Precise fertilization and moisture management system for traditional Chinese medicinal materials

Abstract

The invention discloses a precise fertilization and moisture management system for traditional Chinese medicinal materials, and relates to the technical field of agricultural planting, and the system comprises a data collection module which is used for dividing a planting region into a plurality of sub-regions according to root distribution characteristics and environmental data, and collecting soil parameters, medicinal material growth states and environmental data of each sub-region in real time; the intelligent decision-making module is used for processing the collected data based on a planting prediction model constructed by a deep learning algorithm, dynamically predicting the fertilization demand quantity and the water demand quantity of the medicinal materials, analyzing the demand quantity according to a cost optimization algorithm, and generating a control strategy; wherein the control strategy comprises a fertilization amount, a water consumption amount and a proportioning scheme of an organic fertilizer and a chemical fertilizer; and the execution control module is used for controlling the mixing of the two paths of fertilizer liquid, the self-adaptive adjustment of the fertilizer injection depth and the adjustment operation of the irrigation flow based on the control strategy. The technical problems that root system microenvironment detection is difficult, water and fertilizer prediction is inaccurate, and distribution is uneven are solved.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of agricultural planting, and particularly relates to a precise fertilization and water management system for traditional Chinese medicinal materials. BACKGROUND

[0002] With the development of modern agriculture towards refinement and intelligence, precise fertilization and water management technology has become a key approach to improving resource utilization efficiency and crop yield. In the specific field of traditional Chinese medicinal material planting, precise fertilization and water management face more complex challenges. Traditional Chinese medicinal material planting has special requirements for water and fertilizer management, such as the accumulation of medicinal components, large differences in root characteristics, and complex mountainous planting environments. Traditional fertilization methods have low fertilizer utilization rate and environmental pollution. Traditional fertilization modes usually rely on experience and often result in insufficient or excessive fertilizer input, which not only poses potential pollution risks to the environment but also affects the healthy growth of crops and reduces economic benefits. Although the intelligent fertilization system in the prior art can achieve a certain degree of precise control, it is mostly designed for field crops or economic fruit trees and does not fully consider the particularity of traditional Chinese medicinal material planting.

[0003] In particular, the prior art has obvious deficiencies in the following aspects: first, there is a lack of precise monitoring capability for the root microenvironment of traditional Chinese medicinal materials. Traditional sensors are mostly deployed on the soil surface and cannot reflect the true conditions of the root active layer. Second, existing prediction models are mostly based on a single data type and cannot effectively integrate multi-dimensional information such as soil parameters, crop images, and environmental factors. Third, the adaptability of the fertilizer liquid mixing system to high-viscosity organic fertilizer is poor, and blockage problems easily occur. Fourth, the control effect of water and fertilizer distribution uniformity in mountainous planting environments is not ideal. These deficiencies seriously restrict the standardization and intelligent development of traditional Chinese medicinal material planting, and a targeted precise fertilization and water management system is urgently needed to solve these problems. SUMMARY

[0004] The present application aims to at least solve one of the technical problems existing in the prior art; for this purpose, the present application proposes a precise fertilization and water management system for traditional Chinese medicinal materials to solve the technical problems of difficult detection of root microenvironment, inaccurate prediction of water and fertilizer, and uneven distribution.

[0005] To achieve the above-mentioned purpose, the first aspect of the present application provides a precise fertilization and water management system for traditional Chinese medicinal materials, comprising: a data acquisition module for dividing the planting area into multiple sub-areas according to the root distribution characteristics and environmental data, and acquiring the soil parameters, growth status of medicinal materials, and environmental data of each sub-area in real time; The intelligent decision module is configured to process the collected data based on a planting prediction model constructed by a deep learning algorithm, dynamically predict the fertilizer demand and water demand of the medicinal materials, and analyze the demand according to a cost optimization algorithm to generate a control strategy, wherein the control strategy includes the amount of fertilizer, the amount of water, and the ratio scheme of organic fertilizer and chemical fertilizer. The execution control module is configured to control the mixing of the double-path fertilizer solution, the adaptive adjustment of the fertilizer injection depth, and the adjustment operation of the irrigation flow based on the control strategy.

[0006] Based on the above technical solutions, the traditional Chinese medicinal material precision fertilization and water management system can realize scientific division and real-time monitoring of the planting area through the data acquisition module, ensuring the comprehensiveness and pertinence of data acquisition. The intelligent decision module dynamically analyzes the water and fertilizer demand by using the prediction model constructed by the deep learning algorithm, and generates an economic and efficient control strategy by combining the cost optimization algorithm, which can improve the accuracy of decision-making and the rationality of resource allocation. The execution control module realizes automation and refinement of the operation process by mixing the double-path fertilizer solution, adaptively adjusting the fertilizer injection depth, and accurately controlling the irrigation flow, effectively improving the utilization rate of water and fertilizer, reducing labor costs and environmental risks, and enhancing the adaptability of the system to the diversified growth needs of traditional Chinese medicinal materials, providing reliable technical support for improving the quality and yield of medicinal materials.

[0007] Further, the planting area is divided into a plurality of sub-areas according to the root distribution characteristics and environmental data, including: The sampling depth is determined by the root distribution characteristics, and the sampling density is determined according to the grid distribution method in combination with the environmental data; wherein the environmental data includes slope, aspect, and altitude in geographic environmental data, and rainfall and evaporation in weather data; The soil parameters and terrain indexes are collected according to the sampling density and sampling depth, and a continuous distribution map is generated by the Kriging spatial interpolation method; The termination area is divided into a plurality of sub-areas by using the K-means clustering algorithm; wherein the feature vectors of clustering include the soil nutrient comprehensive index, the root depth coefficient, and the terrain factor; The soil nutrient comprehensive index is calculated by weighting the nitrogen, phosphorus, and potassium NPK concentrations in the soil parameters, the root depth coefficient is the normalized ratio of the actual root depth of the medicinal materials to the standard root depth, and the terrain factor is represented by the first principal component obtained by principal component analysis of the slope, aspect, and altitude in the geographic environmental data.

[0008] Further, the data acquisition module includes: The multi-source sensor unit includes soil nitrogen, phosphorus and potassium sensors, pH electrodes, conductivity sensors, and soil humidity-temperature composite sensors, which collect soil humidity, soil temperature, NPK concentration, pH value, and conductivity data; the pH electrode is used to measure soil acidity and alkalinity, and the conductivity sensor is used to measure soil salinity, and the sensor unit is embedded in the root layer with an embedding depth adjusted according to the root distribution characteristics; The medicinal material growth monitoring unit includes a multispectral camera and a stem micro-variation sensor, which collect crown image data, stem diameter change data, and chlorophyll index data; the multispectral camera is used to collect crown images, and the stem micro-variation sensor is used to measure the stem diameter change rate; The environmental data acquisition unit is used to collect rainfall, evaporation and temperature data; The data acquisition intervals of each unit are consistent, and the data are stored in a time sequence and transmitted to the intelligent decision-making module.

[0009] Further, the data preprocessing process of the data acquisition module includes: The original data collected by the multi-source sensor unit and the environmental data acquisition unit are denoised by the Kalman filtering algorithm to obtain first preprocessed data; The first preprocessed data are reduced in dimension by using the principal component analysis method, and soil feature vectors and environmental feature vectors are generated; the soil feature vectors include soil humidity, soil temperature, NPK concentration, pH value and conductivity, and the environmental feature vectors include rainfall, evaporation, temperature and light intensity.

[0010] Further, the planting prediction model constructed based on the deep learning algorithm includes a feature extraction module, a cross-dimension fusion module and a prediction output module; wherein, The feature extraction module includes a soil feature submodule, an image feature submodule and an environmental feature submodule, which are used to extract features from the input soil feature vectors, multispectral image data and environmental feature vectors, respectively, and output soil time sequence feature vectors, image time sequence feature vectors and environmental time sequence feature vectors; The cross-dimension fusion module includes a feature alignment unit, a dynamic weight calculation unit and an adaptive fusion unit, which are used to process the multi-dimensional time sequence feature vectors, traditional Chinese medicinal material variety labels and growth stage labels, balance the contribution of multi-dimensional information, and output an optimized multi-dimensional fusion feature vector; The prediction output module includes a time sequence encoding unit and a double-task decoding unit, which are used to perform multi-step water and fertilizer demand prediction and output daily fertilizer application amount and water consumption amount.

[0011] Further, the working process of the feature extraction module includes: The input of the soil feature submodule is a soil feature vector, a first soil feature of a soil parameter is extracted through a backbone network, a channel attention mechanism is used to weight the first soil feature, a second soil feature after weighting is obtained, and the second soil feature is input into an adversarial training unit and a fully connected layer to obtain a soil time sequence feature vector; the adversarial training unit is a "generator-discriminator" double-branch structure, the generator and the discriminator are both built based on a multilayer perception mechanism, and the generator is used to generate a soil feature with noise, and the discriminator is used to distinguish the second soil feature and the soil feature with noise; The input of the image feature submodule is multispectral image data, a first image feature is extracted through a backbone network, then the first image feature is input into a Visual Transformer module to extract global features, and at the same time, the first image feature is input into a band attention mechanism for adaptive reinforcement of local features, and then the global features and the local features are spliced to obtain an image time sequence feature vector; The input of the environment feature submodule is an environment feature vector, a first environment feature is extracted through a backbone network, then the first environment feature is input into a multilayer hollow convolution and a multiscale attention mechanism, and after being optimized by an activation function and a Dropout layer, an environment time sequence feature vector is obtained.

[0012] Further, the working process of the cross-dimension fusion module includes: The feature alignment unit aligns the input of each dimension time sequence feature vector, first unifies the vector dimension through a fully connected layer, then performs Z-score standardization and L2 normalization, and outputs a multi-dimensional standardized feature vector with aligned dimensions and consistent distribution; The input of the dynamic weight calculation unit is a Chinese herbal medicine variety label and a growth stage label, an initial weight is given to the multi-dimensional standardized feature vector according to a built-in reference weight library, and the initial weight is adaptively adjusted according to the feature confidence of the multi-dimensional data, and a dynamic weight coefficient of each dimension is output; the reference weight library includes the weight of the Chinese herbal medicine variety, the growth stage, and each dimension feature; The input of the fusion unit is a multi-dimensional standardized feature vector and a dynamic modal weight coefficient, the multi-dimensional standardized feature vector is weighted and summed based on the weight coefficient to obtain a multi-dimensional fusion feature vector.

[0013] Further, the working process of the prediction output module includes: The input of the time sequence encoding unit is a multi-dimensional fusion feature vector, a bidirectional gated recurrent unit is used to time sequence encode the fusion feature vector, and a time sequence encoding feature is output; The input of the double-task decoding is the time series encoding feature, the water and fertilizer threshold of Chinese herbal medicine planting, and the field water holding capacity parameter, a double output head is constructed based on the LSTM module, and a water and fertilizer coupling constraint is introduced, the time series encoding feature is mapped to the water and fertilizer prediction value in the future preset time period, and the daily fertilization demand and water demand are obtained; wherein, the water and fertilizer coupling constraint refers to the original prediction value output by the double output head is optimized and corrected in coordination according to the water and fertilizer threshold and the field water holding capacity parameter.

[0014] Further, the water and fertilizer coupling constraint is based on the water and fertilizer coupling equation The calculation formula is: ; wherein, is the irrigation threshold, is the fertilization threshold, is the field water holding capacity parameter; The fertilization demand and the water demand are calculated based on the theoretical water and fertilizer ratio, and the calculation formula is: , ; wherein, is the original water prediction value output by the double output head, is the original fertilization prediction value output by the double output head.

[0015] Further, the demand amount is analyzed according to the cost optimization algorithm to generate a control strategy, including: A fuzzy rule base is constructed, and the organic fertilizer replacement rate is obtained according to the fuzzy set of input variables and rule logic; wherein, the input variables include fertilizer price, medicinal material market price and current inventory, and the output variable is the organic fertilizer replacement rate; A linear programming model is established according to the organic fertilizer replacement rate, the predicted fertilization demand and the water demand; the linear programming model takes the minimization of the total cost of fertilization and water as the objective function; The linear programming model is solved, and the control strategy is generated according to the calculated fertilization amount, water amount and organic fertilizer and chemical fertilizer ratio scheme.

[0016] Further, the double-path fertilizer liquid mixing represents a double-path independent control system, including a chemical fertilizer pipe and an organic fertilizer pipe; wherein, The chemical fertilizer pipe adopts proportional-integral-derivative PID control algorithm to adjust the opening degree of electric valve, control the flow of chemical fertilizer; The organic fertilizer pipe integrates ultrasonic crushing unit and mechanical stirring unit for processing organic fertilizer liquid; wherein, the ultrasonic crushing unit decomposes fertilizer liquid particles by vibration to improve uniformity, and the mechanical stirrer generates turbulent flow in the mixing chamber to enhance the mixing effect; The double-path fertilizer liquid is mixed at the outlet of the mixing chamber according to the preset ratio.

[0017] It should be noted that the mixing process is monitored in real time by the conductivity sensor to ensure the uniformity of the fertilizer solution.

[0018] Further, the working process of self-adaptive adjustment of the fertilizer injection depth comprises: The ideal fertilizer injection depth D is calculated according to the medicinal material plant age T and the root distribution characteristics, and the calculation formula is: ; wherein, represents the minimum fertilizer injection depth, represents the maximum fertilizer injection depth, represents the complete growth period of the medicinal material. The fuzzy PID control algorithm is adopted to control the actual fertilizer injection depth, wherein the input of the fuzzy PID control algorithm is the depth deviation and the soil hardness value, and the output is the hydraulic cylinder control signal of the fertilizer injection needle; the depth deviation is the difference between the actual fertilizer injection depth and the ideal fertilizer injection depth.

[0019] It should be noted that the fertilizer injection needle and the irrigation system work synchronously, and each medicinal material is independently controlled to ensure that the fertilizer reaches the root active layer.

[0020] Further, the control method of the irrigation flow comprises: The soil water content of each sub-region is monitored by a humidity sensor, and the deviation between the actual water content and the target humidity is calculated; According to the deviation, the evaporation amount and the rainfall, a proportional control strategy is adopted to adjust the irrigation flow; The irrigation pipeline adopts a pressure regulating device, which maintains the pipeline pressure by segmented pressurization, compensates for the pressure loss according to the terrain difference, and combines the adjustable design of the dripper or the spray head to ensure the stability and precise control of the flow.

[0021] Compared with the prior art, the beneficial effects of the present application are: The present application effectively solves the problem of uneven water and fertilizer application caused by soil heterogeneity and terrain difference in traditional planting management through the regional division method combining root distribution characteristics and environmental data. Specifically, the present application generates a continuous distribution map based on the grid point method and the Kriging spatial interpolation, and divides the homogeneous sub-regions by combining the K-means clustering algorithm, so that the internal conditions of the management unit are highly consistent, which can improve the targeting of fertilization and irrigation. The design of directly embedding the multi-source sensor unit into the root layer overcomes the technical problem of response lag of traditional surface monitoring to the microenvironment of Chinese medicinal materials roots, and realizes the stereoscopic perception of the growth state of medicinal materials through synchronous acquisition of soil NPK concentration, pH value, salt content and crown layer multispectral data. The combination of Kalman filtering and principal component analysis in data preprocessing further eliminates sensor noise interference, ensuring the reliability and timeliness of the input data.

[0022] The planting prediction model constructed based on the deep learning algorithm solves the problem of insufficient adaptability of traditional prediction models to complex environmental factors and growth dynamics of medicinal materials through the collaborative work of multiple modules. In the feature extraction module, the soil feature submodule introduces an adversarial training unit to enhance feature robustness, the image feature submodule captures global canopy features and local spectral details through Visual Transformer and band attention mechanisms, and the environmental feature submodule optimizes the temporal feature extraction of meteorological data using a cavity convolution and multi-scale attention. The cross-dimensional fusion module uses a dynamic weight calculation unit to adaptively adjust the contribution of multi-source data according to the medicinal material variety and growth stage, significantly improving the model's generalization ability in variable environments. The prediction output module combines water and fertilizer coupling constraints with a dual-task decoding mechanism to achieve collaborative optimization of fertilizer and irrigation amounts, avoiding nutrient leaching caused by excessive water and preventing root damage from high fertilizer concentrations. This model can accurately predict water and fertilizer requirements at different growth stages, improving fertilizer utilization and irrigation water use efficiency, thereby reducing overall operating costs.

[0023] In addition, the execution control module solves the technical problems of traditional systems in terms of fertilizer solution mixing uniformity, precise control of fertilizer injection depth, and stability of mountain irrigation. The dual-path fertilizer solution mixing system uses PID control in combination with ultrasonic and mechanical stirring to achieve non-clogging processing of high-viscosity organic fertilizers. The fertilizer injection depth self-adaptive adjustment dynamically adjusts the fertilizer injection needle position based on the medicinal material plant age and root system model, and combines a fuzzy PID control algorithm for real-time response to soil hardness, keeping the fertilizer injection depth error within ±2 cm. The irrigation flow control method uses a combination strategy of humidity feedback and feedforward compensation, combined with a terrain pressure compensation mechanism, to ensure the stability of flow control in complex environments. Therefore, the present application promotes the development of standardized and intelligent Chinese medicinal material planting, providing core support for improving medicinal material quality and industrial efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, a brief introduction will be given below to the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.

[0025] Figure 1 A system framework diagram of a precise fertilization and water management system for Chinese medicinal materials is provided for the embodiments of the present application. Figure 2 A working flowchart of a data acquisition module in a precise fertilization and water management system for Chinese medicinal materials is provided for the embodiments of the present application. Figure 3A working process schematic diagram of an intelligent decision module in a traditional Chinese medicinal material precision fertilization and water management system provided by the embodiment of the present application is provided. Figure 4 A working process schematic diagram of an execution control module in a traditional Chinese medicinal material precision fertilization and water management system provided by the embodiment of the present application is provided. DETAILED DESCRIPTION

[0026] The technical solutions of the present application will be clearly and completely described in combination with the embodiments. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0027] Please refer to Figures 1-4 The first aspect embodiment of the present application provides a traditional Chinese medicinal material precision fertilization and water management system, which comprises: A data acquisition module is configured to divide a planting area into a plurality of sub-areas according to root distribution characteristics and environmental data, and to acquire soil parameters, medicinal material growth states and environmental data of each sub-area in real time. An intelligent decision module is configured to process the acquired data based on a planting prediction model constructed based on a deep learning algorithm, to dynamically predict fertilization demand and water demand of the medicinal material, and to analyze the demand based on a cost optimization algorithm to generate a control strategy; wherein the control strategy comprises a fertilization amount, a water amount and a proportioning scheme of organic fertilizer and chemical fertilizer. An execution control module is configured to control mixing of double-path fertilizer liquid, adaptive adjustment of fertilization depth and adjustment operation of irrigation flow based on the control strategy.

[0028] Based on this, the embodiment of the present application can realize precision, automation and refinement of traditional Chinese medicinal material planting water and fertilizer management, effectively improve water and fertilizer utilization rate, reduce manual intervention cost and environmental risk, and enhance the adaptability of the system to different traditional Chinese medicinal material varieties and complex growth environments To solve the technical problems of uneven water and fertilizer application, root system microenvironment monitoring lag, insufficient adaptability of the prediction model to complex environmental factors and medicinal material growth dynamics, easy clogging of uneven mixing of high-viscosity organic fertilizer, inability of fertilization depth to adapt to root growth, and unstable irrigation flow under complex terrain in the traditional water and fertilizer management, the embodiment of the present application provides a traditional Chinese medicinal material precision fertilization and water management system, which comprises: A data acquisition module is configured to divide a planting area into a plurality of sub-areas according to root distribution characteristics and environmental data, and to acquire soil parameters, medicinal material growth states and environmental data of each sub-area in real time.

[0029] The data acquisition module can include a soil parameter acquisition unit, a medicinal material growth state monitoring unit, an environment data acquisition unit, and a region division unit. The soil parameter acquisition unit is used to obtain soil-related core indicators. The medicinal material growth state monitoring unit captures dynamic information of medicinal material growth. The environment data acquisition unit collects planting environment-related data. The region division unit is responsible for completing the sub-region division of the planting area based on preset conditions.

[0030] In some implementations, the sub-regions can be divided according to any one or more combinations of soil texture, topographical differences, medicinal material growth period, etc. The acquisition unit can use existing conventional sensing equipment, image acquisition equipment, environmental monitoring equipment, etc. The acquisition interval can be flexibly set according to planting needs, and the data can be transmitted to subsequent modules through wired or wireless transmission.

[0031] It should be noted that the division of the sub-regions needs to ensure the relative consistency of the planting conditions in each region, and the acquired data needs to cover the soil, plant, and environment dimensions to ensure the comprehensiveness and effectiveness of the data.

[0032] For example, the soil nitrogen, phosphorus, and potassium content, humidity, etc. can be obtained by manual sampling combined with a conventional soil detector. The growth state data can be obtained by shooting plant images with an ordinary camera and measuring stem thickness with a ruler. The environmental data such as rainfall and temperature can be acquired by a small weather station. The land is divided into three sub-regions according to the landform slope.

[0033] In a possible implementation of the embodiments of the present application, the data acquisition module can be combined with the data processing module. Figure 1 As shown in Figure 2 The data acquisition module can be implemented through the following S101, S102, and S103, which will be described in detail below. S101, according to the root distribution characteristics and the environmental data, the sub-region division of the planting area is completed.

[0034] The division process focuses on improving the water and fertilizer management pertinence. Through the collaborative operation of sampling design, data modeling, and cluster analysis, the homogeneity division of the planting area is realized.

[0035] In some implementations, the division of the sub-regions includes the following steps: First, determine the sampling depth and sampling density. The sampling depth is determined by the root distribution characteristics of the traditional Chinese medicinal materials to ensure that the sampling covers the active root zone. Meanwhile, the sampling density is determined by using the grid point method in combination with the environmental data, which includes the slope, slope direction, and elevation in the geographical environmental data, and the rainfall and evaporation in the weather data, so that the sampling point distribution can adapt to the environmental differences of the planting area.

[0036] Secondly, collect relevant data and generate continuous distribution map. Collect soil parameters (including NPK concentration, etc.) and terrain index according to the determined sampling density and sampling depth; then process the collected data by using Kriging spatial interpolation method to generate the continuous distribution map of soil parameters and terrain index, which intuitively presents the distribution difference in the planting area.

[0037] Thirdly, divide the sub-regions by using K-means clustering algorithm. Take the soil nutrient comprehensive index, root depth coefficient and terrain factor as the feature vector of clustering, and divide the planting area into multiple homogeneous sub-regions by using K-means clustering algorithm; wherein, the soil nutrient comprehensive index is calculated by weighting the NPK concentration in the soil parameters, the root depth coefficient is the normalized ratio of the actual root depth of medicinal materials to the standard root depth, and the terrain factor is represented by the first principal component obtained by principal component analysis on the slope, slope direction and altitude in the geographical environment data, so as to ensure that the planting conditions in the divided sub-regions are highly consistent.

[0038] It should be pointed out that this division method can effectively solve the problem of uneven water and fertilizer application caused by soil heterogeneity and terrain difference in traditional planting, so that the planting conditions in each sub-region are highly consistent.

[0039] S102, acquire soil parameters, growth status of medicinal materials and environmental data of each sub-region by multi-unit cooperative collection.

[0040] In some implementations, multi-source sensor unit, medicinal material growth monitoring unit and environmental data acquisition unit are used for division collection: The multi-source sensor unit needs to determine the embedding depth according to the root distribution characteristics of Chinese medicinal materials first, so as to ensure that the sensor probe reaches the root active layer, and then fix the soil nitrogen, phosphorus and potassium sensor, pH electrode, conductivity sensor and soil humidity-temperature composite sensor at this depth; wherein, the soil nitrogen, phosphorus and potassium sensor is used to collect the concentration of nitrogen, phosphorus and potassium elements in the soil, the pH electrode is used to measure the soil acidity and alkalinity, the conductivity sensor is used to monitor the soil salt content, and the soil humidity-temperature composite sensor is used to synchronously collect the soil humidity and temperature data.

[0041] The medicinal material growth monitoring unit needs to deploy the multi-spectral camera on the fixed bracket above the planting area, adjust the shooting angle to completely cover the medicinal material canopy in the sub-region, and bind the stem micro-change sensor at the middle part of the medicinal material stem. The multi-spectral camera collects canopy image data to extract the chlorophyll index, and the stem micro-change sensor records the stem diameter change rate in real time.

[0042] The environmental data collection unit is installed in an open and unobstructed position in the planting area to ensure that the data collection is not disturbed by the local environment and to collect rainfall, evaporation and environmental temperature data. In addition, the data collection intervals of the three units need to be unified in advance, and all collected data are stored in time sequence data format in chronological order and transmitted to the intelligent decision module synchronously through wired or wireless transmission.

[0043] It should be noted that the design of directly embedding the multi-source sensor unit into the root layer overcomes the technical problem of response lag of traditional ground monitoring on the microenvironment of the roots of Chinese herbal medicines.

[0044] S103, preprocessing the collected raw data to generate a standardized feature vector.

[0045] The preprocessing aims to eliminate data interference and extract key features to provide reliable data input for subsequent intelligent decision-making.

[0046] In some implementations, the generation process of the standardized feature vector includes: First, the raw data is denoised. The soil humidity, temperature, NPK concentration, pH value, conductivity data collected by the multi-source sensor unit, and the rainfall, evaporation, temperature data collected by the environmental data collection unit are filtered by the Kalman filter algorithm. Further, the Kalman filter establishes the state equation and observation equation of the data, uses the estimated value at the previous time and the observed value at the current time to recursively calculate the optimal estimate at the current time, which can effectively eliminate the random noise of the sensor itself, the abnormal fluctuations caused by external environmental interference and other interference factors, and finally obtain the first preprocessed data after removing the noise.

[0047] Then, the first preprocessed data is dimensionally reduced and a feature vector is constructed. The first preprocessed data after denoising is classified and sorted according to data types, and then principal component analysis is used for dimension reduction. Specifically, the principal component analysis calculates the covariance matrix of the data, solves the eigenvalues and eigenvectors, and selects the first few principal components with higher variance contribution to replace the original high-dimensional data, which not only retains the core information but also eliminates redundant variables; on this basis, soil feature vectors and environmental feature vectors are extracted, wherein the soil feature vectors include soil humidity, soil temperature, NPK concentration, pH value, conductivity and other core soil parameters, and the environmental feature vectors include rainfall, evaporation, temperature, light intensity and other key environmental parameters, forming a standardized feature vector.

[0048] It should be noted that the cooperative use of Kalman filtering and principal component analysis not only solves the problem of noise interference in the original data, ensuring the reliability and accuracy of the data, but also simplifies the data dimension through dimension reduction, reduces the computational complexity of the subsequent planting prediction model, and improves the model processing efficiency.

[0049] It should be noted that the traditional technology often ignores soil heterogeneity and terrain differences, and unified management leads to uneven water and fertilizer application, and relies on surface monitoring, and the response to root microenvironment is lagging, and the original data noise interference is large and the redundant information is much. The application determines the sampling density by grid distribution method, generates continuous distribution map by Kriging spatial interpolation, and divides the sub-regions by combining K-means clustering algorithm, so that the internal conditions of each management unit are highly consistent, solving the problem of ignoring the heterogeneity of root microenvironment of traditional planting area division and the influence of mountain terrain, avoiding the imbalance of water and fertilizer distribution caused by uneven soil conditions, and improving the management pertinence; The multi-source sensor is directly embedded in the root layer, and the soil NPK concentration, pH value, salt and crown layer multispectral data are synchronously collected, so as to realize the stereoscopic perception of the growth state of medicinal materials and overcome the defect of lagging response of surface monitoring; The cooperative use of Kalman filter denoising and principal component analysis dimensionality reduction effectively eliminates the sensor noise interference, refines the core features, and ensures the reliability and timeliness of the input data, providing high-quality data support for subsequent decision-making.

[0050] Based on the above technical scheme, the data acquisition module realizes the homogeneity of the planting area management unit, the comprehensiveness and reliability of the collected data through sub-regional division, multi-source data acquisition and data preprocessing, lays a solid foundation for the intelligent decision-making module to dynamically predict water and fertilizer demand and generate precise control strategy, and effectively supports the fine management of Chinese herbal medicine planting.

[0051] The intelligent decision-making module is used for processing the collected data based on the planting prediction model constructed based on the deep learning algorithm, dynamically predicting the fertilizer demand and water demand of the medicinal materials, and analyzing the demand amount according to the cost optimization algorithm to generate the control strategy.

[0052] Among them, the control strategy includes the amount of fertilizer, the amount of water, and the proportioning scheme of organic fertilizer and chemical fertilizer; It can also include water and fertilizer application time, application frequency, application method and other related parameters; The planting prediction model is constructed based on the deep learning algorithm, which can realize the comprehensive analysis and demand prediction of multi-source data, and the cost optimization algorithm is used to optimize the resource input cost under the premise of meeting the growth demand of medicinal materials.

[0053] In some implementations, the deep learning algorithm can select any one or a combination of convolutional neural network (CNN), recurrent neural network (RNN), multilayer perceptron (MLP), etc., and the cost optimization algorithm can select linear programming, genetic algorithm, particle swarm optimization algorithm, etc. Optimization algorithm; The model training data can include historical planting data of different Chinese herbal medicine varieties and different growth environments.

[0054] It should be noted that the planting prediction model needs to be adaptable to different varieties of Chinese medicinal herbs and different growth stages, and the cost optimization algorithm needs to comprehensively consider the acquisition costs of fertilizer and water resources as well as the output value of medicinal herbs.

[0055] In some implementations, the specific network structure of the planting prediction model built based on deep learning algorithms includes: a feature extraction module, a cross-dimensional fusion module, and a prediction output module, realizing end-to-end water and fertilizer demand prediction. The specific structure and core process are as follows: The feature extraction module, as the basic data processing unit of the model, includes a soil feature submodule, an image feature submodule, and an environmental feature submodule, which respectively perform in-depth mining of time-series features for three types of heterogeneous data: soil, plants, and environment.

[0056] Among them, the soil feature submodule uses soil feature vectors As input, the first soil features are extracted through the backbone network. ;in, These correspond to parameters such as soil moisture, temperature, and NPK concentration. This is the weight matrix. For bias terms, The activation function is then used; channel attention mechanism is then applied to... We perform weighted summation to obtain , Here is the channel attention weight vector. It is an element-wise product; finally, a two-branch adversarial training unit, "generator-discriminator," is introduced, where the generator... Noisy soil features generated based on MLP Discriminator Through binary classification loss After optimization, the soil temporal feature vector is finally output through a fully connected layer. This significantly enhances the robustness of features; among them, This represents the mathematical expectation, used to calculate the average of the objective function within the parentheses over the distribution of the input data.

[0057] The image feature submodule uses multispectral image data As input, H, W, and C represent the image height, width, and number of spectral bands, respectively. The first image features are extracted through the backbone network. ; then Enter the VisualTransformer (ViT) module and embed it in chunks. Location encoding and multi-head self-attention mechanism Capturing global features of the canopy Simultaneously, band weights are generated through a band attention mechanism. , The c-th band features are used; adaptive enhancement of local spectral details is performed to obtain... Finally, the global and local features are concatenated to output the temporal feature vector of the image. .

[0058] The environmental feature submodule uses environmental feature vectors As input, the first environmental features are extracted via the backbone network. Then, input multiple dilated convolution layers respectively. With multi-scale attention mechanisms By capturing temporal dependence and scale features through different receptive fields, and then through activation functions... The Dropout layer suppresses overfitting, and the final output is a temporal feature vector of the environment. ; where the environmental feature vector corresponds to parameters such as rainfall and evaporation, and r is the porosity.

[0059] The cross-dimensional fusion module addresses the heterogeneity issue of multi-source temporal features, achieving feature optimization through three steps: feature alignment, dynamic weight calculation, and adaptive fusion. The feature alignment unit first uses a fully connected layer to... , , Map to the same dimension d, then perform Z-score standardization. with L2 normalization Output multi-dimensional standardized feature vectors with aligned dimensions and consistent distribution. The dynamic weight calculation unit uses the labels of Chinese medicinal materials. Growth stage labels As input, from the benchmark weight library Calling the initial weights Then, based on the feature confidence of multi-dimensional data The initial weights are adaptively adjusted to obtain dynamic weight coefficients. This ensures a reasonable allocation of contributions from various dimensions of features across different varieties and growth stages. The adaptive fusion unit performs a weighted summation of standardized features based on dynamic weights, i.e. The output is an optimized multi-dimensional fused feature vector; where, The characteristic mean, The characteristic standard deviation, This is the variance calculation function.

[0060] The prediction output module adopts a "time-series coding-dual-task decoding" architecture to achieve multi-step water and fertilizer demand prediction. The time-series coding unit uses... As input, long-term dependencies of features are captured through a bidirectional gated recurrent unit (BiGRU), and its hidden state update formula is: , , , , final output timing encoding feature ; wherein, , are update gate, reset gate, is the hidden state at time t; The double-task decoding unit constructs a double-output head based on the LSTM module, inputs , water and fertilizer threshold of Chinese herbal medicine planting , and field water holding capacity parameters , and introduces a water and fertilizer coupling constraint - first, the theoretical water and fertilizer ratio is calculated by formula , and then the original prediction value output by the double-output head is co-optimized and corrected based on the ratio, to obtain the final fertilizer demand and water demand , which not only avoids nutrient leaching caused by excessive water, but also prevents root damage caused by excessive fertilizer concentration, achieving precise co-optimization prediction of water and fertilizer demand. It should be noted that the traditional prediction model has insufficient adaptability to complex environmental factors and growth dynamics of medicinal materials, is difficult to balance the contribution of multi-source data, and lacks water and fertilizer co-optimization design, which is easy to cause nutrient leaching or root damage. In the feature extraction module of the present application, the soil feature submodule introduces an adversarial training unit to enhance the robustness of the features, the image feature submodule captures the global features of the canopy and the local spectral details through the Visual Transformer and the band attention mechanism, respectively, and the environmental feature submodule optimizes the time series feature extraction of meteorological data using the hollow convolution and multi-scale attention, realizing deep mining of heterogeneous data; the cross-dimensional fusion module adaptively adjusts the dynamic weight of multi-source data according to the variety and growth stage of Chinese herbal medicine, which significantly improves the generalization ability of the model in a variable environment; the prediction output module introduces a water and fertilizer coupling constraint, which co-optimizes and corrects the original prediction value through the theoretical water and fertilizer ratio, which not only avoids nutrient leaching caused by excessive water, but also prevents root damage caused by excessive fertilizer concentration, making the water and fertilizer demand prediction more accurate, effectively improving the fertilizer utilization rate and irrigation water use efficiency, and reducing the comprehensive operation cost.

[0061] It should be noted that the traditional prediction model has insufficient adaptability to complex environmental factors and growth dynamics of medicinal materials, is difficult to balance the contribution of multi-source data, and lacks water and fertilizer co-optimization design, which is easy to cause nutrient leaching or root damage. In the feature extraction module of the present application, the soil feature submodule introduces an adversarial training unit to enhance the robustness of the features, the image feature submodule captures the global features of the canopy and the local spectral details through the Visual Transformer and the band attention mechanism, respectively, and the environmental feature submodule optimizes the time series feature extraction of meteorological data using the hollow convolution and multi-scale attention, realizing deep mining of heterogeneous data; the cross-dimensional fusion module adaptively adjusts the dynamic weight of multi-source data according to the variety and growth stage of Chinese herbal medicine, which significantly improves the generalization ability of the model in a variable environment; the prediction output module introduces a water and fertilizer coupling constraint, which co-optimizes and corrects the original prediction value through the theoretical water and fertilizer ratio, which not only avoids nutrient leaching caused by excessive water, but also prevents root damage caused by excessive fertilizer concentration, making the water and fertilizer demand prediction more accurate, effectively improving the fertilizer utilization rate and irrigation water use efficiency, and reducing the comprehensive operation cost.

[0062] In a possible implementation manner of the embodiments of the present application, in combination Figure 1 , as shown in Figure 3 , the above-mentioned intelligent decision module can be implemented by the following S201, S202 and S203, which will be described in detail as follows: S201, receiving the preprocessed data transmitted by the data acquisition module, providing standardized input for the planting prediction model. ​

[0063] In some implementations, the received data is in a time series storage format, including time series data of soil parameters such as soil moisture, NPK concentration, pH value, time series data of environmental parameters such as rainfall and evaporation, growth state data such as canopy image and stem diameter change rate, and pre-set information of medicinal material varieties and current growth stage. In addition, the received data will be format checked to ensure that the data dimensions are consistent and the time series are continuous. If there is data missing, linear interpolation method is used for supplement to ensure the integrity of the input data.

[0064] It should be noted that the standardization of the input data has been completed by the data acquisition module. This step only focuses on the reception, verification and supplement of the data to avoid repeated preprocessing and reduce efficiency.

[0065] S202, processing the input data by a planting prediction model to dynamically output the corrected fertilizer demand and water demand.

[0066] The planting prediction model can comprehensively analyze multi-source data, first obtain the original water and fertilizer prediction value, then perform collaborative optimization and correction through water and fertilizer coupling constraints, and finally output accurate water and fertilizer demand.

[0067] In some implementations, the planting prediction model first extracts and fuses the input data through a feature extraction module and a cross-dimension fusion module, and then generates the original water prediction value and the original fertilizer prediction value by a double-task decoding unit of a prediction output module. Then, the water and fertilizer coupling constraints are introduced for correction to obtain the corrected fertilizer demand and water demand .

[0068] It should be noted that the core role of the water and fertilizer coupling constraints is to realize the collaborative optimization of the fertilizer amount and the irrigation amount, which not only avoids the leaching of nutrients caused by excessive water, but also prevents the damage to the root system caused by high fertilizer concentration.

[0069] S203, analyzing the water and fertilizer demand based on a cost optimization algorithm to generate an economic and efficient control strategy.

[0070] The core of step S203 is to combine market cost factors and water and fertilizer demand data to determine the fertilizer amount, water amount, and organic fertilizer and chemical fertilizer ratio scheme through algorithm optimization, so as to realize the balance between cost minimization and growth demand.

[0071] In some implementations, the analysis of the water and fertilizer demand by the cost optimization algorithm includes the following steps: Firstly, the input variables and output variables of the fuzzy rule base were defined. The input variables included the price of chemical fertilizer, the price of organic fertilizer, the market price of medicinal materials, and the current inventory of organic fertilizer and chemical fertilizer. The output variable was the organic fertilizer replacement rate, which was the proportion of organic fertilizer in the total amount of fertilizer. Then, the fuzzy set definition was performed for each input variable. For example, the price of chemical fertilizer was divided into three fuzzy subsets: "low price", "medium price", and "high price". The inventory of organic fertilizer was divided into three fuzzy subsets: "sufficient", "general", and "shortage". The fuzzy rule logic was preset, such as "if the price of chemical fertilizer is high, the inventory of organic fertilizer is sufficient, and the market price of medicinal materials is high, then the organic fertilizer replacement rate is high". "If the price of chemical fertilizer is low, and the inventory of organic fertilizer is short, then the organic fertilizer replacement rate is low". A complete fuzzy rule base was formed. Subsequently, the actual collected input variable data was fuzzy processed and mapped to the corresponding fuzzy subsets. The fuzzy output result was obtained through rule matching and reasoning. The fuzzy output was converted to a specific organic fertilizer replacement rate value through defuzzification processing, such as the center of gravity method.

[0072] Secondly, a linear programming model was established to minimize the cost. The corrected fertilizer demand and water demand output by S202 were used as the basic data. The model variables were defined as follows: let be the amount of organic fertilizer applied (kg / acre), be the amount of chemical fertilizer applied (kg / acre), and W be the actual irrigation water volume (m³ / acre). Then, the objective function was constructed as follows: where is the unit price of organic fertilizer (yuan / kg), is the unit price of chemical fertilizer (yuan / kg), is the cost of unit water volume (yuan / m³). The constraint conditions were set as follows: first, the total fertilizer amount constraint, , and , where r is the organic fertilizer replacement rate obtained in the first step; second, the nutrient supply constraint, , is the unit nutrient content of organic fertilizer, is the unit nutrient content of chemical fertilizer, is the total nutrient required for the growth of medicinal materials; third, the water volume constraint, to ensure that the water demand is met. If there is a water resource limit, it can be adjusted to , which is the maximum water supply upper limit; fourth, the variable non-negative constraint, .

[0073] It should be noted that the nutrient supply constraint is the core constraint condition, which needs to be accurately set by detecting the actual nutrient content of organic fertilizer and chemical fertilizer to avoid nutrient deficiency caused by simply pursuing the lowest cost.

[0074] The third step is to solve the linear programming model and generate the control strategy. A linear programming algorithm (such as the simplex method) is used to solve the above model to obtain... , The optimal solution for W is the organic fertilizer application rate, chemical fertilizer application rate, and irrigation water volume that minimize costs and meet all constraints. Then, by integrating the optimal solution with the organic fertilizer substitution rate, the core parameters of the control strategy are defined: total fertilizer application rate (…). Water consumption (W), ratio of organic fertilizer to chemical fertilizer ( At the same time, supplementary parameters such as fertilization frequency and irrigation time are added to form a complete, economical and efficient control strategy.

[0075] It should be noted that after solving the problem, it is necessary to verify whether the optimal solution meets the nutrient supply constraint. If it does not, it is necessary to backtrack and adjust the organic fertilizer substitution rate in the fuzzy rule base, and rebuild and solve the model again to ensure the feasibility of the strategy.

[0076] Based on the above technical solutions, the intelligent decision-making module, through data input verification, water and fertilizer demand prediction, and cost optimization strategy generation process, not only achieves dynamic and accurate prediction of water and fertilizer demand by relying on the planting prediction model, but also takes into account the economics of planting through cost optimization algorithms. The generated control strategy not only conforms to the growth law of Chinese medicinal materials, but also effectively reduces the overall operating cost, providing a scientific and reliable decision-making basis for the automated operation of the execution control module.

[0077] The execution control module is used to control the mixing of the two fertilizer solutions, the adaptive adjustment of the fertilization depth, and the adjustment of the irrigation flow rate based on the control strategy.

[0078] The dual-path fertilizer-liquid mixing is achieved through independent chemical fertilizer delivery pipelines and organic fertilizer delivery pipelines. The fertilizer injection depth is adaptively adjusted through an adjustable fertilizer injection mechanism. The irrigation flow rate is adjusted through a flow control device. All three work together to execute the control strategy generated by the intelligent decision-making module.

[0079] In some implementation methods, fertilizer solution mixing can be carried out using conventional mixing methods such as mechanical stirring and fluid mixing. The fertilizer injection mechanism can adjust the depth through drive components such as motors, hydraulic cylinders, and air cylinders. The irrigation flow rate can be controlled by equipment such as electric regulating valves and flow pumps. Feedback mechanisms can be set up for each control link to correct control parameters in real time.

[0080] It should be noted that the mixing of the two fertilizer solutions must ensure that the fertilizer concentration is uniform and stable, the fertilization depth must match the distribution of the medicinal herb roots, and the irrigation flow rate must be precisely adjusted according to the differences in soil moisture in the sub-regions.

[0081] Exemplary, the delivery flow of chemical fertilizer and organic fertilizer can be controlled by an electric regulating valve, and the mixed fertilizer is delivered to the fertilizer injection mechanism after mixing in the stirring tank; the fertilizer injection needle is driven to move up and down by a stepping motor to adjust the injection depth; the water flow rate in the irrigation pipeline is adjusted by an electromagnetic flow valve to realize precise control of the irrigation flow rate.

[0082] In a possible implementation of the embodiment of the application, in combination with Figure 1 As shown in Figure 4 The execution control module can be implemented by the following S301, S302 and S303, which are described in detail as follows: S301, based on the matching scheme in the control strategy, the precise mixing of the double-path fertilizer liquid is completed.

[0083] The core of step S301 is to deliver the fertilizer liquid in a preset ratio and realize uniform mixing through the independently controlled chemical fertilizer pipeline and organic fertilizer pipeline, so as to ensure that the concentration of the fertilizer liquid meets the growth requirements of the traditional Chinese medicinal materials.

[0084] In some implementations, the mixing of the double-path fertilizer liquid uses a double-path independent control system to regulate the delivery of chemical fertilizer and organic fertilizer, wherein: The chemical fertilizer pipeline adjusts the opening of the electric valve through a proportional-integral-derivative (PID) control algorithm, calculates the target flow rate according to the chemical fertilizer consumption in the control strategy, and corrects the valve opening in real time through the PID algorithm, with the formula being , is the proportional coefficient, is the integral coefficient, is the differential coefficient, and e(t) is the deviation between the actual flow rate and the target flow rate; The organic fertilizer pipeline integrates an ultrasonic crushing unit and a mechanical stirring unit, wherein the ultrasonic crushing unit decomposes large particle impurities in the organic fertilizer liquid through a vibration frequency of 20-40 kHz, and the mechanical stirrer generates turbulent flow in the mixing chamber at a speed of 300-500 r / min, thereby improving the uniformity and flowability of the organic fertilizer liquid and avoiding pipeline blockage.

[0085] The double-path fertilizer liquid is combined at the outlet of the mixing chamber, and the concentration of the fertilizer liquid is monitored in real time by an electric conductivity sensor during the mixing process to dynamically feedback and adjust the delivery speed of the two pipelines.

[0086] It should be noted that the cooperative design of ultrasonic crushing and mechanical stirring can effectively solve the technical problems of uneven mixing and easy blockage of high-viscosity organic fertilizer, and the real-time monitoring of the electric conductivity sensor ensures the stability of the mixed fertilizer liquid concentration.

[0087] S302, according to the control strategy and the growth state of the medicinal materials, the injection depth is self-adaptively adjusted.

[0088] The core of step S302 is to first calculate the ideal fertilization depth matched with the root system of medicinal materials, and then accurately adjust the actual fertilization depth to the ideal value through a closed-loop control algorithm, so as to ensure that the fertilizer reaches the root active layer.

[0089] In some implementations, the ideal fertilization depth D is first calculated according to the plant age T and the root distribution characteristics of medicinal materials, and the calculation formula is , is the minimum fertilization depth, which is set according to the root depth of the seedling stage of medicinal materials, is the maximum fertilization depth, which is set according to the root depth of the mature stage of medicinal materials, is the complete growth period of medicinal materials; Then, a fuzzy PID control algorithm is used to perform closed-loop control on the actual fertilization depth: taking the difference between the actual fertilization depth and the ideal fertilization depth and the soil hardness value as input variables, the input variables are fuzzily processed through a fuzzy rule base, the proportional, integral and differential coefficients of the PID controller are dynamically adjusted, the hydraulic cylinder control signal of the fertilization needle is output, and the fertilization needle is driven to move up and down to correct the depth deviation.

[0090] It should be noted that the fertilization needle and the irrigation system work synchronously, a single plant independent control mode is adopted, and the fertilization depth error can be controlled within ±2 cm, which can accurately adapt to the root distribution of medicinal materials at different growth stages.

[0091] S303, based on the soil moisture and environmental data of the sub-regions, the irrigation flow is accurately adjusted.

[0092] The core of step S303 is to dynamically adjust the irrigation flow according to the water consumption in the control strategy, combined with the real-time monitored soil moisture deviation and environmental factors, to ensure that the soil moisture of each sub-region reaches the target value.

[0093] In some implementations, the soil moisture content of each sub-region is first monitored in real time by a humidity sensor, and the deviation between the actual moisture content and the target humidity is calculated , is the target humidity, is the actual moisture content; then, combined with the evaporation E and rainfall R obtained by the environmental data acquisition unit, the irrigation flow is adjusted by using a proportional control strategy, and the control formula is , is the proportional control coefficient, is the environmental compensation coefficient; At the same time, the irrigation pipeline is provided with a pressure regulating device, which maintains the pressure in the pipeline stable through segmented pressure boosting technology, and calculates the pressure loss according to the terrain difference , is the density of water, g is the acceleration of gravity, h is the terrain height difference; and the pressure loss is supplemented by a pressure compensation valve, and the adjustable design of the dripper or the spray head is combined to realize accurate distribution of the flow.

[0094] It should be noted that this adjustment mode combines feedback control and feedforward compensation, which not only responds to real-time changes in soil humidity, but also anticipates environmental disturbances such as evaporation and rainfall, and solves the problem of uneven flow caused by terrain differences.

[0095] It should be noted that the traditional system has problems such as poor uniformity of fertilizer solution mixing, easy clogging of pipelines for high-viscosity organic fertilizer, fixed fertilization depth that cannot adapt to root growth, and unstable irrigation flow in complex terrains such as mountains. The double-path fertilizer solution mixing system of the present application controls the flow of chemical fertilizer through PID, processes organic fertilizer through an ultrasonic crushing unit and a mechanical stirring unit, and then monitors the concentration in real time through an electrical conductivity sensor, achieving uniform mixing of high-viscosity organic fertilizer without clogging; the fertilization depth self-adaptive adjustment calculates the ideal depth based on the characteristics of the medicinal material age and root distribution, and combines a fuzzy PID control algorithm to respond to soil hardness in real time, which can control the fertilization depth error within ±2cm, ensuring that the fertilizer reaches the root active layer; the irrigation flow control is achieved through humidity feedback and feedforward compensation of evaporation and rainfall, combined with a segmented pressure regulation device and a terrain height difference pressure loss compensation mechanism, and combined with an adjustable dripper / spray head design, solving the problem of uneven flow in complex terrains and achieving stable and accurate control of irrigation flow.

[0096] Based on the above technical solutions, the execution control module realizes the conversion of the control strategy of the intelligent decision module into automatic and refined operation actions through accurate mixing of double-path fertilizer solution, self-adaptive adjustment of fertilization depth, and dynamic calibration of irrigation flow. Among them, the double-path fertilizer solution mixing system solves the problems of uniform mixing and anti-clogging of high-viscosity organic fertilizer, the fertilization depth adjustment realizes the accurate matching of fertilizer and root active layer, and the irrigation flow control adapts to the terrain differences and environmental dynamic changes, which together improves the water and fertilizer utilization rate, reduces the labor intervention cost and environmental risk, and enhances the adaptability of the system to different Chinese herbal medicine varieties, different growth stages and complex planting environments, providing reliable execution guarantee for the improvement of the quality and yield of Chinese herbal medicines.

[0097] Some data in the above formula are calculated by removing the dimension and taking the numerical value, and the formula is obtained by software simulation of a large amount of collected data to obtain a formula closest to the real situation; the preset parameters and preset thresholds in the formula are set by a person skilled in the art according to the actual situation or obtained by a large amount of data simulation.

[0098] The working principle of the present application is as follows: Firstly, the data acquisition module as the foundation support, first combines the root system distribution characteristics of traditional Chinese medicinal materials and environmental data such as geography, weather, and divides the planting area into homogeneous sub-regions through a scientific regional division method, ensuring the pertinence of management; then through the multi-source sensor unit, the medicinal material growth monitoring unit and the environmental data acquisition unit, the soil core parameters, the medicinal material growth state data and the environmental data of each sub-region are synchronously collected, the collected data is stored and transmitted in time sequence form; then the original data is preprocessed such as denoising and dimensionality reduction, and the key feature vector is extracted, providing reliable data input for subsequent decision-making.

[0099] Then, the intelligent decision-making module receives the preprocessed data, constructs a planting prediction model relying on deep learning algorithm, extracts and fuses the multi-dimensional features of soil, plant and environment, dynamically adjusts the feature weight combining the variety and growth stage of traditional Chinese medicinal materials, accurately predicts the fertilizer demand and water demand in different periods, and avoids nutrient leaching or root damage through water and fertilizer synergistic optimization mechanism; on this basis, combining factors such as fertilizer price, medicinal material market price and inventory, the cost optimization algorithm is used to determine the replacement rate of organic fertilizer, a planning model with the goal of minimizing cost is constructed, and finally the optimization control strategy including fertilizer amount, water amount and organic fertilizer and chemical fertilizer ratio is generated.

[0100] Finally, the execution control module processes chemical fertilizer and organic fertilizer through a double-path independent control system respectively, realizes uniform mixing through a mixing device and monitors the concentration in real time; the fertilizer injection depth is dynamically adjusted according to the plant age and root system distribution, and the fertilizer injection position is accurately controlled combining soil conditions, to ensure that the fertilizer reaches the active layer of root system; at the same time, the soil humidity deviation, environmental evaporation and rainfall conditions are referred to, the terrain difference compensation pressure loss is combined, the irrigation flow is accurately adjusted, and the soil humidity of each sub-region is ensured to meet the standard. Through the collaborative work of each module, the automation, precision and economy of water and fertilizer application are realized, the diversified growth needs of traditional Chinese medicinal materials are effectively adapted, the water and fertilizer utilization rate is improved, and the environmental risk and labor cost are reduced, providing protection for medicinal material quality and yield.

[0101] The above embodiments are only used to illustrate the technical method of the present application and not to limit it, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical method of the present application.

Claims

1. A precision fertilization and water management system for traditional Chinese medicinal materials, characterized in that, include: The data acquisition module is used to divide the planting area into multiple sub-regions based on root distribution characteristics and environmental data, and to collect soil parameters, medicinal herb growth status and environmental data of each sub-region in real time. The intelligent decision-making module is used to process the collected data by the planting prediction model built based on deep learning algorithms, dynamically predict the fertilizer and water requirements of medicinal materials, and analyze the requirements according to the cost optimization algorithm to generate control strategies; wherein, the control strategies include fertilizer amount, water amount, and the ratio of organic fertilizer to chemical fertilizer. The execution control module is used to control the mixing of the two-channel fertilizer solution, the adaptive adjustment of the fertilization depth, and the adjustment of the irrigation flow rate based on the control strategy.

2. The precision fertilization and water management system for traditional Chinese medicinal materials according to claim 1, characterized in that, The planting area is divided into multiple sub-regions based on root distribution characteristics and environmental data, including: The sampling depth is determined by the root distribution characteristics, and the sampling density is determined by the grid method in combination with environmental data; wherein, the environmental data includes slope, aspect, and altitude from the geographical environment data, and rainfall and evaporation from the weather data; Soil parameters and topographic indices were collected based on sampling density and sampling depth, and a continuous distribution map was generated using the Kriging spatial interpolation method. The K-means clustering algorithm was used to divide the termination region into multiple sub-regions; the feature vectors of the clustering included the comprehensive soil nutrient index, root depth coefficient and topographic factor. The comprehensive soil nutrient index is calculated by weighting the concentrations of nitrogen, phosphorus, potassium (NPK) in the soil parameters. The root depth coefficient is the normalized ratio of the actual root depth of the medicinal herb to the standard root depth. The topographic factor is represented by the first principal component obtained by dimensionality reduction of the slope, aspect, and altitude in the geographical environment data through principal component analysis.

3. The precision fertilization and water management system for traditional Chinese medicinal materials according to claim 1, characterized in that, The data preprocessing process of the data acquisition module includes: The raw data acquired by the multi-source sensor unit and the environmental data acquisition unit are denoised using the Kalman filter algorithm to obtain the first preprocessed data. Principal component analysis was used to reduce the dimensionality of the first preprocessed data and generate soil feature vectors and environmental feature vectors. The soil feature vectors include soil moisture, soil temperature, NPK concentration, pH value, and electrical conductivity. The environmental feature vectors include rainfall, evaporation, temperature, and light intensity.

4. The precision fertilization and water management system for traditional Chinese medicinal materials according to claim 1, characterized in that, The planting prediction model based on deep learning algorithms includes a feature extraction module, a cross-dimensional fusion module, and a prediction output module; wherein, The feature extraction module includes a soil feature submodule, an image feature submodule, and an environmental feature submodule, which are used to extract features from the input soil feature vector, multispectral image data, and environmental feature vector, respectively, and output soil time-series feature vector, image time-series feature vector, and environmental time-series feature vector; The cross-dimensional fusion module includes a feature alignment unit, a dynamic weight calculation unit, and a fusion unit. The input consists of temporal feature vectors of each dimension, Chinese medicinal herb variety labels, and growth stage labels. The module processes these inputs to balance the contribution of multi-dimensional information and outputs an optimized multi-dimensional fused feature vector. The prediction output module includes a time-series encoding unit and a dual-task decoding unit. The inputs are the multi-dimensional fusion feature vector, the water and fertilizer threshold for medicinal herb cultivation, and the field water holding capacity parameter. It is used to perform multi-step water and fertilizer demand prediction and output the daily fertilizer application amount and water consumption.

5. The precision fertilization and water management system for traditional Chinese medicinal materials according to claim 4, characterized in that, The workflow of the feature extraction module includes: The input to the soil feature submodule is a soil feature vector. The first soil feature is extracted from the soil parameters through the backbone network. Then, the first soil feature is weighted by the channel attention mechanism to obtain the weighted second soil feature. The second soil feature is then input into the adversarial training unit and the fully connected layer to obtain the soil temporal feature vector. The adversarial training unit is a dual-branch structure of "generator-discriminator". Both the generator and the discriminator are built based on the multilayer perceptron. The generator is used to generate noisy soil features, and the discriminator is used to distinguish the second soil feature from the noisy soil feature. The input to the image feature submodule is multispectral image data. The first image feature is extracted through the backbone network. Then, the first image feature is input into the Visual Transformer module to extract global features. At the same time, the first image feature is input into the band attention mechanism for adaptive enhancement of local features. Finally, the global features and local features are concatenated to obtain the image temporal feature vector. The input to the environmental feature submodule is an environmental feature vector. The first environmental feature is extracted through the backbone network. Then, the first environmental feature is input into a multi-layer dilated convolution and a multi-scale attention mechanism. After optimization by activation function and Dropout layer, the environmental temporal feature vector is obtained.

6. The precision fertilization and water management system for traditional Chinese medicinal materials according to claim 4, characterized in that, The workflow of the cross-dimensional fusion module includes: The feature alignment unit first unifies the vector dimensions of the input temporal feature vectors of each dimension through a fully connected layer, and then performs Z-score standardization and L2 normalization to output multi-dimensional standardized feature vectors with aligned dimensions and consistent distribution. The input to the dynamic weight calculation unit is the Chinese medicinal herb variety label and growth stage label. It assigns initial weights to the multi-dimensional standardized feature vectors according to the built-in benchmark weight library, and then adaptively adjusts the initial weights according to the feature confidence of the multi-dimensional data, and outputs the dynamic weight coefficients of each dimension. The benchmark weight library includes the weights of Chinese medicinal herb varieties, growth stages, and features of each dimension. The input to the fusion unit is a multi-dimensional standardized feature vector and dynamic modal weight coefficients. The multi-dimensional standardized feature vector is weighted and summed based on the weight coefficients to obtain a multi-dimensional fused feature vector.

7. The precision fertilization and water management system for traditional Chinese medicinal materials according to claim 4, characterized in that, The workflow of the prediction output module includes: The input to the temporal coding unit is a multi-dimensional fused feature vector. A bidirectional gated recurrent unit is used to temporally encode the fused feature vector and output temporally encoded features. The input to the dual-task decoding is the temporal coding features, the water and fertilizer threshold for medicinal herb cultivation, and the field water holding capacity parameter. A dual output head is constructed based on the LSTM module, and a water and fertilizer coupling constraint is introduced to map the temporal coding features to the predicted water and fertilizer values ​​for a preset future time period, thereby obtaining the daily fertilizer and water requirements. The water and fertilizer coupling constraint refers to the collaborative optimization and correction of the original predicted values ​​output by the dual output heads based on the water and fertilizer threshold and the field water holding capacity parameter.

8. A precision fertilization and water management system for traditional Chinese medicinal materials according to claim 7, characterized in that, The water-fertilizer coupling constraint is based on the calculation of the theoretical water-fertilizer ratio using the water-fertilizer coupling equation. The calculation formula is: ;in, For irrigation threshold, This is the fertilization threshold. This refers to field water holding capacity parameters; The fertilizer requirement and water demand Based on the theoretical water-fertilizer ratio calculation, the formula is as follows: , ;in, This is the raw water usage prediction value output by the dual-output head. This is the raw fertilization prediction value output by the dual-output head.

9. A precision fertilization and water management system for traditional Chinese medicinal materials according to claim 1, characterized in that, The step of analyzing demand based on a cost optimization algorithm and generating a control strategy includes: A fuzzy rule base is constructed, and the organic fertilizer substitution rate is obtained based on the fuzzy set of input variables and rule logic; wherein, the input variables include fertilizer price, medicinal herb market price and current inventory, and the output variable is the organic fertilizer substitution rate; A linear programming model is established based on the organic fertilizer substitution rate, the predicted fertilizer demand, and the water demand; the linear programming model takes minimizing the total cost of fertilizer and water as the objective function. Solve the linear programming model and generate a control strategy based on the calculated fertilizer application rate, water consumption, and the ratio of organic fertilizer to chemical fertilizer.

10. A precision fertilization and water management system for traditional Chinese medicinal materials according to claim 1, characterized in that, The workflow of the adaptive adjustment of fertilizer injection depth includes: The ideal fertilization depth D is calculated based on the plant age T and root distribution characteristics of the medicinal herb. The calculation formula is as follows: ;in, Indicates the minimum injection depth. Indicates the maximum fertilizer injection depth. Indicates the complete growth period of medicinal materials; A fuzzy PID control algorithm is used to control the actual fertilization depth. The inputs of the fuzzy PID control algorithm are the depth deviation and the soil hardness value, and the output is the hydraulic cylinder control signal of the fertilization needle. The depth deviation is the difference between the actual fertilization depth and the ideal fertilization depth.

Images