Intelligent fertilization and irrigation regulation and control device and method based on deep learning

Abstract

The invention discloses an intelligent fertilization and irrigation regulation and control device and method based on deep learning. The method comprises the following steps: S1, constructing an agricultural environment original data set; s2, preprocessing the original data set to generate an agricultural data set with consistent time and space; s3, constructing a multi-layer neuromorphic calculation model which comprises a photosynthetic layer, a root system layer and a transpiration layer; s4, performing correlation analysis based on the photosynthesis dynamic curve, the root system water and fertilizer absorption curve and the transpiration-moisture supply curve, and constructing a photosynthesis-root system-transpiration synergistic curve; s5, on the basis of the photosynthesis-root system-transpiration synergistic curve, calculating a promoting factor of photosynthesis on root system absorption and a regulating factor of root system absorption on transpiration intensity, and constructing a self-adaptive regulation and control function; and S6, carrying out time sequence analysis on the self-adaptive regulation and control function, and formulating a final fertilization and irrigation regulation and control scheme. According to the method, a multi-layer neuromorphic calculation method is adopted, the photosynthesis-root system-transpiration cooperative relation is accurately modeled, and intelligent fertilization and irrigation regulation and control are achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of agricultural intelligence, and particularly relates to an intelligent fertilization and irrigation regulation device and method based on deep learning. Background Art

[0002] With the growth of the global population and the intensification of climate change, agricultural production is facing multiple challenges of increasing food production, improving water and fertilizer use efficiency, and reducing environmental pollution. Fertilization and irrigation, as key links in crop growth, their reasonable regulation is of great significance for the sustainable development of agriculture. However, the traditional fertilization and irrigation management mode mainly relies on farmers' experience or simple timing and quantitative strategies, and usually fails to fully consider the actual needs of crop growth and the dynamic changes of the environment. This method is prone to low utilization rate of water and fertilizer resources. Excessive fertilization may cause soil degradation and water eutrophication, while insufficient fertilization may affect crop yield and quality. In addition, due to the different water and fertilizer requirements of different crop varieties at different growth stages and the complex changes in environmental conditions (such as temperature, humidity, light, soil nutrients), it is difficult for manual regulation to meet the needs of precision agriculture. Therefore, there is an urgent need for a regulation method that can dynamically sense the water and fertilizer requirements of crops and achieve precise fertilization and irrigation based on intelligent algorithms.

[0003] In recent years, with the development of technologies such as the Internet of Things, big data, and artificial intelligence, intelligent agriculture has been rapidly promoted. Agricultural monitoring systems based on sensor technology can collect crop growth environment data in real time, such as soil moisture, nitrogen, phosphorus, and potassium content, and meteorological conditions, providing basic information support for precise fertilization and irrigation. Some studies have proposed using mechanism models, statistical analysis, or machine learning methods to optimize water and fertilizer management strategies. For example, the irrigation regulation method based on soil moisture can automatically turn on irrigation when the humidity is lower than the threshold by setting the soil humidity threshold, avoiding over-irrigation. However, such methods usually ignore the actual growth state of crops, such as key physiological parameters such as root absorption capacity and leaf transpiration rate, and it is difficult to achieve efficient water and fertilizer linkage regulation. On the other hand, although the fertilization decision-making method based on crop growth models can combine soil water and fertilizer data with crop physiological requirements, due to the complex crop growth process and numerous variables, the traditional mechanism models have limited accuracy in describing the interaction between crops, soil, and environment, and it is difficult to meet the real-time regulation needs of large-scale agricultural production.

[0004] The rise of deep learning technology has provided new ideas for precise fertilization and irrigation optimization. By constructing data-driven intelligent regulation models and using historical data of crop growth for learning and prediction, it is possible to optimize the supply of water and fertilizer under complex environmental conditions. However, most existing deep learning methods focus on single-task modeling, such as crop yield prediction, soil moisture estimation, or pest and disease detection, lacking in-depth modeling of key physiological processes such as photosynthesis, root water and fertilizer absorption, and transpiration. At the same time, deep learning models usually rely on large-scale labeled data, which is difficult to obtain in the agricultural field, and data noise and missing phenomena are relatively common, making the application of traditional deep learning methods in agricultural intelligent regulation face certain challenges. In addition, most existing intelligent fertilization and irrigation methods adopt static optimization strategies and are difficult to adapt to the dynamic changes of the environment and crop physiological states, resulting in a large optimization space for water and fertilizer use efficiency.

[0005] Therefore, how to provide an intelligent fertilization and irrigation regulation device and method based on deep learning is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0006] An object of the present invention is to provide an intelligent fertilization and irrigation regulation device and method based on deep learning. The present invention adopts a multi-layer neuromorphic computing method, combines a spiking neural network, a topological graph attention network, and a neural tensor network to accurately model crop photosynthesis, root water and fertilizer absorption, and transpiration, constructs a photosynthesis-root-transpiration synergistic curve, and dynamically optimizes fertilization and irrigation strategies. Through non-linear dynamic mapping and an adaptive regulation function, intelligent adjustment of water and fertilizer supply is realized, water and fertilizer use efficiency is improved, resource waste and environmental pollution are reduced, the stability and sustainability of agricultural production are enhanced, and it has high efficiency, accuracy, and environmental adaptability, and is applicable to various crops and different environmental conditions.

[0007] An intelligent fertilization and irrigation regulation method based on an embodiment of the present invention includes the following steps:

[0008] S1. Arrange a sensor network and an unmanned aerial vehicle remote sensing device in the target farmland, collect soil-crop-meteorological data, and construct an original agricultural environment dataset;

[0009] S2. Preprocess the original agricultural environment dataset, denoise the data, and perform standardization processing on the time scale and space scale to generate a spatio-temporally consistent agricultural dataset;

[0010] S3. Based on the spatio-temporally consistent agricultural dataset, construct a multi-layer neuromorphic computing model, and the multi-layer neuromorphic computing model includes a photosynthesis layer, a root layer, and a transpiration layer:

[0011] S31. Based on the pulse neural network combined with synaptic weight adaptive adjustment, the photosynthetic layer predicts the dynamic response of crop leaves to environmental factors and generates a photosynthesis dynamic curve;

[0012] S32. The root layer uses a topological graph attention network and variational inference to simulate the water and fertilizer absorption capacity of roots at different depths, predict the dynamic water and fertilizer in the rhizosphere soil, and generate a root water and fertilizer absorption curve;

[0013] S33. The transpiration layer uses a neural tensor network combined with a dynamic environment perception mechanism to generate a transpiration-water supply curve based on the crop transpiration rate and soil moisture status;

[0014] S4. Conduct correlation analysis based on the photosynthesis dynamic curve, root water and fertilizer absorption curve, and transpiration-water supply curve, calculate the coupling relationship between photosynthetic rate, root absorption capacity, and transpiration intensity, and construct a photosynthesis-root-transpiration synergy curve;

[0015] S5. Based on the photosynthesis-root-transpiration synergy curve, use a non-linear dynamic mapping method to calculate the promotion factor of photosynthesis on root absorption and the regulation factor of root absorption on transpiration intensity, and construct an adaptive regulation function;

[0016] S6. Conduct time series analysis on the adaptive regulation function to formulate a final fertilization and irrigation regulation plan.

[0017] Optionally, the S31 specifically includes:

[0018] S311. Based on the spatio-temporally consistent agricultural dataset, obtain the light intensity, air temperature, air humidity, carbon dioxide concentration, and leaf temperature of crop leaves in the target area, and construct a photosynthetic environment state vector;

[0019] S312. Construct a photosynthetic calculation model based on a pulse neural network, use the photosynthetic environment state vector as the input, the input layer neurons receive the photosynthetic environment state vector, the hidden layer is composed of pulse neurons, and the dynamic equation of the membrane potential of each pulse neuron changing with time is:

[0020]

[0021] Among them, U represents the membrane potential, τ m represents the time constant, I syn represents the synaptic input current, and a gated pulse triggering mechanism is used to control the neuron firing frequency;

[0022] S313. Calculate the leaf photosynthesis intensity based on the pulse neural network and calculate the instantaneous photosynthetic rate of the crop:

[0023]

[0024] Among them, P t represents the instantaneous photosynthetic rate of crops, A max represents the maximum photosynthetic rate, α represents the light response coefficient, I represents the light intensity, β represents the carbon dioxide response coefficient, represents the carbon dioxide concentration, R d represents the dark respiration rate;

[0025] S314. Optimize the synaptic weights based on the historical data of photosynthesis and dynamically adjust the synaptic weights of the spiking neurons:

[0026] w ij (t + 1) = w ij (t) + η·S(i, j)·[λE t + γP t ;

[0027] Among them, w ij (t + 1) represents the synaptic weight between the i-th neuron and the j-th neuron at time step t + 1, w ij (t) represents the synaptic weight between the i-th neuron and the j-th neuron at time step t, η represents the learning rate, S(i, j) represents the synaptic connection strength between the i-th neuron and the j-th neuron, λ and γ represent the adjustment coefficients, E t represents the photosynthetic environment state vector;

[0028] S315. Perform curve fitting on the instantaneous photosynthetic rates of crops at multiple time steps, and use the spiking neural network with optimized synaptic weights to generate a dynamic photosynthesis curve.

[0029] Optionally, the S32 specifically includes:

[0030] S321. Based on the spatio-temporally consistent agricultural dataset, obtain the soil moisture content, soil nutrient concentration, and electrical conductivity at different depths of the root system, and construct a root system topology graph G = (V, E) in combination with the crop root system morphology information, where the node V represents the positions at different depths of the root system, and the edge E represents the water and fertilizer transfer relationship;

[0031] S322. Use a topological graph attention network to perform weighted calculation on the root system topology graph, and obtain the attention coefficient of the node to the neighbor nodes based on the attention mechanism;

[0032] S323. Estimate the water and fertilizer absorption rate of the root system at the node through variational inference:

[0033] A p (t) = k w W d + k n N d + k p Pd +k k K d -ζ p t;

[0034] Among them, A p (t) represents the water and fertilizer absorption rate of the p-th node at time step t, k w represents the water absorption coefficient, k n represents the nitrogen absorption coefficient, k p represents the phosphorus absorption coefficient, k k represents the potassium absorption coefficient, W d represents the soil water content, N d represents the soil nitrogen concentration, P d represents the soil phosphorus concentration, K d represents the soil potassium concentration, ζ p represents the root decay factor of the p-th node, and t represents the time step;

[0035] S324. Model the posterior distribution of the water and fertilizer absorption rate, and define the posterior distribution p(A p (t)|E r,p,t ) of the water and fertilizer absorption at the node of the root system, where E r,p,t represents the rhizosphere environmental state vector of the p-th node at time step t, and use the variational inference method for iterative update to minimize the deviation between the measured value and the estimated value to obtain the optimal absorption coefficient;

[0036] S325. Combine the attention coefficient and the optimal absorption coefficient, summarize the water and fertilizer absorption rates of each node at different time steps to form a dynamic sequence of root system water and fertilizer, use the edge weight in the root system topology map to correct the cross-node water and fertilizer transfer error, and re-estimate the nodes with noise or anomalies to obtain a unified water and fertilizer absorption sequence;

[0037] S326. Conduct time series analysis on the unified water and fertilizer absorption sequence, and generate a root system water and fertilizer absorption curve according to the change trends of the water and fertilizer absorption rates of multiple nodes at different depths.

[0038] Optionally, the S33 specifically includes:

[0039] S331. Based on the spatio-temporally consistent agricultural dataset, obtain the transpiration rate, soil water content, environmental temperature, air humidity and meteorological fitness of the target farmland crops to form a transpiration environment vector E tr,t ;

[0040] S332. Perform feature encoding on the transpiration environment vector, and input the encoding result into a neural tensor network to establish a multi-factor interaction relationship, where the core tensor mapping of the neural tensor network satisfies:

[0041]

[0042] Where f represents the neural tensor network, x1 and x2 represent different parts of the transpiration environment vector, and U[ k represents the kth component of the tensor core, σ represents the activation function, V represents the weight matrix, and b represents the offset term;

[0043] S333. Combined with the dynamic environment perception mechanism, the neural tensor network mapping weights are adjusted according to the changes in wind speed, light and soil moisture, so that the sensitivity of transpiration to external conditions is updated over time, and the prediction error caused by environmental fluctuations is reduced through multiple iterations;

[0044] S334. Define the transpiration-water supply coupling function to comprehensively describe the nonlinear relationship between transpiration rate and water supply:

[0045] T couple (t) = α1exp(θ·E l (t))-β1ln(1+γ1W d );

[0046] Among them, T couple (t) represents the nonlinear coupling degree between crop transpiration rate and soil water supply at time step t, α1 represents the transpiration amplification coefficient, β1 represents the water suppression coefficient, γ1 represents the soil water gain factor, exp represents the exponential function, E l (t) represents the transpiration rate of crops, W d Indicates soil moisture content;

[0047] S335: Based on the transpiration-water supply coupling function, a stochastic gradient descent algorithm is used to minimize the transpiration-water supply residual, and the T obtained at each time step t is couple (t) Integrated into the transpiration-water supply curve.

[0048] Optionally, the S4 specifically includes:

[0049] S41, obtaining the photosynthesis dynamic curve, the root water and fertilizer absorption curve and the transpiration-water supply curve, and aligning the three curves at the same time step to construct a time series data set;

[0050] S42, interpolating and smoothing the time series data set, removing outliers, and using a dynamic time warping method to correct local peak misalignment to obtain a complete and comparable three-curve sequence;

[0051] S43. Calculate the coupling degree ρ between photosynthetic rate, root absorption capacity and transpiration intensity i,j :

[0052]

[0053] Among them, T1 represents the total number of time steps, and X i (t) and X j (t) respectively represent the values of any two of the three curves at time step t. represents the average value of the curve where X i (t) is located. represents the average value of the curve where X j (t) is located.

[0054] S44. Construct a non - linear mapping function according to the coupling degree, perform multi - dimensional mapping on the three curves and output a coupling index sequence, and extract the key coupling characteristics of photosynthesis, root absorption and transpiration intensity.

[0055] S45. Define a collaborative optimization function:

[0056] F(ξ1, ξ2, ξ3) = ξ1ρ P,A + ξ2ρ A,T + ξ3ρ P,T ;

[0057] Among them, F(ξ1, ξ2, ξ3) represents the collaborative optimization function, ξ1 represents the weight coefficient of the coupling degree between photosynthetic rate and root absorption ability, ξ2 represents the weight coefficient of the coupling degree between root absorption ability and transpiration intensity, ξ3 represents the weight coefficient of the coupling degree between photosynthetic rate and transpiration intensity, and ρ P,A represents the coupling degree between photosynthetic rate P and root absorption ability A, and ρ A,T represents the coupling degree between root absorption ability A and transpiration intensity T, and ρ P,T represents the coupling degree between photosynthetic rate P and transpiration intensity T. By iteratively solving ξ1, ξ2, ξ3, maximize the said collaborative optimization function to obtain the optimal coupling coefficient combination of photosynthesis - root - transpiration.

[0058] S46. Substitute the optimal coupling coefficient combination of photosynthesis - root - transpiration into the coupling index sequence to generate a photosynthesis - root - transpiration collaborative curve.

[0059] Optionally, the specific content of S5 includes:

[0060] S51. Obtain the coupling relationship among the three represented by the photosynthesis - root - transpiration collaborative curve, discretize the photosynthesis - root - transpiration collaborative curve on the unified time axis to form a time - series sample set for calculating the promotion factor and regulation factor.

[0061] S52. Based on the non - linear dynamic mapping method, define the promotion factor φ P→A (t) of photosynthesis on root absorption:

[0062]

[0063] Among them, P(τ) represents the photosynthetic rate at time τ, κ1 represents the photosynthesis influence coefficient, μ1 represents the root absorption threshold, and (x) + represents taking 0 when x < 0, ensuring that the promotion factor only accumulates when the photosynthetic rate is greater than the threshold;

[0064] S53. Based on the non-linear dynamic mapping method, define the regulation factor φ A→T (t) of root absorption on transpiration intensity:

[0065]

[0066] Among them, A(τ) represents the root water and fertilizer absorption amount at time τ, κ2 represents the root regulation coefficient, and μ2 represents the transpiration trigger threshold;

[0067] S54. Combine the promotion factor φ P→A (t) of photosynthesis on root absorption and the regulation factor φ A→T (t) of root absorption on transpiration intensity to construct an adaptive regulation function Ψ control (t):

[0068] Ψ control (t) = ζ1φ P→A (t) + ζ2φ A→T (t);

[0069] Among them, ζ1 represents the weight coefficient of the promotion factor, ζ2 represents the weight coefficient of the regulation factor, and ζ1 and ζ2 are updated by minimizing the fertilization-irrigation deviation.

[0070] An intelligent fertilization and irrigation regulation device based on deep learning according to an embodiment of the present invention includes the following modules:

[0071] A data acquisition module, which is used to deploy a sensor network and an unmanned aerial vehicle remote sensing device in the target farmland, collect soil-crop-meteorological data, and obtain an original agricultural environment dataset;

[0072] A data preprocessing module, which is used to perform denoising processing on the original agricultural environment dataset, and standardize the time scale and space scale to generate a spatio-temporally consistent agricultural dataset;

[0073] A multi-layer neuromorphic computing module, including:

[0074] A photosynthesis layer modeling unit, which is used to predict the dynamic response of crop leaves to environmental factors based on a spiking neural network combined with adaptive adjustment of synaptic weights, and generate a photosynthesis dynamic curve;

[0075] A root system layer modeling unit, which uses a topological graph attention network and variational inference to simulate the water and fertilizer absorption capacity of roots at different depths, predict the dynamics of water and fertilizer in the rhizosphere soil, and generate a root system water and fertilizer absorption curve;

[0076] A transpiration layer modeling unit, which uses a neural tensor network combined with a dynamic environment perception mechanism to generate a transpiration-water supply curve based on the crop transpiration rate and soil moisture status;

[0077] A curve correlation analysis module, which performs correlation analysis based on the photosynthesis dynamic curve, the root system water and fertilizer absorption curve, and the transpiration-water supply curve, calculates the coupling relationship between the photosynthetic rate, the root system absorption capacity, and the transpiration intensity, and constructs a photosynthesis-root system-transpiration synergy curve;

[0078] A non-linear dynamic mapping module, which calculates the promotion factor of photosynthesis on root system absorption and the regulation factor of root system absorption on transpiration intensity based on the photosynthesis-root system-transpiration synergy curve, and constructs an adaptive regulation function;

[0079] A time series regulation module, which performs time series analysis on the adaptive regulation function, and formulates a final fertilization and irrigation regulation plan according to the crop growth cycle, environmental changes, and soil water and fertilizer status.

[0080] The beneficial effects of the present invention are:

[0081] The present invention realizes the collaborative modeling of crop photosynthesis, root system water and fertilizer absorption, and transpiration by constructing an intelligent fertilization and irrigation regulation method based on deep learning, and optimizes the accuracy and adaptability of water and fertilizer supply. Compared with traditional regulation methods that rely on empirical judgment or fixed rules, the present invention is based on a multi-layer neuromorphic computing model, fully considers the physiological mechanism of crops, can dynamically perceive environmental changes, and adjusts fertilization and irrigation strategies in real time. By dynamically modeling photosynthesis through a spiking neural network, it simulates the response of crop leaves to factors such as light, temperature, humidity, and carbon dioxide, predicts the instantaneous photosynthetic rate of crops, and ensures that the water and fertilizer supply meets the growth needs of crops. The root system layer modeling uses a topological graph attention network combined with a variational inference method to accurately simulate the water and fertilizer absorption capacity of roots at different depths, predict the dynamics of water and fertilizer in the rhizosphere soil, so that the water and fertilizer supply strategy can match the absorption efficiency of crop roots, and avoid crop growth problems or resource waste caused by uneven water and fertilizer supply. At the same time, the present invention introduces a neural tensor network combined with a dynamic environment perception mechanism to perform non-linear modeling on the relationship between crop transpiration rate and soil water supply, accurately predict the water demand during crop transpiration, optimize the water supply plan, and improve the utilization rate of water and fertilizer.

[0082] In terms of water and fertilizer regulation decision-making, the present invention breaks through the limitation of traditional irrigation and fertilization methods that are only adjusted based on soil moisture or crop growth. It proposes a method based on the collaborative analysis of photosynthesis-root-transpiration, conducts a correlation analysis on the dynamic curve of photosynthesis, the water and fertilizer absorption curve of roots, and the transpiration-water supply curve, calculates the coupling relationship between photosynthetic rate, root absorption capacity, and transpiration intensity, and constructs a photosynthesis-root-transpiration collaborative curve. This collaborative analysis method can quantify the dynamic characteristics of crop water and fertilizer utilization, making up for the defect that existing deep learning methods fail to consider the physiological interaction of crops. Based on this collaborative curve, a non-linear dynamic mapping method is further used to calculate the promotion factor of photosynthesis on root absorption and the regulation factor of root absorption on transpiration intensity, and an adaptive regulation function is constructed, enabling the water and fertilizer supply to be automatically optimized according to environmental changes and crop physiological needs, improving the accuracy and stability of water and fertilizer supply. Finally, the regulation parameters are optimized through time series analysis, so that the intelligent fertilization and irrigation strategies can not only adapt to short-term environmental fluctuations but also optimize water and fertilizer management in the long term, enhancing the stability and sustainability of agricultural production.

[0083] The implementation of the present invention effectively improves the utilization efficiency of water and fertilizer resources, reduces water resource waste and soil pollution problems caused by excessive fertilization. Compared with traditional irrigation and fertilization systems, the present invention not only increases crop yields but also reduces the environmental impact of agricultural production, achieving the core goal of precision agriculture. In addition, the multi-layer neuromorphic computing model constructed by the present invention can adapt to the water and fertilizer regulation requirements of different crop varieties, different growth stages, and different meteorological conditions, has strong versatility and popularization value, and provides a new technical approach for the development of agricultural intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0085] Figure 1 is the overall flowchart of an intelligent fertilization and irrigation regulation method based on deep learning proposed by the present invention;

[0086] Figure 2 is the structural schematic diagram of an intelligent fertilization and irrigation regulation device based on deep learning proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0087] Now, the present invention will be further described in detail with reference to the accompanying drawings. These drawings are all simplified schematic diagrams, only showing the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0088] Reference Figure 1, An intelligent fertilization and irrigation regulation method based on deep learning, comprising the following steps:

[0089] S1. Deploy a sensor network and an unmanned aerial vehicle (UAV) remote sensing device in the target farmland, collect soil-crop-meteorological data, and construct an original agricultural environment dataset;

[0090] S2. Preprocess the original agricultural environment dataset, denoise the data, and perform standardization processing on the time scale and spatial scale to generate a spatiotemporally consistent agricultural dataset;

[0091] S3. Based on the spatiotemporally consistent agricultural dataset, construct a multi-layer neuromorphic computing model, and the multi-layer neuromorphic computing model includes a photosynthesis layer, a root system layer, and a transpiration layer:

[0092] S31. The photosynthesis layer is based on a spiking neural network combined with adaptive adjustment of synaptic weights to predict the dynamic response of crop leaves to environmental factors and generate a photosynthesis dynamic curve;

[0093] S32. The root system layer uses a topological graph attention network and variational inference to simulate the water and fertilizer absorption capacity of the root system at different depths, predict the dynamics of water and fertilizer in the rhizosphere soil, and generate a root system water and fertilizer absorption curve;

[0094] S33. The transpiration layer uses a neural tensor network combined with a dynamic environment perception mechanism to generate a transpiration-water supply curve based on the crop transpiration rate and soil moisture conditions;

[0095] S4. Conduct correlation analysis based on the photosynthesis dynamic curve, the root system water and fertilizer absorption curve, and the transpiration-water supply curve, calculate the coupling relationship between the photosynthetic rate, the root system absorption capacity, and the transpiration intensity, and construct a photosynthesis-root system-transpiration synergy curve;

[0096] S5. Based on the photosynthesis-root system-transpiration synergy curve, use a non-linear dynamic mapping method to calculate the promotion factor of photosynthesis on root system absorption and the regulation factor of root system absorption on transpiration intensity, and construct an adaptive regulation function;

[0097] S6. Conduct a time series analysis of the adaptive regulation function to formulate a final fertilization and irrigation regulation plan.

[0098] In this embodiment, the S31 specifically includes:

[0099] S311. Based on the spatiotemporally consistent agricultural dataset, obtain the light intensity, air temperature, air humidity, carbon dioxide concentration, and leaf temperature of crop leaves in the target area, and construct a photosynthetic environment state vector;

[0100] S312. Construct a photosynthetic computing model based on a spiking neural network, take the photosynthetic environment state vector as the input, the input layer neurons receive the photosynthetic environment state vector, the hidden layer consists of spiking neurons, and the kinetic equation for the membrane potential of each spiking neuron changing with time is:

[0101]

[0102] Among them, U represents the membrane potential, τ m represents the time constant, I syn represents the synaptic input current, and a gated pulse triggering mechanism is used to control the neuron firing frequency;

[0103] S313. Calculate the photosynthesis intensity of the leaf based on the spiking neural network, and calculate the instantaneous photosynthetic rate of the crop:

[0104]

[0105] Among them, P t represents the instantaneous photosynthetic rate of the crop, A max represents the maximum photosynthetic rate, α represents the light response coefficient, I represents the light intensity, β represents the carbon dioxide response coefficient, represents the carbon dioxide concentration, R d represents the dark respiration rate;

[0106] S314. Optimize the synaptic weights based on the photosynthesis historical data, and dynamically adjust the synaptic weights of the spiking neurons:

[0107] w ij (t + 1) = w ij (t) + η·S(i, j)·[λE t + γP t ;

[0108] Among them, w ij (t + 1) represents the synaptic weight between the i-th neuron and the j-th neuron at time step t + 1, w ij (t) represents the synaptic weight between the i-th neuron and the j-th neuron at time step t, η represents the learning rate, S(i, j) represents the synaptic connection strength between the i-th neuron and the j-th neuron, λ and γ represent the adjustment coefficients, and E t represents the photosynthetic environment state vector;

[0109] S315. Perform curve fitting on the instantaneous photosynthetic rates of the crop at multiple time steps, and use the spiking neural network with optimized synaptic weights to generate a photosynthesis dynamic curve.

[0110] In this embodiment, the S32 specifically includes:

[0111] S321. Based on the spatio-temporally consistent agricultural dataset, obtain the soil moisture content, soil nutrient concentration, and electrical conductivity at different root depths, and construct a root topology graph G=(V, E) by combining crop root morphology information, where the nodes V represent the positions at different root depths, and the edges E represent the water and fertilizer transfer relationships;

[0112] S322. Use the topological graph attention network to perform weighted calculation on the root topology graph, and obtain the attention coefficient of the node to its neighbor nodes based on the attention mechanism;

[0113] S323. Estimate the water and fertilizer absorption rate of the roots at the nodes through variational inference:

[0114] A p (t)=k w W d +k n N d +k p P d +k k K d -ζ p t;

[0115] where, A p (t) represents the water and fertilizer absorption rate of the p-th node at time step t, k w represents the absorption coefficient of water, k n represents the absorption coefficient of nitrogen, k p represents the absorption coefficient of phosphorus, k k represents the absorption coefficient of potassium, W d represents the soil moisture content, N d represents the soil nitrogen concentration, P d represents the soil phosphorus concentration, K d represents the soil potassium concentration, ζ p represents the root decay factor of the p-th node, and t represents the time step;

[0116] S324. Perform posterior distribution modeling on the water and fertilizer absorption rate, define the posterior distribution p(A p (t)|E r,p,t ), where E r,p,t represents the rhizosphere environmental state vector of the p-th node at time step t, and use the variational inference method to perform iterative update to minimize the deviation between the measured value and the estimated value to obtain the optimal absorption coefficient;

[0117] S325. Combine the attention coefficient and the optimal absorption coefficient, summarize the water and fertilizer absorption rates of each node at different time steps to form a dynamic sequence of root water and fertilizer, correct the cross-node water and fertilizer transfer error using the edge weights in the root topology map, and re-estimate the nodes with noise or anomalies to obtain a unified water and fertilizer absorption sequence;

[0118] S326. Perform time series analysis on the unified water and fertilizer absorption sequence, and generate a root water and fertilizer absorption curve based on the change trends of the water and fertilizer absorption rates of multiple nodes at different depths.

[0119] In this embodiment, the S33 specifically includes:

[0120] S331. Based on the spatio-temporally consistent agricultural dataset, obtain the transpiration rate, soil moisture content, ambient temperature, air humidity, and meteorological fitness of the target farmland crops to form a transpiration environment vector E tr,t ;

[0121] S332. Perform feature encoding on the transpiration environment vector, and input the encoding result into a neural tensor network to establish a multi-factor interaction relationship, where the core tensor mapping of the neural tensor network satisfies:

[0122]

[0123] where f represents the neural tensor network, x1 and x2 respectively represent different parts of the transpiration environment vector, U k represents the k-th component of the tensor core, σ represents the activation function, V represents the weight matrix, and b represents the offset term;

[0124] S333. Combine the dynamic environment perception mechanism, adjust the mapping weights of the neural tensor network according to the changes in wind speed, light, and soil humidity, so that the sensitivity of transpiration to external conditions is updated over time, and reduce the prediction error caused by environmental fluctuations through multiple iterations;

[0125] S334. Define a transpiration-water supply coupling function to comprehensively describe the non-linear relationship between the transpiration rate and water supply:

[0126] T couple (t) = α1exp(θ·E l (t)) - β1ln(1 + γ1W d );

[0127] where T couple (t) represents the degree of non-linear coupling between the crop transpiration rate and soil water supply at time step t, α1 represents the transpiration amplification coefficient, β1 represents the water inhibition coefficient, γ1 represents the soil water gain factor, exp represents the exponential function, E l (t) represents the transpiration rate of the crop, and Wd represents the soil moisture content;

[0128] S335. Based on the transpiration-water supply coupling function, the stochastic gradient descent algorithm is used to minimize the transpiration-water supply residual, and the T obtained at each time step t couple (t) is integrated into the transpiration-water supply curve.

[0129] In this embodiment, the S4 specifically includes:

[0130] S41. Obtain the photosynthesis dynamic curve, the root water and fertilizer absorption curve, and the transpiration-water supply curve, align the three curves at the same time step, and construct a time series data set;

[0131] S42. Interpolate and smooth the time series data set, remove outliers, and use the dynamic time warping method to correct the local peak misalignment to obtain a complete and comparable three-curve sequence;

[0132] S43. Calculate the coupling degree ρ between any two of the photosynthetic rate, the root absorption ability, and the transpiration intensity i,j :

[0133]

[0134] where T1 represents the total number of time steps, and X i (t) and X j (t) respectively represent the values of any two of the three curves at time step t, represents the average value of the curve where X i (t) is located, represents the average value of the curve where X j (t) is located;

[0135] S44. Construct a non-linear mapping function according to the coupling degree, perform multi-dimensional mapping on the three curves and output a coupling index sequence, and extract the key coupling characteristics of photosynthesis, root absorption, and transpiration intensity;

[0136] S45. Define the collaborative optimization function:

[0137] F(ξ1, ξ2, ξ3) = ξ1ρ P,A + ξ2ρ A,T + ξ3ρ P,T ;

[0138] where F(ξ1, ξ2, ξ3) represents the collaborative optimization function, ξ1 represents the weight coefficient of the coupling degree between the photosynthetic rate and the root absorption ability, ξ2 represents the weight coefficient of the coupling degree between the root absorption ability and the transpiration intensity, ξ3 represents the weight coefficient of the coupling degree between the photosynthetic rate and the transpiration intensity, and ρ P,ARepresents the coupling degree ρ between the photosynthetic rate P and the root absorption capacity A A,T Represents the coupling degree ρ between the root absorption capacity A and the transpiration intensity T P,T Represents the coupling degree between the photosynthetic rate P and the transpiration intensity T. By iteratively solving ξ1, ξ2, and ξ3 to maximize the collaborative optimization function, the optimal coupling coefficient combination of photosynthesis-root-transpiration is obtained;

[0139] S46. Substitute the optimal coupling coefficient combination of photosynthesis-root-transpiration into the coupling index sequence to generate a photosynthesis-root-transpiration collaborative curve.

[0140] In this embodiment, S5 specifically includes:

[0141] S51. Obtain the coupling relationship among the three represented by the photosynthesis-root-transpiration collaborative curve, discretize the photosynthesis-root-transpiration collaborative curve on the unified time axis to form a time series sample set for calculating the promotion factor and the regulation factor;

[0142] S52. Based on the non-linear dynamic mapping method, define the promotion factor φ P→A (t) of photosynthesis on root absorption:

[0143]

[0144] where P(τ) represents the photosynthetic rate at time τ, κ1 represents the photosynthesis influence coefficient, μ1 represents the root absorption threshold, and (x) + represents taking 0 when x < 0 to ensure that the promotion factor accumulates only when the photosynthetic rate is greater than the threshold;

[0145] S53. Based on the non-linear dynamic mapping method, define the regulation factor φ A→T (t) of root absorption on transpiration intensity:

[0146]

[0147] where A(τ) represents the root water and fertilizer absorption amount at time τ, κ2 represents the root regulation coefficient, and μ2 represents the transpiration trigger threshold;

[0148] S54. Combine the promotion factor φ P→A (t) of photosynthesis on root absorption and the regulation factor φ A→T (t) of root absorption on transpiration intensity to construct an adaptive regulation function Ψ control (t):

[0149] Ψ control (t) = ζ1φ P→A (t) + ζ2φ A→T (t);

[0150] Among them, ζ1 represents the weight coefficient of the promotion factor, ζ2 represents the weight coefficient of the regulation factor, and ζ1 and ζ2 are updated by minimizing the fertilization-irrigation deviation.

[0151] Reference Figure 2 , an intelligent fertilization and irrigation regulation device based on deep learning, includes the following modules:

[0152] The data acquisition module is used to deploy a sensor network and an unmanned aerial vehicle remote sensing device in the target farmland, collect soil-crop-meteorological data, and obtain the original agricultural environment dataset.

[0153] The data preprocessing module is used to denoise the original agricultural environment dataset, standardize the time scale and space scale, and generate a spatio-temporally consistent agricultural dataset.

[0154] The multi-layer neuromorphic computing module includes:

[0155] The photosynthesis layer modeling unit is used to predict the dynamic response of crop leaves to environmental factors based on a spiking neural network combined with adaptive adjustment of synaptic weights, and generate a photosynthesis dynamic curve.

[0156] The root system layer modeling unit is used to simulate the water and fertilizer absorption capacity of the root system at different depths by using a topological graph attention network and variational inference, predict the water and fertilizer dynamics in the rhizosphere soil, and generate a root system water and fertilizer absorption curve.

[0157] The transpiration layer modeling unit is used to generate a transpiration-water supply curve based on the crop transpiration rate and soil moisture status by using a neural tensor network combined with a dynamic environment perception mechanism.

[0158] The curve correlation analysis module is used to perform correlation analysis based on the photosynthesis dynamic curve, the root system water and fertilizer absorption curve, and the transpiration-water supply curve, calculate the coupling relationship between the photosynthetic rate, the root system absorption capacity, and the transpiration intensity, and construct a photosynthesis-root system-transpiration synergy curve.

[0159] The non-linear dynamic mapping module is used to calculate the promotion factor of photosynthesis on root system absorption and the regulation factor of root system absorption on transpiration intensity based on the photosynthesis-root system-transpiration synergy curve, and construct an adaptive regulation function.

[0160] The time series regulation module is used to perform time series analysis on the adaptive regulation function, and formulate a final fertilization and irrigation regulation plan according to the crop growth cycle, environmental changes, and soil water and fertilizer status.

[0161] Example 1:

[0162] To verify the feasibility of the present invention in implementation, the present invention was applied to a wheat planting base in a national modern agricultural demonstration area. The base is located in a temperate monsoon climate zone, with an average annual precipitation of about 600 mm and an evaporation rate as high as 1200 mm. The soil type is loamy soil, which has good water retention, but is prone to local waterlogging and uneven soil nutrients. Traditional fertilization and irrigation methods mainly rely on empirical judgment or simple fixed patterns, such as irrigating and fertilizing at fixed time intervals, and cannot accurately match the real-time water and fertilizer requirements of crops, resulting in low water and fertilizer utilization rates, affecting crop growth, and at the same time increasing the risks of water resource waste and environmental pollution. To solve this problem, the intelligent fertilization and irrigation regulation method of the present invention was tested in this wheat planting base.

[0163] In the test in this base, a sensor network and an unmanned aerial vehicle (UAV) remote sensing device were installed in a 30-hectare test field to collect real-time soil-crop-meteorological data, including soil moisture (water content, capillary water), soil nutrients (nitrogen, phosphorus, potassium content), electrical conductivity, pH value, organic matter content, leaf chlorophyll content (SPAD value), photosynthetic rate, transpiration, air temperature and humidity, precipitation, wind speed, solar radiation and other data. All data was transmitted to the intelligent fertilization and irrigation regulation system through a wireless network and was subjected to denoising, data completion and standardization processing to generate a high-quality, spatio-temporally consistent agricultural data set.

[0164] After the data processing was completed, the system was analyzed based on a multi-layer neuromorphic computing model. First, the photosynthetic layer used a spiking neural network combined with synaptic weight adaptive adjustment to calculate the dynamic curve of crop leaf photosynthesis in real time and predict the change trend of the photosynthetic rate in the next 24 hours. The root layer used a topological graph attention network combined with variational inference to model the water and fertilizer absorption capabilities of roots at different depths and calculate the real-time requirements of roots for water and fertilizer. The transpiration layer established a non-linear coupling relationship between the transpiration rate and soil moisture through a neural tensor network combined with a dynamic environment perception mechanism and predicted the future water evaporation rate. After obtaining the dynamic curves of photosynthesis, root absorption and transpiration, the system performed correlation analysis, calculated the coupling relationship between the photosynthetic rate, root absorption capacity and transpiration intensity, and constructed a photosynthesis-root-transpiration synergy curve to optimize the fertilization and irrigation strategies.

[0165] The experiment was conducted for four months, covering the jointing stage, booting stage, filling stage, and maturity stage of wheat growth. The intelligent fertilization and irrigation control system dynamically adjusts the irrigation frequency, fertilization amount, and water-fertilizer ratio according to the water and fertilizer requirements at different growth stages. For example, during the jointing stage, the system detects that the roots have a high demand for nitrogen fertilizer and a relatively fast transpiration rate. Therefore, the system adopts a high-nitrogen and low-phosphorus mode and combines drip irrigation to reduce water evaporation loss. During the filling stage, the system finds that the photosynthetic rate reaches its peak and the roots' demand for phosphorus and potassium increases. Therefore, the fertilization plan is adjusted to a high-phosphorus and high-potassium mode, and intermittent irrigation is used to prevent waterlogging and root hypoxia caused by excessive water. Throughout the process, the system continuously optimizes the water and fertilizer parameters through an adaptive control function, making the water and fertilizer supply highly match the crop growth requirements.

[0166] To evaluate the effectiveness of the present invention, a control group (using traditional fertilization and irrigation methods) and an experimental group (applying the intelligent control system of the present invention) were set up in the experimental field. The experimental results show that the average wheat yield of the experimental group increased by 8.6%, and the maximum yield increase reached 12.3%. The water and fertilizer use efficiency of the experimental group was significantly improved, with the irrigation water usage reduced by 21.7% and the fertilization amount reduced by 18.9%, while the crop growth status was better than that of the control group. In addition, the residual amounts of nitrogen, phosphorus, and potassium in the soil of the experimental group decreased by 12.4%, and the pH value fluctuation range was more stable, indicating that the intelligent fertilization and irrigation system effectively reduced nutrient loss and improved soil health. Through multiple tests, the system showed strong adaptability to the crop growth conditions under different climate conditions (sunny days, cloudy days, rainy days), indicating that the present invention can effectively improve the accuracy and stability of water and fertilizer management.

[0167] Table 1 Comparison test data table between traditional mode and intelligent control

[0168]

[0169]

[0170] The experimental results in Table 1 above show that the intelligent fertilization and irrigation control system of the present invention has significant advantages in multiple key indicators. First, in terms of wheat yield, the average yield of the experimental group was 554 kg / mu, which was 8.6% higher than that of the control group (510 kg / mu), and the maximum yield increase rate reached 12.3%. This result indicates that with the support of intelligent water and fertilizer regulation, the crop growth received more scientific and reasonable water and fertilizer supply, thus improving the overall yield. In addition, the SPAD value (chlorophyll content) of the plants in the experimental group reached 42.6, which was 8.9% higher than that of the control group (39.1), indicating that the photosynthesis efficiency of the crop was higher and the leaf nutrient status was healthier.

[0171] The improvement of water and fertilizer use efficiency is a major highlight of the present invention. The irrigation water consumption of the experimental group was only 344 m 3 / mu, a decrease of 21.7% compared to 440 m per mu in the control group, fully demonstrating the precise water resource management ability of the intelligent regulation system. Traditional irrigation methods often rely on fixed cycles and farmers' experience, while this system can monitor the transpiration rate of crops, soil moisture dynamics, and environmental factors in real time, thereby reasonably adjusting irrigation strategies to effectively reduce water resource waste while ensuring crop growth. Similarly, in terms of the total fertilization amount, the fertilization amount in the experimental group was 61 kg / mu, a decrease of 18.9% compared to 75 kg / mu in the control group, indicating that the present invention can reduce chemical fertilizer input while maintaining high yields, reduce agricultural production costs, and at the same time reduce the risk of environmental pollution. 3 From the perspective of soil health, the intelligent regulation system of the present invention significantly reduces the nitrogen, phosphorus, and potassium residues in the soil. Among them, the nitrogen fertilizer residue decreases from 26.5 mg / kg in the control group to 23.3 mg / kg, the phosphorus fertilizer residue decreases from 19.2 mg / kg to 16.8 mg / kg, and the potassium fertilizer residue decreases from 15.7 mg / kg to 13.8 mg / kg, a reduction of 12.1%, 12.5%, and 12.1% respectively. These data indicate that the intelligent fertilization regulation system can accurately match the nutrient requirements of crops, avoid soil pollution problems caused by excessive fertilization, and at the same time reduce the risk of soil salinization. In addition, the soil pH value in the experimental group fluctuates less, while that in the control group fluctuates more, indicating that the intelligent fertilization plan can effectively maintain the acid-base balance of the soil and is helpful for improving soil health in the long term.

[0172]

[0173]

[0174] 2 In terms of the physiological performance of crops, the transpiration rate of crops in the experimental group was 2.63 mmol / m 2 / s, while that in the control group was 2.45 mmol / m / s, an increase of 7.3%. The increase in transpiration rate means that crops can absorb water and nutrients more efficiently and maintain a better physiological metabolism level. The present invention constructs a photosynthesis-root-transpiration synergistic curve, accurately calculates the relationship between photosynthetic rate, root water and fertilizer absorption ability, and transpiration intensity, and dynamically adjusts fertilization and irrigation strategies, thereby optimizing the water and fertilizer utilization efficiency of crops and enabling them to obtain optimal water and fertilizer supply at different growth stages.

[0175] Overall, the intelligent fertilization and irrigation regulation system of the present invention shows superiority in improving crop yields, optimizing water and fertilizer management, reducing soil pollution, and enhancing crop growth status. Compared with the traditional mode, this system realizes a precise, intelligent, and dynamically regulated water and fertilizer management plan, making agricultural production more scientific and sustainable. This innovative technology not only improves agricultural production efficiency, but also reduces water and fertilizer input costs and provides effective support for ecological environmental protection.

[0175] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, should be covered by the protection scope of the present invention.

Claims

1. An intelligent fertilization and irrigation regulation method based on deep learning, characterized in that, It includes the following steps: S1. Deploy a sensor network and an unmanned aerial vehicle (UAV) remote sensing device in the target farmland, collect soil-crop-meteorological data, and construct an original agricultural environment dataset; S2. Preprocess the original agricultural environment dataset, denoise the data, and perform standardization processing on the time scale and space scale to generate a spatiotemporally consistent agricultural dataset; S3. Based on the spatiotemporally consistent agricultural dataset, construct a multi-layer neuromorphic computing model, and the multi-layer neuromorphic computing model includes a photosynthesis layer, a root system layer, and a transpiration layer: S31. The photosynthesis layer is based on a spiking neural network combined with adaptive adjustment of synaptic weights to predict the dynamic response of crop leaves to environmental factors and generate a photosynthesis dynamic curve; S32. The root system layer uses a topological graph attention network and variational inference to simulate the water and fertilizer absorption capacity of roots at different depths, predict the dynamic water and fertilizer in the rhizosphere soil, and generate a root system water and fertilizer absorption curve; S33. The transpiration layer uses a neural tensor network combined with a dynamic environment perception mechanism to generate a transpiration-water supply curve based on the crop transpiration rate and soil moisture status; S4. Conduct correlation analysis based on the photosynthesis dynamic curve, the root system water and fertilizer absorption curve, and the transpiration-water supply curve, calculate the coupling relationship between the photosynthetic rate, the root absorption capacity, and the transpiration intensity, and construct a photosynthesis-root system-transpiration synergy curve; S5. Based on the photosynthesis-root system-transpiration synergy curve, use a non-linear dynamic mapping method to calculate the promotion factor of photosynthesis on root absorption and the regulation factor of root absorption on transpiration intensity, and construct an adaptive regulation function; S6. Conduct a time series analysis on the adaptive regulation function to formulate a final fertilization and irrigation regulation plan.

2. The intelligent fertilization and irrigation regulation method based on deep learning according to claim 1, wherein, The specific content of S31 includes: S311. Based on the spatiotemporally consistent agricultural dataset, obtain the light intensity, air temperature, air humidity, carbon dioxide concentration, and leaf temperature of crop leaves in the target area, and construct a photosynthetic environment state vector; S312. Construct a photosynthetic computing model based on a spiking neural network, use the photosynthetic environment state vector as the input, the input layer neurons receive the photosynthetic environment state vector, and the hidden layer is composed of spiking neurons. The kinetic equation for the membrane potential of each spiking neuron changing with time is: Where U represents the membrane potential, τ m represents the time constant, I syn represents the synaptic input current, and uses a gated pulse trigger mechanism to control the neuron firing frequency; S313. Calculate the leaf photosynthesis intensity based on the spiking neural network and calculate the instantaneous photosynthetic rate of the crop; Among them, P t represents the instantaneous photosynthetic rate of crops, A max represents the maximum photosynthetic rate, α represents the light response coefficient, I represents the light intensity, β represents the carbon dioxide response coefficient, represents the carbon dioxide concentration, R d represents the dark respiration rate; S314. Optimize the synaptic weights based on the photosynthesis historical data and dynamically adjust the synaptic weights of the spiking neurons; w ij (t + 1) = w ij (t) + η·S(i, j)·[λE t + γP t ; where, w ij (t + 1) represents the synaptic weight between the i-th neuron and the j-th neuron at time step t + 1, w ij (t) represents the synaptic weight between the i-th neuron and the j-th neuron at time step t, η represents the learning rate, S(i, j) represents the synaptic connection strength between the i-th neuron and the j-th neuron, λ and γ represent adjustment coefficients, E t represents the photosynthetic environmental state vector; S315. Perform curve fitting on the instantaneous photosynthetic rates of the crop at multiple time steps, and use the spiking neural network with optimized synaptic weights to generate a photosynthesis dynamic curve.

3. The intelligent fertilization and irrigation regulation method based on deep learning according to claim 1, wherein The specific content of S32 includes: S321. Based on the spatiotemporally consistent agricultural dataset, obtain the soil moisture content, soil nutrient concentration, and conductivity at different depths of the root system, and construct a root system topological graph G=(V, E) in combination with the crop root system morphology information, where the node V represents the positions of each depth of the root system, and the edge E represents the water and fertilizer transfer relationship; S322. Use a topological graph attention network to perform weighted calculation on the root system topological graph, and obtain the attention coefficient of a node to its neighbor nodes based on the attention mechanism; S323. Estimate the water and fertilizer absorption rate of the root system at the nodes through variational inference: A p (t) = k w W d + k n N d + k p P d + k k K d - ζ p t; Among them, A p (t) represents the water and fertilizer absorption rate of the p-th node at time step t, k w represents the water absorption coefficient, k n represents the nitrogen absorption coefficient, k p represents the phosphorus absorption coefficient, k k represents the potassium absorption coefficient, W d represents the soil water content, N d represents the soil nitrogen concentration, P d represents the soil phosphorus concentration, K d represents the soil potassium concentration, ζ p represents the root decay factor of the p-th node, and t represents the time step; S324. Posterior distribution modeling is performed on the water and fertilizer absorption rate, and the posterior distribution p(A p (t)|E r,p,t ) of water and fertilizer absorption at the node by the root system is defined, where E r,p,t represents the rhizosphere environmental state vector of the p-th node at time step t. The variational inference method is used for iterative update to minimize the deviation between the measured value and the estimated value, and the optimal absorption coefficient is obtained; S325. Combine the attention coefficient and the optimal absorption coefficient, summarize the water and fertilizer absorption rates of each node at different time steps to form a dynamic water and fertilizer sequence of the root system, correct the cross-node water and fertilizer transfer error using the edge weights in the root system topological graph, and re-estimate the nodes with noise or anomalies to obtain a unified water and fertilizer absorption sequence; S326. Perform time series analysis on the unified water and fertilizer absorption sequence, and generate a water and fertilizer absorption curve of the root system according to the change trends of the water and fertilizer absorption rates of multiple nodes at different depths.

4. An intelligent fertilization and irrigation regulation method based on deep learning according to claim 1, characterized in that, The specific steps of S33 are as follows: S331. Based on the spatio-temporally consistent agricultural dataset, obtain the transpiration rate, soil moisture content, ambient temperature, air humidity, and meteorological fitness of the target farmland crops to form a transpiration environment vector E tr,t ; S332. Perform feature encoding on the transpiration environment vector, and input the encoded result into a neural tensor network to establish multi-factor interaction relationships, where the core tensor mapping of the neural tensor network satisfies: where f represents the neural tensor network, x1 and x2 respectively represent different parts of the transpiration environment vector, U k represents the k-th component of the tensor core, σ represents the activation function, V represents the weight matrix, and b represents the bias term; S333. Combine the dynamic environment perception mechanism, adjust the mapping weights of the neural tensor network according to the changes in wind speed, light, and soil humidity, so that the sensitivity of transpiration to external conditions is updated over time, and reduce the prediction error caused by environmental fluctuations through multiple iterations; S334. Define a transpiration-water supply coupling function for comprehensively describing the non-linear relationship between the transpiration rate and water supply: T couple (t) = α1exp(θ·E l (t)) - β1ln(1 + γ1W d ); Among them, T couple (t) represents the degree of non-linear coupling between the crop transpiration rate and the soil water supply at time step t, α1 represents the transpiration amplification coefficient, β1 represents the water inhibition coefficient, γ1 represents the soil water gain factor, exp represents the exponential function, E l (t) represents the transpiration rate of the crop, W d represents the soil water content; S335. Based on the transpiration-water supply coupling function, the stochastic gradient descent algorithm is used to minimize the transpiration-water supply residual, and the T couple (t) obtained at each time step t is integrated into a transpiration-water supply curve.

5. The intelligent fertilization and irrigation regulation method based on deep learning according to claim 1, characterized in that The specific steps of S4 are as follows: S41. Obtain the dynamic curve of photosynthesis, the water and fertilizer absorption curve of the root system, and the transpiration-water supply curve, align the three curves at the same time step, and construct a time series data set; S42. Perform interpolation and smoothing processing on the time series data set, remove outliers, and use the dynamic time warping method to correct the misalignment of local peaks to obtain a complete and comparable three-curve sequence; S43. Calculate the coupling degree ρ between any two of the photosynthetic rate, root absorption capacity, and transpiration intensity i,j : Among them, T1 represents the total number of time steps, X i (t) and X j (t) respectively represent the values of any two of the three curves at time step t, represents the average value of the curve where X i (t) is located, represents the average value of the curve where X j (t) is located; S44. Construct a non-linear mapping function according to the coupling degree, perform multi-dimensional mapping on the three curves and output a coupling index sequence, and extract the key coupling characteristics of photosynthesis, root absorption, and transpiration intensity; S45. Define a collaborative optimization function: F(ξ1, ξ2, ξ3) = ξ1ρ P,A + ξ2ρ A,T + ξ3ρ P,T ; Among them, F(ξ1, ξ2, ξ3) represents the collaborative optimization function, ξ1 represents the weight coefficient of the coupling degree between photosynthetic rate and root absorption capacity, ξ2 represents the weight coefficient of the coupling degree between root absorption capacity and transpiration intensity, ξ3 represents the weight coefficient of the coupling degree between photosynthetic rate and transpiration intensity, and ρ P,A represents the coupling degree between photosynthetic rate P and root absorption capacity A, and ρ A,T represents the coupling degree between root absorption capacity A and transpiration intensity T, and ρ P,T represents the coupling degree between photosynthetic rate P and transpiration intensity T. By iteratively solving ξ1, ξ2, and ξ3 to maximize the collaborative optimization function, the optimal coupling coefficient combination of photosynthesis-root-transpiration is obtained; S46. Substitute the optimal coupling coefficient combination of the photosynthesis-root-transpiration into the coupling index sequence to generate a photosynthesis-root-transpiration collaborative curve.

6. The intelligent fertilization and irrigation regulation method based on deep learning according to claim 1, wherein The specific steps of S5 are as follows: S51. Obtain the coupling relationship among the three represented by the photosynthesis-root-transpiration collaborative curve, discretize the photosynthesis-root-transpiration collaborative curve on a unified time axis to form a time series sample set for calculating the promotion factor and the regulation factor; S52. Define the promotion factor φ P→A (t) of photosynthesis on root absorption based on the non-linear dynamic mapping method: Among them, P(τ) represents the photosynthetic rate at time τ, κ1 represents the photosynthesis influence coefficient, μ1 represents the root absorption threshold, and (x) + represents taking 0 when x < 0, ensuring that the promotion factor accumulates only when the photosynthetic rate is greater than the threshold; S53. Based on the non-linear dynamic mapping method, define the regulation factor φ A→T (t) of root absorption on transpiration intensity: Among them, A(τ) represents the water and fertilizer absorption amount of the root system at time τ, κ2 represents the root system regulation coefficient, and μ2 represents the transpiration trigger threshold; S54. Combine the promotion factor φ P→A (t) of photosynthesis on root absorption and the regulation factor φ A→T (t) of root absorption on transpiration intensity to construct an adaptive regulation function Ψ control (t): Ψ control (t) = ζ1φ P→A (t) + ζ2φ A→T (t); Among them, ζ1 represents the weight coefficient of the promotion factor, ζ2 represents the weight coefficient of the regulation factor, and ζ1 and ζ2 are updated by minimizing the fertilization-irrigation deviation.

7. An intelligent fertilization and irrigation regulation device based on deep learning, which executes the method for intelligent fertilization and irrigation regulation based on deep learning according to any one of claims 1 to 6, characterized in that, It includes the following modules: A data acquisition module for deploying a sensor network and an unmanned aerial vehicle remote sensing device in the target farmland to collect soil-crop-meteorological data and obtain an original agricultural environment data set; A data preprocessing module for denoising the original agricultural environment data set and standardizing the time scale and space scale to generate a spatio-temporally consistent agricultural data set; Multi-layer neuromorphic computing module, comprising: A photosynthetic layer modeling unit, configured to predict the dynamic response of crop leaves to environmental factors based on a spiking neural network combined with adaptive adjustment of synaptic weights, and generate a photosynthesis dynamic curve; A root system layer modeling unit, configured to simulate the water and fertilizer absorption capacity of the root system at different depths and predict the dynamic water and fertilizer in the rhizosphere soil by using a topological graph attention network and variational inference, and generate a root system water and fertilizer absorption curve; A transpiration layer modeling unit, configured to generate a transpiration-water supply curve based on the crop transpiration rate and soil moisture condition by using a neural tensor network combined with a dynamic environment perception mechanism; A curve correlation analysis module, configured to perform correlation analysis based on the photosynthesis dynamic curve, the root system water and fertilizer absorption curve, and the transpiration-water supply curve, calculate the coupling relationship between the photosynthetic rate, the root system absorption capacity, and the transpiration intensity, and construct a photosynthesis-root system-transpiration synergy curve; A non-linear dynamic mapping module, configured to calculate the promotion factor of photosynthesis on root system absorption and the regulation factor of root system absorption on transpiration intensity based on the photosynthesis-root system-transpiration synergy curve, and construct an adaptive regulation function; A timing regulation module, configured to perform timing analysis on the adaptive regulation function, and formulate a final fertilization and irrigation regulation plan according to the crop growth cycle, environmental changes, and soil water and fertilizer status.

Images