effect feedback data are fed back to the data processing module for microbial network reconstruction and to the fertilization decision module for DQN reward function calculation and parameter calibration, forming a closed loop of collection, processing, decision-making, execution, and feedback.

[0051] The closed-loop control mechanism relies on its feedback-driven adaptive parameter adjustment. For example, when feedback data shows that ΔNDVI < -0.15 in a certain area, indicating inhibited crop growth, the system automatically triggers the reconstruction of the microbial co-occurrence network in the data processing module: in the Bi-GCN-STA model, the soil parameter weights for that area increase from 0.45 to 0.72 (increasing the pH weight to compensate for the decrease in microbial activity), while the b-value of the temperature-pH formula is dynamically adjusted from 0.015 to 0.017. In a 100-mu experimental field, the closed-loop mechanism improved the accuracy of fertilization decisions by 28.7%, shortened the system's autonomous optimization cycle from 7 days to 2 days, and significantly reduced the need for manual intervention.

[0052] In the above embodiments, fertilization decisions are transformed into visualized zonal prescription maps, and a drone precision spraying scheme is provided, which solves the problem of the difficulty in implementing traditional fertilization decisions and achieves a seamless connection from theoretical models to field applications. Zonal differentiated fertilization based on prescription maps, compared with traditional uniform fertilization methods, can supply fertilizer on demand according to the soil-microorganism-crop characteristics of different sub-regions, further improving fertilizer utilization efficiency and reducing the risk of non-point source pollution from chemical fertilizers. Drone operation path planning enables automated and large-scale spraying, which is more efficient and has better fertilization accuracy than manual fertilization or mechanical spreading, and is suitable for intensive management of large-area planting areas.

[0053] In this embodiment, metabolic potential and stability are quantified through microbial co-occurrence network topology parameters, overcoming the limitations of traditional diversity indicators. A temperature-pH dual-factor dynamic fertilization model is constructed to achieve real-time matching of fertilization amount and microbial activity. A dual-flow graph convolution-attention model (Bi-GCN-STA) is used to fuse multimodal data, and a self-governing closed-loop control mechanism is constructed by combining reinforcement learning. This design effectively solves the core problems of missing microbial functional information, insufficient dynamic modeling accuracy, and lack of system autonomy, filling the technical gap in microbial-soil collaborative modeling and dynamic decision-making.

[0054] The embodiments of this application can significantly improve the efficiency of precision fertilization and soil nutrient utilization: by identifying the functional zoning of microbial co-occurrence networks and quantifying the contribution of key species, combined with the fusion of multidimensional features of soil and microorganisms, a high degree of matching between fertilization schemes and actual crop needs can be achieved, reducing excessive fertilization and resource waste, while optimizing soil nutrient conversion efficiency and enhancing environmental adaptability. The embodiments of this application can enhance the system's autonomous decision-making ability: based on a temperature-pH metabolic model with dynamic parameter calibration and a reinforcement learning framework, the system can respond to environmental changes such as precipitation and temperature in real time, autonomously adjust fertilization strategies, maintain stable decision-making ability under extreme climatic conditions, and reduce the need for manual intervention.

[0055] The embodiments of this application can achieve a synergistic improvement in crop yield and ecological benefits: accurately matching nutrient supply with the needs of crop growth stages, simultaneously improving yield and quality (such as protein content and disease resistance), while reducing excessive application of chemical fertilizers, alleviating agricultural non-point source pollution, and promoting eco-friendly agricultural practices.

[0056] This application, by constructing a microbial co-occurrence network and calculating the modularity index (quantifying the strength of functional zoning) and node centrality (quantifying the role of key species), transforms the metabolic potential and stability of microbial communities into quantifiable parameters for the first time. This method overcomes the limitation of traditional diversity indicators that only focus on the number of species, effectively compensating for the shortcomings of related technologies in analyzing microbial communities at the functional level, and providing deeper bioinformatics support for precision fertilization decisions.

[0057] This application constructs a dual data processing stream. The microbial stream extracts the topological features (such as node degree distribution and clustering coefficient) of the microbial co-occurrence network through a graph convolutional network (GCN) to form a microbial network feature vector. The soil stream extracts the spatial distribution patterns and temporal dynamic features of soil physicochemical parameters through a convolutional neural network (CNN) to form a soil modal feature vector. A dynamic weight allocation strategy is designed to automatically adjust the weight ratio of microorganisms and soil features based on environmental changes (such as precipitation and temperature fluctuations), strengthen key influencing factors (such as dynamically increasing the pH weight in high pH gradient areas), and weaken secondary interference factors, thereby significantly improving the pertinence and adaptability of multimodal data fusion and solving the bottleneck of traditional fixed weight methods that cannot cope with dynamic environmental changes.

[0058] Based on the Metabolic Ecosystem Theory (MTE), this application proposes a temperature-pH dual-factor metabolic formula to calculate microbial metabolic rates in real time. And combine the particle swarm optimization (PSO) algorithm to dynamically adjust the parameters. (Target temperature coefficient) and (Target pH coefficient). This formula, for the first time, incorporates both temperature and pH as environmental factors into the metabolic model, achieving precise matching between fertilizer application rate and microbial activity, filling a gap in existing technologies regarding dynamic environmental response mechanisms. Through closed-loop feedback of the reinforcement learning model (DQN), the parameter calibration process further adapts to actual field conditions, significantly improving the environmental adaptability and autonomous optimization capability of fertilization decisions.

[0059] Based on the foregoing embodiments, this application provides a fertilization decision system based on soil microbial diversity analysis. The system includes various modules and sub-modules, and each unit of each sub-module can be implemented by a processor in an electronic device; of course, it can also be implemented by specific logic circuits. In the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.

[0060] Figure 2 A schematic diagram of the composition structure of a fertilization decision-making system based on soil microbial diversity analysis is provided for an embodiment of this application, as shown below. Figure 2 As shown, the system 200 includes: The system comprises a data acquisition module 21, a data processing module 22, and a fertilization decision module 23, wherein: The data acquisition module 21 is used to acquire soil samples, environmental parameters and crop growth data of the target planting area, and to analyze the soil samples to obtain microbial community composition data and soil physicochemical parameters. The data processing module 22 is used to determine the functional structure characteristics of the microbial community based on the microbial community composition data; generate soil parameter distribution characteristics based on the soil physicochemical parameters; and perform weighted fusion of the functional structure characteristics and the soil parameter distribution characteristics based on the dual-flow graph convolution-attention model to obtain soil microbial fusion characteristics. The fertilization decision module 23 is used to generate microbial metabolic rate and crop nutrient demand increment based on the functional structure characteristics of the microbial community, the environmental parameters, the soil parameter distribution characteristics and the crop growth data; and input the microbial metabolic rate, crop nutrient demand increment, soil microbial fusion characteristics, the environmental parameters and the soil parameter distribution characteristics into a reinforcement learning model to obtain the target fertilization decision output by the reinforcement learning model.

[0061] In some possible embodiments, the microbial community composition data includes an operational taxonomic unit (OTU) abundance table, an amplicon sequence variant (ASV) abundance table, and an average color change rate (AWCD), wherein the AWCD is used to characterize the carbon source utilization activity of the microbial community; the functional structural features include a modularity index and node centrality. The data processing module 22 is used to determine co-occurrence associations between species based on the OTU abundance table and the ASV abundance table using the Jaccard similarity coefficient; when the Jaccard similarity coefficient between species is greater than a preset association threshold, it is determined that there is a co-occurrence relationship between the corresponding species; using species as nodes, edges are added between species with co-occurrence relationships to construct a co-occurrence network of the microbial community; an initial feature vector is configured for each species node in the co-occurrence network, the initial feature vector including the average abundance of the corresponding species in each soil sample, the carbon source utilization capacity score of the corresponding species determined based on the AWCD, and the functional label annotation; the Louvain algorithm is used to iteratively optimize the community division of the microbial community to obtain the functional partitions of the microbial community; the modularity index of the functional partitions and the node centrality of the species nodes in the co-occurrence network are determined, the modularity index is used to quantify the functional partition strength of the microbial community, and the node centrality is used to locate the contribution degree of key species in the microbial community.

[0062] In some possible embodiments, the data processing module 22 is used to normalize the soil physicochemical parameters to obtain normalized physicochemical parameters; to interpolate the normalized physicochemical parameters using the Kriging spatial interpolation algorithm to generate a spatial distribution map of soil parameters in the target planting area; and to analyze the spatial distribution map of soil parameters to obtain the distribution characteristics of soil parameters.

[0063] In some possible embodiments, the environmental parameters include ambient temperature, the soil parameter distribution characteristics include soil pH, the crop growth data includes crop functional gene abundance and soil nutrient supply, and the fertilization decision module 23 is used to determine the microbial metabolic rate based on the functional structure characteristics of the microbial community, the ambient temperature, the soil pH, the target temperature coefficient, the target pH coefficient, and a preset basal metabolic rate constant; calculate a first change between the microbial metabolic rate and the basal metabolic rate constant, and calculate the difference between the first change and a second change in the soil nutrient supply; and determine the incremental crop nutrient demand based on the target microbial metabolic rate at which the difference is minimized, the crop functional gene abundance, and a preset nutrient conversion coefficient.

[0064] In some possible embodiments, the fertilization decision module 23 is used to update the target temperature coefficient for the next iteration based on the target temperature coefficient, temperature sensitivity coefficient, ambient temperature, and target microbial metabolic rate of the current iteration; and to update the target pH coefficient for the next iteration based on the target pH coefficient, pH sensitivity coefficient, soil pH, and target microbial metabolic rate of the current iteration.

[0065] In some possible embodiments, the crop growth data includes the normalized vegetation index and the current soil nutrient content, and the reward function of the reinforcement learning model is expressed by the following formula: ; in, It is the change in the normalized vegetation index. yes The corresponding weights It is the cost of candidate fertilization decisions. yes The corresponding weights This is the increase in the crop's nutrient requirements. This refers to the current nutrient content of the soil. It is a maximum value function. yes The corresponding weights It is the reward value for candidate fertilization decisions; The reinforcement learning model is used to determine the candidate fertilization decision with the highest reward value as the target fertilization decision.

[0066] In some possible embodiments, the target fertilization decision includes a real-time fertilization prescription map, which contains the fertilization amount for each sub-region of the target planting area. The system further includes an execution module for planning the operation path of a drone based on the fertilization prescription map; and using the drone to spray the corresponding amount of fertilizer onto each sub-region sequentially according to the operation path.

[0067] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0068] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0069] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0070] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected to achieve the purpose of the embodiments of this application according to actual needs. In addition, each functional unit in the embodiments of this application may be fully integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the integrated unit may be implemented in hardware or in the form of hardware plus software functional units.

[0071] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause the device automatic test line to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0072] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined to obtain new method embodiments without conflict. The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined to obtain new method embodiments or device embodiments without conflict.

[0073] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A fertilization decision-making method based on soil microbial diversity analysis, applied to a fertilization decision-making system based on soil microbial diversity analysis, the system comprising a data acquisition module, a data processing module, and a fertilization decision-making module, the method comprising: The data acquisition module is used to acquire soil samples, environmental parameters and crop growth data of the target planting area. The soil samples are then analyzed to obtain microbial community composition data and soil physicochemical parameters. The functional structure characteristics of the microbial community are determined using the data processing module based on the microbial community composition data. Based on the soil physicochemical parameters, soil parameter distribution characteristics are generated; based on the dual-flow graph convolution-attention model, the functional structure characteristics and the soil parameter distribution characteristics are weighted and fused to obtain soil microbial fusion characteristics. Using the fertilization decision module, based on the functional structure characteristics of the microbial community, the environmental parameters, the distribution characteristics of the soil parameters, and the crop growth data, microbial metabolic rate and crop nutrient demand increment are generated; the microbial metabolic rate, crop nutrient demand increment, soil microbial fusion characteristics, environmental parameters, and soil parameter distribution characteristics are input into a reinforcement learning model to obtain the target fertilization decision output by the reinforcement learning model.

2. The method according to claim 1, characterized in that, The microbial community composition data includes an operational taxonomic unit (OTU) abundance table, an amplicon sequence variant (ASV) abundance table, and an average color change rate (AWCD), wherein the AWCD is used to characterize the carbon source utilization activity of the microbial community; the functional structure features include a modularity index and node centrality. The step of using the data processing module to determine the functional structure characteristics of the microbial community based on the microbial community composition data includes: Based on the OTU abundance table and the ASV abundance table, co-occurrence associations between species are determined by the Jaccard similarity coefficient; when the Jaccard similarity coefficient between species is greater than a preset association threshold, it is determined that there is a co-occurrence relationship between the corresponding species; using species as nodes, edges are added between species with co-occurrence relationships to construct a co-occurrence network of the microbial community. An initial feature vector is configured for each species node in the co-occurrence network. The initial feature vector includes the average abundance of the corresponding species in each soil sample, the carbon source utilization capacity score of the corresponding species determined based on the AWCD, and the functional label annotation. The Louvain algorithm is used to iteratively optimize the community partitioning of the microbial community to obtain functional partitions. The modularity index of the functional partitions and the node centrality of species nodes in the co-occurrence network are determined. The modularity index is used to quantify the strength of the functional partitions of the microbial community, and the node centrality is used to locate the contribution of key species in the microbial community.

3. The method according to claim 1, characterized in that, The process of generating soil parameter distribution characteristics based on the soil physicochemical parameters includes: The soil physicochemical parameters were normalized to obtain normalized physicochemical parameters; The normalized physicochemical parameters are interpolated using the Kriging spatial interpolation algorithm to generate a spatial distribution map of soil parameters in the target planting area. The spatial distribution map of soil parameters is then analyzed to obtain the distribution characteristics of soil parameters.

4. The method according to claim 1, characterized in that, The environmental parameters include ambient temperature; the soil parameter distribution characteristics include soil pH; the crop growth data includes crop functional gene abundance and soil nutrient supply; the functional structure characteristics include modularity index and node centrality; the fertilization decision module, based on the functional structure characteristics of the microbial community, the environmental parameters, the soil parameter distribution characteristics, and the crop growth data, generates microbial metabolic rate and incremental crop nutrient demand, including: The microbial metabolic rate is determined based on the functional structure characteristics of the microbial community, the ambient temperature, the soil pH, the target temperature coefficient, the target pH coefficient, and the preset basal metabolic rate constant. Calculate the first change between the microbial metabolic rate and the basal metabolic rate constant, and calculate the difference between the first change and the second change in soil nutrient supply; The incremental crop nutrient requirements are determined based on the target microbial metabolic rate at which the difference is minimized, the abundance of crop functional genes, and the preset nutrient conversion coefficient.

5. The method according to claim 4, characterized in that, The method further includes: Based on the target temperature coefficient, temperature sensitivity coefficient, ambient temperature, and target microbial metabolic rate of the current iteration, the target temperature coefficient for the next iteration is updated. Based on the target pH coefficient, pH sensitivity coefficient, soil pH, and target microbial metabolic rate of the current iteration, the target pH coefficient for the next iteration is updated.

6. The method according to claim 1, characterized in that, The crop growth data includes the normalized vegetation index and the current soil nutrient content. The reward function of the reinforcement learning model is expressed by the following formula: ； in, It is the change in the normalized vegetation index. yes The corresponding weights It is the cost of candidate fertilization decisions. yes The corresponding weights This is the increase in the crop's nutrient requirements. This refers to the current nutrient content of the soil. It is a maximum value function. yes The corresponding weights It is the reward value for candidate fertilization decisions; The reinforcement learning model is used to determine the candidate fertilization decision with the highest reward value as the target fertilization decision.

7. The method according to claim 1, characterized in that, The target fertilization decision includes a real-time fertilization prescription map, which contains the fertilization amount for each sub-region of the target planting area. The system also includes an execution module, and the method further includes: Using the execution module, the operation path of the drone is planned based on the fertilizer prescription map; the drone is then used to spray the corresponding amount of fertilizer onto each sub-area in sequence according to the operation path.

8. A fertilization decision-making system based on soil microbial diversity analysis, the system comprising: The system comprises a data acquisition module, a data processing module, and a fertilization decision module, among which: The data acquisition module is used to acquire soil samples, environmental parameters and crop growth data of the target planting area, and to analyze the soil samples to obtain microbial community composition data and soil physicochemical parameters. The data processing module is used to determine the functional structure characteristics of the microbial community based on the microbial community composition data; generate soil parameter distribution characteristics based on the soil physicochemical parameters; and perform weighted fusion of the functional structure characteristics and the soil parameter distribution characteristics based on the dual-flow graph convolution-attention model to obtain soil microbial fusion characteristics. The fertilization decision module is used to generate microbial metabolic rate and crop nutrient demand increment based on the functional structure characteristics of the microbial community, the environmental parameters, the soil parameter distribution characteristics, and the crop growth data; and input the microbial metabolic rate, crop nutrient demand increment, soil microbial fusion characteristics, the environmental parameters, and the soil parameter distribution characteristics into a reinforcement learning model to obtain the target fertilization decision output by the reinforcement learning model.

9. The system according to claim 8, characterized in that, The environmental parameters include ambient temperature; the soil parameter distribution characteristics include soil pH; the crop growth data include crop functional gene abundance and soil nutrient supply; and the functional structure characteristics include modularity index and node centrality. The fertilization decision module is used for: The microbial metabolic rate is determined based on the functional structure characteristics of the microbial community, the ambient temperature, the soil pH, the target temperature coefficient, the target pH coefficient, and the preset basal metabolic rate constant. Calculate the first change between the microbial metabolic rate and the basal metabolic rate constant, and calculate the difference between the first change and the second change in soil nutrient supply; The incremental crop nutrient requirements are determined based on the target microbial metabolic rate at which the difference is minimized, the abundance of crop functional genes, and the preset nutrient conversion coefficient.

10. The system according to claim 8, characterized in that, The target fertilization decision includes a real-time fertilization prescription map, which contains the fertilization amount for each sub-region of the target planting area. The system also includes: The execution module is used to plan the operation path of the drone based on the fertilizer prescription map; and to use the drone to spray the corresponding amount of fertilizer onto each of the sub-regions in sequence according to the operation path.

Images