Cultivation method of safflower and cumin interplanting in sandy land

Abstract

The application relates to a cultivation method of sand land medicinal rose interplanting cumin, and belongs to the technical field of intelligent agriculture. The method fuses internet of things sensing, big data analysis and artificial intelligence decision, and constructs a closed loop system of "data acquisition-model construction-intelligent decision-precise execution-feedback optimization". Through laying soil-weather-remote sensing sensor networks, a rose-cumin competition-complementary coupling model is established, particle swarm algorithm is adopted to dynamically optimize planting layout, sowing period, water and fertilizer schemes, and precise regulation is realized. The system has knowledge self-learning ability, and model parameters are iteratively updated every season. The application solves the problems of traditional interplanting relying on experience, fixed parameters and being unable to adapt to environmental variation, improves land equivalent ratio by 0.6-0.8, improves water and fertilizer utilization rate by more than 35%, improves decision accuracy by more than 5% per year, and provides a digital solution for efficient ecological planting on sandy land.

Description

Technical Field

[0001] This invention relates to the field of smart agriculture and agricultural information technology, and in particular to a cultivation method for intercropping cumin with medicinal roses in sandy soil. Background Technology

[0002] Current methods for intercropping cumin with medicinal roses in sandy soils rely heavily on traditional agronomic experience, revealing significant technical bottlenecks in long-term production. These methods typically employ fixed row spacing, static sowing dates, and extensive water and fertilizer management, failing to establish dynamic monitoring and quantitative control mechanisms for crop growth. This results in poor environmental adaptability, low resource utilization efficiency, and unstable economic benefits, making it difficult to meet the precision and intelligent development needs of modern agriculture. Specifically, traditional methods lack scientific decision-making tools when addressing core issues such as the weak water and fertilizer retention capacity of sandy soils, large interannual climate fluctuations, and complex intercropping competition. A fundamental shift from experience-driven to data-driven approaches is urgently needed through technological innovation.

[0003] Existing technologies suffer from three major flaws: First, parameter settings rely on empirical methods. Key decisions such as planting density, row spacing, and staggered sowing dates depend on static historical experience, failing to respond in real time to interannual climate anomalies (such as late spring frosts and extreme summer temperatures). This results in an interannual coefficient of variation (CV) for yield exceeding 25%, indicating severe instability of the model. Second, resource competition regulation is qualitative. The competition for light, water, and fertilizer between roses (deep-rooted perennials) and cumin (shallow-rooted annuals) lacks dynamic quantitative methods. The failure to establish a competition coefficient model based on real-time monitoring data limits the release of interspecific synergistic effects, resulting in land equivalent ratios (LER) generally below 1.4 and the intercropping potential far from being fully realized. Third, water and soil management is inefficient. Existing integrated water and fertilizer systems lack a real-time feedback loop between crops, soil, and the environment. Irrigation uniformity is less than 70%, and fertilizer utilization is below 30%, not only wasting resources but also exacerbating the ecological vulnerability of sandy soils.

[0004] While smart agriculture technologies offer new pathways for precision intercropping, existing solutions still have significant limitations. Although monoculture optimization systems based on crop growth models exist both domestically and internationally, research on coupled models for interspecies intercropping systems like rose-cumin is severely lacking. Machine learning algorithms are widely used in agricultural forecasting, but there is a lack of dedicated algorithmic frameworks applicable to the specific physiological and ecological characteristics of medicinal plant-spice crops. An intelligent intercropping decision-making system integrating competitive ecology and data science has not yet been established in this field, failing to address the fundamental shortcomings of traditional methods in dynamic optimization, precise control, and self-learning evolution. Therefore, constructing a smart cultivation method for intercropping medicinal rose and cumin in sandy soil, capable of real-time sensing, intelligent decision-making, and precise execution, has become an urgent need to overcome industrial technology bottlenecks and improve the sustainability of sandy agriculture. Summary of the Invention

[0005] The main objective of this invention is to overcome the shortcomings of existing cultivation methods for intercropping medicinal roses with cumin in sandy soil, which rely on farmers' experience, cannot quantify interspecific competition, and have extensive planting layout decisions lacking dynamic optimization capabilities. Instead, this invention provides a new cultivation method for intercropping medicinal roses with cumin in sandy soil. The technical problem to be solved is to enable the intelligent generation and dynamic optimization of planting parameters through multi-source environmental data acquisition and crop growth coupling modeling, thereby making it more practical and having industrial application value.

[0006] Another objective of this invention is to provide a cultivation method for intercropping cumin with medicinal roses in sandy soil. The technical problem to be solved is to construct an intelligent decision engine based on particle swarm optimization algorithm, which combines historical big data and real-time monitoring data to output the optimal narrow row to wide row ratio, sowing density and staggered planting days. This engine can be replaced by genetic algorithm or simulated annealing algorithm, making it more suitable for practical use.

[0007] Another objective of this invention is to provide a cultivation method for intercropping cumin with medicinal roses in sandy soil. The technical problem to be solved is to enable it to form a closed-loop control system through an Internet of Things drip irrigation system and drone remote sensing, which automatically triggers the timing and amount of water and fertilizer application based on the soil matrix potential threshold and the real-time value of the leaf area index, thus making it more suitable for practical use.

[0008] Another objective of this invention is to provide a cultivation method for intercropping cumin with medicinal roses in sandy soil. The technical problem to be solved is to establish a cloud-edge collaborative knowledge base self-learning and updating mechanism, which feeds back the actual yield, quality, and cost data of each season to the long short-term memory neural network for incremental training, thereby reducing the decision error rate year by year and making it more suitable for practical use.

[0009] The objective of this invention and the technical problem it solves are achieved by the following technical solutions. According to the present invention, a smart cultivation method for intercropping medicinal roses with cumin in sandy soil is characterized by the following five collaborative steps: (1) collecting multi-source environmental data of at least the first 3-5 complete growing seasons of the target plot, covering soil moisture, micro-meteorological information and crop growth image information; (2) constructing a growth coupling model of the rose-cumin intercropping system, which uses land equivalent ratio, water use efficiency and nutrient competition coefficient as multi-objective optimization functions; (3) dynamically generating the optimal planting layout scheme through intelligent algorithms based on historical data and real-time monitoring data, and outputting the narrow row to wide row ratio, sowing density and staggered planting days; (4) performing precise planting and dynamic water and fertilizer synergistic regulation based on the decision results, and automatically adjusting the application time and amount according to the growth stage model through the Internet of Things drip irrigation system; (5) feeding back the actual production data to the cloud platform, and achieving self-learning update through incremental learning to optimize model parameters; wherein the intelligent algorithm includes multi-objective optimization algorithms such as particle swarm optimization algorithm, genetic algorithm or simulated annealing algorithm, and the expert rule engine is activated when the historical data coverage is less than 30%, and the data-driven model is switched when it reaches more than 30%.

[0010] The objectives of this invention and the solutions to its technical problems can be further achieved by the following technical measures. In the aforementioned cultivation method of intercropping cumin with medicinal roses in sandy soil, the multi-source environmental data acquisition is completed by deploying soil moisture sensors, meteorological monitoring stations, and UAV remote sensing systems. The soil data sampling density is one point per 0.5-1 acre with stratified depth. The meteorological data acquisition frequency is no less than once per hour. Crop growth data is obtained through UAV multispectral remote sensing inversion with a resolution ≤5cm. The input layer feature vector of the crop growth coupling model includes at least 15 parameters such as accumulated temperature, soil moisture content, organic matter content, and expected market price. The output layer decision parameters include a wide row spacing of 120-150cm and a narrow row spacing of 30-4... The sowing period is staggered by 5-15 days. The dynamic water and fertilizer synergistic regulation involves burying double-layer soil moisture sensors at depths of 20cm and 40cm at 15cm intervals in the center of the wide rows of roses and on both sides of the narrow rows of cumin. Irrigation is triggered when the substrate potential is below -30kPa. Fertilization decisions are based on the real-time leaf area index value retrieved from UAV multispectral images. The decision tree model determines the growth stage and automatically matches the nitrogen, phosphorus, and potassium formula. The knowledge base self-learning update adopts a long short-term memory neural network with 64 hidden layer nodes. Through transfer learning, the weights of the bottom layer are frozen and only the top layer is finely adjusted. Data is uploaded within 7 days after each harvest to trigger model retraining, thereby reducing the decision error rate year by year.

[0011] The objectives of this invention and the solutions to its technical problems are further achieved using the following technical solutions. According to this invention, a smart agricultural management system for implementing the above-mentioned method is proposed, characterized by comprising: a perception layer, consisting of soil sensors, a weather station, a drone, and a high-definition camera, with a data acquisition frequency ≥ 1 time / hour; a transmission layer, based on a LoRaWAN or NB-IoT wireless transmission network, with a data packet loss rate < 5%; a decision layer, deploying big data analysis and AI models on an edge computing gateway or cloud server, with a response latency < 30 seconds; and an execution layer, comprising an intelligent water and fertilizer integrated machine, an automated seeder, and a variable fertilization device, with a control accuracy ≥ 95%. The decision layer includes a dual-mode decision engine comprising an expert knowledge base and a machine learning model, and a manual review mechanism is activated when the difference between the output results of the two modes exceeds 15%.

[0012] The objectives of this invention and the solutions to its technical problems can be further achieved by the following technical measures. In the aforementioned smart agricultural management system, the execution layer fertigation machine has a three-channel mother liquor dispensing system that stores concentrated nitrogen, phosphorus, and potassium solutions. It uses PID control for real-time fertilizer application, and EC / pH is monitored and adjusted online with accuracies of ±0.1 mS / cm and ±0.2 mS / cm, respectively. The minimum control unit area for plot grid management is ≤0.5 mu. The UAV remote sensing system performs topographic mapping before sowing to generate a high-precision DEM to assist in ridging operations with an accuracy of ±2 cm. During the growing season, it acquires multispectral images weekly to calculate NDVI and GNDVI vegetation indices to monitor growth differences. Later, it identifies pod maturity and sends a harvest warning when the proportion of yellow pixels in the image is ≥70%.

[0013] In summary, this invention presents a unique cultivation method for intercropping cumin with medicinal roses in sandy soil. This method aims to address the technical challenges of sandy soil, including poor water and fertilizer retention, low resource utilization, significant competition from other crops, reliance on experience in cultivation management, and insufficient market risk control. It achieves efficient utilization of sandy soil resources and synergistic high-quality and high-yield cultivation of medicinal roses and cumin. Furthermore, intelligent decision-making and dynamic control improve cultivation precision and reduce production risks, balancing ecological adaptability with industrial economic benefits. This method possesses numerous advantages and practical value, and is truly innovative as no similar design has been publicly disclosed or used in the field. It represents a significant improvement in both method and function, demonstrating substantial technological advancement and producing user-friendly and practical results. Compared to existing methods for intercropping cumin with medicinal roses in sandy soil, it offers several enhanced benefits, making it more suitable for practical application and possessing broad industrial value. It is indeed a novel, progressive, and practical design.

[0014] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, the preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0015] The specific methods and structures of the present invention are given in detail in the following embodiments and accompanying drawings. Attached Figure Description

[0016] Figure 1 : Schematic diagram of intercropping medicinal roses in sandy areas;

[0017] In the picture: 1. Roses with a wide row spacing of 135cm, 2. Roses with a wide row spacing, 3. Roses with a narrow row spacing of 35cm, 4. Plant spacing of 90cm, 5. Cumin inoculation with a row spacing of 18cm. Detailed Implementation

[0018] To further illustrate the technical means and effects of the present invention in achieving the intended purpose, the following detailed description, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation methods, steps, structures, characteristics, and effects of the cultivation method for intercropping cumin with medicinal roses in sandy soil according to the present invention.

[0019] Please see Figure 1 The preferred embodiment of the present invention describes a cultivation method for intercropping cumin with medicinal roses in sandy soil, which mainly includes the following steps:

[0020] Step 1: Basic Survey and Data Collection of the Target Site. A sandy soil plot in Northwest China (soil organic matter content 0.5%-1.2%, pH 7.5-8.5) was selected as the experimental site. Multi-source data collection was completed using sensing layer equipment: soil sensors (20m×20m density) were used to collect data on humidity, temperature, and nitrogen, phosphorus, and potassium content in the 0-20cm and 20-40cm soil layers, with a collection frequency of once per hour; meteorological data such as accumulated temperature, precipitation, and wind speed were collected using a small weather station; simultaneously, data on successful intercropping cases of medicinal rose and cumin in the region over the past 5 years (60 sets in total, including fields such as yield, quality, and planting parameters) were collected as the basic data for model construction.

[0021] Step 2: Construction and Training Optimization of the Growth Coupling Model. Based on the 60 sets of case data collected in Step 1, the random forest algorithm was used to select five key influencing factors: accumulated temperature (T), moisture (W), planting density (D), and canopy closure (H). The relative growth rate competition equations for rose and cumin were established: RGR_rose = f(T,W,N,D,H)×(1-α×D_cumin) and RGR_cumin = g(T,W,P,D,L)×(1-β×H_rose). The initial values ​​of the competition coefficients α and β were set to 0.3-0.5 and 0.2-0.4, respectively. Through nonlinear least squares fitting optimization, α = 0.38 and β = 0.26 were finally determined, and the model prediction error was controlled within 8%.

[0022] Step 3: Intelligent Decision-Making Scheme Generation. Input the current year's weather forecast data (accuracy ≥ 90%), soil measurement data, and expected market prices for roses and cumin. Call the historical similar year dataset (selecting years with accumulated temperature deviation ±5% and precipitation deviation ±10%) to extract the optimal planting pattern; run the multi-objective particle swarm optimization algorithm (iteration count set to 100 times, convergence accuracy ≤ 0.01), and finally output the optimal parameter combination: roses with a wide row spacing of 140cm, a narrow row spacing of 35cm, and a plant spacing of 90cm; cumin sowing row spacing of 18cm; and a 10-day staggered replanting period for roses and cumin; simultaneously access futures market data, calculate the market volatility to be 15%, and select a balanced planting scheme.

[0023] Step 4: Precision Planting Implementation. Based on the decision results of Step 3, an automated seeder was used for planting: first, medicinal rose seedlings (15-20cm tall, survival rate ≥95%) were planted in the wide-row-spacing area; 10 days later, cumin was sown in the narrow-row-spacing area (sowing depth 2-3cm, sowing rate 1.5kg / mu); after planting, the intelligent water and fertilizer integrated machine was started to complete the initial irrigation (matrix potential adjusted to -20kPa) and base fertilizer application (nitrogen, phosphorus, and potassium ratio 3:1:2).

[0024] Step 5: Dynamic regulation throughout the entire growth period. Soil moisture sensors monitor matrix potential in real time; irrigation (drip irrigation, 20m³) is triggered when the value falls below -30kPa. 3 / mu); weekly images are acquired via drones (equipped with 5-band multispectral cameras), leaf area index (LAI) is retrieved (using the NDVI-LAI model, accuracy ≥92%), and growth stage is determined using a decision tree model: nitrogen fertilizer (concentration 200mg / L) is applied during the vegetative growth stage, and phosphorus and potassium fertilizer (concentration 300mg / L) is applied during the flowering stage; in the later stage, pod maturity is identified through drone images, and a harvest warning is sent when the proportion of yellow pixels is ≥70%.

[0025] Step 6: Data Feedback and Model Self-Learning Update. After each harvest, record the measured yield (rose flower yield ≥ 800 kg / mu, cumin yield ≥ 120 kg / mu), quality indicators (rose essential oil content ≥ 0.03%, cumin volatile oil content ≥ 4.5%), and input costs, and upload the data to the system. Using an LSTM neural network (64 hidden layer nodes, learning rate 0.001), input the data from seasons t-3, t-2, and t-1, predict the optimal parameters for season t, and automatically trigger model retraining to reduce the decision error rate by ≥ 5% compared to the previous season.

[0026] Example 1:

[0027] Step 1: Basic Survey and Data Collection of the Target Site. A sandy soil plot in Northwest China (soil organic matter content 0.5%, pH 7.5) was selected as the experimental site. Multi-source data collection was completed using sensing layer equipment: soil sensors (20m×20m density) were used to collect data on humidity, temperature, and nitrogen, phosphorus, and potassium content in the 0-20cm and 20-40cm soil layers, with a collection frequency of once per hour; meteorological data such as accumulated temperature, precipitation, and wind speed were collected using a small weather station; simultaneously, data on successful intercropping cases of medicinal rose and cumin in the region over the past 5 years (a total of 50 sets, including fields such as yield, quality, and planting parameters) were collected as the basic data for model construction.

[0028] Step 2: Construction and Training Optimization of the Growth Coupling Model. Based on the 50 sets of case data collected in Step 1, the random forest algorithm was used to select five key influencing factors: accumulated temperature (T), moisture (W), planting density (D), and canopy closure (H). The relative growth rate competition equations for rose and cumin were established: RGR_rose = f(T,W,N,D,H)×(1-α×D_cumin) and RGR_cumin = g(T,W,P,D,L)×(1-β×H_rose). The initial values ​​of the competition coefficients α and β were set to 0.3 and β to 0.2, respectively. The model was optimized using nonlinear least squares fitting, ultimately determining α = 0.32 and β = 0.21, with the prediction error controlled within 8%.

[0029] Step 3: Intelligent Decision-Making Scheme Generation. Input the current year's weather forecast data (accuracy ≥ 90%), soil measurement data, and expected market prices for roses and cumin. Call the historical similar year dataset (selecting years with accumulated temperature deviation ±5% and precipitation deviation ±10%) to extract the optimal planting pattern; run the multi-objective particle swarm optimization algorithm (iteration count set to 80 times, convergence accuracy ≤ 0.01), and finally output the optimal parameter combination: roses with a wide row spacing of 120cm, a narrow row spacing of 30cm, and a plant spacing of 80cm; cumin sowing row spacing of 15cm; and a 5-day stagger between the replanting periods of roses and cumin; simultaneously access futures market data, calculate the market volatility to 25%, and select a conservative planting scheme (reducing the planting density of both roses and cumin by 10%).

[0030] Step 4: Precision Planting Implementation. Based on the decision results of Step 3, an automated seeder was used for planting: first, medicinal rose seedlings (15cm tall, survival rate ≥95%) were planted in the wide-row-spacing area; cumin was sown in the narrow-row-spacing area 5 days later (sowing depth 2cm, sowing rate 1.2kg / mu); after planting, the intelligent water and fertilizer integrated machine was started to complete the initial irrigation (matrix potential adjusted to -20kPa) and base fertilizer application (nitrogen, phosphorus and potassium ratio 3:1:2).

[0031] Step 5: Dynamic regulation throughout the entire growth period. Soil moisture sensors monitor matrix potential in real time; irrigation (drip irrigation, irrigation volume 18m³) is triggered when the value falls below -30kPa. 3 / mu); weekly images are acquired via drones (equipped with 4-band multispectral cameras), leaf area index (LAI) is retrieved (using the NDVI-LAI model, accuracy ≥90%), and growth stage is determined using a decision tree model: nitrogen fertilizer (concentration 180mg / L) is applied during the vegetative growth stage, and phosphorus and potassium fertilizer (concentration 280mg / L) is applied during the flowering stage; in the later stage, pod maturity is identified through drone images, and a harvest warning is sent when the proportion of yellow pixels is ≥70%.

[0032] Step 6: Data Feedback and Model Self-Learning Update. After each harvest, record the measured yield (rose flower yield ≥750kg / mu, cumin yield ≥100kg / mu), quality indicators (rose essential oil content ≥0.028%, cumin volatile oil content ≥4.2%), and input costs, and upload the data to the system. Using an LSTM neural network (64 hidden layer nodes, learning rate 0.001), input the data from seasons t-3, t-2, and t-1, predict the optimal parameters for season t, and automatically trigger model retraining to reduce the decision error rate by ≥5% compared to the previous season.

[0033] Example 2:

[0034] Step 1: Basic Survey and Data Collection of the Target Site. A sandy soil plot in Northwest China (soil organic matter content 0.8%, pH 8.0) was selected as the experimental site. Multi-source data collection was completed using sensing layer equipment: soil sensors (20m×20m density) were used to collect data on humidity, temperature, and nitrogen, phosphorus, and potassium content in the 0-20cm and 20-40cm soil layers, with a collection frequency of once per hour; meteorological data such as accumulated temperature, precipitation, and wind speed were collected using a small weather station; simultaneously, data on successful intercropping cases of medicinal rose and cumin in the region over the past 5 years (60 sets in total, including fields such as yield, quality, and planting parameters) were collected as the basic data for model construction.

[0035] Step 2: Construction and Training Optimization of the Growth Coupling Model. Based on the 60 sets of case data collected in Step 1, the random forest algorithm was used to select five key influencing factors: accumulated temperature (T), moisture (W), planting density (D), and canopy closure (H). The relative growth rate competition equations for rose and cumin were established: RGR_rose = f(T,W,N,D,H)×(1-α×D_cumin) and RGR_cumin = g(T,W,P,D,L)×(1-β×H_rose). The initial values ​​of the competition coefficients α and β were set to 0.4 and 0.3, respectively. Through nonlinear least squares fitting optimization, α = 0.38 and β = 0.26 were finally determined, and the model prediction error was controlled within 7%.

[0036] Step 3: Intelligent Decision-Making Scheme Generation. Input the current year's weather forecast data (accuracy ≥ 90%), soil measurement data, and expected market prices for roses and cumin. Call the historical similar year dataset (selecting years with accumulated temperature deviation ±5% and precipitation deviation ±10%) to extract the optimal planting pattern; run the multi-objective particle swarm optimization algorithm (iteration count set to 100 times, convergence accuracy ≤ 0.01), and finally output the optimal parameter combination: roses with a wide row spacing of 135cm, a narrow row spacing of 35cm, and a plant spacing of 90cm; cumin sowing row spacing of 18cm; and a 10-day staggered replanting period for roses and cumin; simultaneously access futures market data, calculate the market volatility to be 15%, and select a balanced planting scheme.

[0037] Step 4: Precision Planting Implementation. Based on the decision results of Step 3, an automated seeder was used for planting: first, medicinal rose seedlings (18cm tall, survival rate ≥95%) were planted in the wide-row-spacing area; cumin was sown in the narrow-row-spacing area 10 days later (sowing depth 2.5cm, sowing rate 1.5kg / mu); after planting, the intelligent water and fertilizer integrated machine was started to complete the initial irrigation (matrix potential adjusted to -20kPa) and base fertilizer application (nitrogen, phosphorus, and potassium ratio 3:1:2).

[0038] Step 5: Dynamic regulation throughout the entire growth period. Soil moisture sensors monitor matrix potential in real time; irrigation (drip irrigation, 20m³) is triggered when the value falls below -30kPa. 3 / mu); weekly images are acquired via drones (equipped with 5-band multispectral cameras), leaf area index (LAI) is retrieved (using the NDVI-LAI model, accuracy ≥92%), and growth stage is determined using a decision tree model: nitrogen fertilizer (concentration 200mg / L) is applied during the vegetative growth stage, and phosphorus and potassium fertilizer (concentration 300mg / L) is applied during the flowering stage; in the later stage, pod maturity is identified through drone images, and a harvest warning is sent when the proportion of yellow pixels is ≥70%.

[0039] Step 6: Data Feedback and Model Self-Learning Update. After each harvest, record the measured yield (rose flower yield ≥ 800 kg / mu, cumin yield ≥ 120 kg / mu), quality indicators (rose essential oil content ≥ 0.03%, cumin volatile oil content ≥ 4.5%), and input costs, and upload the data to the system. Using an LSTM neural network (64 hidden layer nodes, learning rate 0.001), input the data from seasons t-3, t-2, and t-1, predict the optimal parameters for season t, and automatically trigger model retraining to reduce the decision error rate by ≥ 5% compared to the previous season.

[0040] Example 3:

[0041] Step 1: Basic Survey and Data Collection of the Target Site. A sandy soil plot in Northwest China (soil organic matter content 1.2%, pH 8.5) was selected as the experimental site. Multi-source data collection was completed using sensing layer equipment: soil sensors (20m×20m density) were used to collect data on humidity, temperature, and nitrogen, phosphorus, and potassium content in the 0-20cm and 20-40cm soil layers, with a collection frequency of once per hour; meteorological data such as accumulated temperature, precipitation, and wind speed were collected using a small weather station; simultaneously, data on successful intercropping cases of medicinal rose and cumin in the region over the past 5 years (70 sets in total, including fields such as yield, quality, and planting parameters) were collected as the basic data for model construction.

[0042] Step 2: Construction and Training Optimization of the Growth Coupling Model. Based on the 70 sets of case data collected in Step 1, the random forest algorithm was used to select five key influencing factors: accumulated temperature (T), moisture (W), planting density (D), and canopy closure (H). The relative growth rate competition equations for rose and cumin were established: RGR_rose = f(T,W,N,D,H)×(1-α×D_cumin) and RGR_cumin = g(T,W,P,D,L)×(1-β×H_rose). The initial values ​​of the competition coefficients α and β were set to 0.5 and β to 0.4, respectively. Through nonlinear least squares fitting optimization, α = 0.45 and β = 0.32 were finally determined, and the model prediction error was controlled within 6%.

[0043] Step 3: Intelligent Decision-Making Scheme Generation. Input the current year's weather forecast data (accuracy ≥ 90%), soil measurement data, and expected market prices for roses and cumin. Call the historical similar year dataset (selecting years with accumulated temperature deviation ±5% and precipitation deviation ±10%) to extract the optimal planting pattern; run the multi-objective particle swarm optimization algorithm (iteration count set to 120 times, convergence accuracy ≤ 0.01), and finally output the optimal parameter combination: roses with a wide row spacing of 150cm, a narrow row spacing of 40cm, and a plant spacing of 100cm; cumin sowing row spacing of 20cm; and a 15-day stagger between the replanting periods of roses and cumin. Simultaneously access futures market data, calculate the market volatility to be 18%, and select an aggressive planting scheme (increasing the planting density of both roses and cumin by 10%).

[0044] Step 4: Precision Planting Implementation. Based on the decision results of Step 3, an automated seeder was used for planting: first, medicinal rose seedlings (20cm tall, survival rate ≥95%) were planted in the wide-row-spacing area; cumin was sown in the narrow-row-spacing area 15 days later (sowing depth 3cm, sowing rate 1.8kg / mu); after planting, the intelligent water and fertilizer integrated machine was started to complete the initial irrigation (matrix potential adjusted to -20kPa) and base fertilizer application (nitrogen, phosphorus and potassium ratio 3:1:2).

[0045] Step 5: Dynamic regulation throughout the entire growth period. Soil moisture sensors monitor matrix potential in real time; irrigation (drip irrigation, irrigation volume 22m³) is triggered when the value falls below -30kPa. 3 / mu); weekly images are acquired via drones (equipped with 6-band multispectral cameras), leaf area index (LAI) is retrieved (using the NDVI-LAI model, accuracy ≥93%), and growth stage is determined using a decision tree model: nitrogen fertilizer (concentration 220mg / L) is applied during the vegetative growth stage, and phosphorus and potassium fertilizer (concentration 320mg / L) is applied during the flowering stage; in the later stage, pod maturity is identified through drone images, and a harvest warning is sent when the proportion of yellow pixels is ≥70%.

[0046] Step 6: Data Feedback and Model Self-Learning Update. After each harvest, record the measured yield (rose flower yield ≥ 850 kg / mu, cumin yield ≥ 140 kg / mu), quality indicators (rose essential oil content ≥ 0.032%, cumin volatile oil content ≥ 4.8%), and input costs, and upload the data to the system. Using an LSTM neural network (64 hidden layer nodes, learning rate 0.001), input the data from seasons t-3, t-2, and t-1, predict the optimal parameters for season t, and automatically trigger model retraining to reduce the decision error rate by ≥ 5% compared to the previous season.

[0047] Table 1 shows the experimental data for the above embodiments.

[0048] Table 1: Experimental Data Table for Example 1

[0049]

[0050]

[0051] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the methods and techniques disclosed above without departing from the scope of the present invention to create equivalent embodiments. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A smart cultivation method for intercropping cumin with medicinal roses in sandy soil, characterized in that, Includes the following steps: Collect multi-source environmental data and crop growth information of the target site; A growth coupling model for the rose-cumin intercropping system was constructed and trained and optimized based on historical data. By combining real-time environmental and predictive data, the optimal planting layout and water and fertilizer plan are dynamically generated through intelligent algorithms. Precision planting and dynamic regulation are implemented based on the decision-making results; Actual production data is fed back into the model to enable self-learning and updates.

2. The method according to claim 1, characterized in that, The method for constructing the crop growth coupling model in step (2) is as follows: Collect data from ≥50 successful intercropping cases in the target area, use the random forest algorithm to screen out key factors affecting yield, and establish a competition equation for the relative growth rates of rose and cumin: RGR_rose = f(T,W,N,D,H)×(1-α×D_cumin) RGR_Cumin=g(T,W,P,D,L)×(1-β×H_Rose) Where T is accumulated temperature, W is moisture, N / P is nutrients, D is planting density, H is canopy closure, and α / β is competition coefficient, which are determined by fitting using nonlinear least squares method.

3. The method according to claim 1, characterized in that, The optimization process executed by the intelligent decision-making system in step (3) is as follows: ① Input the weather forecast data, soil test data, and expected market price for the year; ②Use historical similar year datasets to extract optimal planting patterns; ③ Run the multi-objective particle swarm optimization algorithm and iterate until the convergence accuracy is ≤0.01; ④ Output the optimal parameter combination: wide row spacing 120-150cm, narrow row spacing 30-40cm, rose plant spacing 80-100cm, cumin sowing row spacing 15-20cm, and sowing period staggered by 5-15 days.

4. The method according to claim 1, characterized in that, The specific implementation of the dynamic water and fertilizer synergistic regulation in step (5) is as follows: Soil moisture sensors were buried 15cm away from the center of the wide row of roses and 15cm away from both sides of the narrow row of cumin (two layers at depths of 20cm and 40cm respectively); The control system reads sensor data and triggers irrigation when the substrate potential is below -30 kPa. Fertilization decisions are based on real-time crop leaf area index (LAI) values. The decision tree model determines the growth stage and automatically matches the fertilizer solution formula. The LAI is obtained by inversion from UAV multispectral images with an accuracy of ≥90%.

5. The method according to claim 1, characterized in that, The knowledge base self-learning update mechanism in step (6) is as follows: Construct a time series forecasting model, input data from seasons t-3, t-2, and t-1, and predict the optimal parameters for season t. The Long Short-Term Memory (LSTM) neural network was used with 64 hidden layer nodes and a learning rate of 0.

001. After each harvest, the measured yield, quality indicators, and input cost data are uploaded, automatically triggering model retraining and updating the weight matrix, so that the decision error rate is reduced by ≥5% year by year.

6. The method according to claim 1, characterized in that, The UAV remote sensing system is used for: ① Pre-sowing topographic mapping generates a high-precision DEM, which assists in ridging operations with an accuracy of ±2cm; ② Acquire multispectral images weekly during the growing season, calculate NDVI and GNDVI vegetation indices, and monitor differences in growth. ③Then, identify the maturity of the pods in the later stage, and send a harvest warning when the proportion of yellow pixels in the image is ≥70%.

7. A smart agricultural management system for implementing the methods of claims 1-6, characterized in that, include: Sensing layer: Composed of soil sensors, weather stations, drones, and high-definition cameras, with a data acquisition frequency of ≥1 time / hour; Transport layer: Wireless transmission network based on LoRaWAN or NB-IoT, with a data packet loss rate of <5%; Decision-making level: Big data analytics and AI models deployed in edge computing gateways or cloud servers with a response latency of <30 seconds; Execution layer: intelligent water and fertilizer integrated machine, automated seeder, variable fertilizer application device, with control accuracy ≥95%.

8. The system according to claim 7, characterized in that, The decision-making layer comprises a dual-mode decision engine consisting of an expert knowledge base and a machine learning model. When the historical data coverage is less than 30%, the rule engine is activated to calculate parameters based on expert experience formulas. When the data coverage is ≥30%, switch to the data-driven model to achieve intelligent decision-making; If the difference between the output results of the two modes is greater than 15%, a manual review mechanism will be initiated.

9. The system according to claim 7, characterized in that, The integrated water and fertilizer machine in the execution layer has the following features: The three-channel mother liquor injection system stores nitrogen, phosphorus, and potassium concentrates separately and uses PID control to dispense fertilizer in real time. EC / pH online monitoring feedback, with adjustment accuracy of EC ±0.1 mS / cm and pH ±0.2; The land parcels are managed in a grid system, with the smallest control unit area ≤ 0.5 mu.

10. The method according to claim 1, characterized in that, This also includes market risk aversion decisions: When making the decision in step (3), the cumin and rose futures market data and the Chinese medicinal materials price index are accessed simultaneously; Construct a risk-return model: Expected return = ∑(Expected output × Expected price) - Input cost - Risk loss; When market volatility exceeds 20%, the planting density is automatically adjusted by 10% as a buffer strategy, generating three options for users to choose from: conservative, balanced, and aggressive.

Images