Honeysuckle polyploidy cultivation environment regulation and control system based on intelligent perception

Abstract

The invention relates to the technical field of agricultural automation control, and particularly discloses a honeysuckle polyploidy cultivation environment regulation and control system based on intelligent sensing, and the system comprises a multi-source fusion sensing module which is used for collecting environment and plant physiological data; the physiological state interpretation and stage recognition module is used for judging a growth stage through a deep learning model; the dynamic environment strategy generation module is used for dynamically generating an environment target value according to the stages and the physiological parameters; and the cooperative execution control module is used for controlling a cooperative executor through hierarchical prediction to realize accurate regulation and control. According to the method, accurate matching of environment regulation and real-time physiological needs of plants can be achieved, the induction success rate and the seedling robustness are improved, and meanwhile energy consumption is reduced.

Description

Technical Field

[0001] This invention belongs to the field of agricultural automation control technology, specifically relating to a honeysuckle polyploid cultivation environment control system based on intelligent sensing. Background Technology

[0002] In the fields of agricultural modernization and precision agriculture, plant factories and facility agriculture are important models for achieving year-round, standardized, and efficient crop production. Their core lies in creating optimal physicochemical conditions for crop growth through artificial environmental control technology. The environmental control system, which integrates sensors, actuators, and controllers to monitor and regulate key environmental factors such as temperature, humidity, light, and nutrient solution, is the foundation for achieving precision agriculture.

[0003] The industrialized seedling production and breeding of medicinal plants, especially the polyploid induction and cultivation stage, has special requirements regarding the sensitivity, response threshold, and dynamic changes of environmental factors. Polyploid plants, due to the doubling of their chromosome sets, often exhibit significant differences from diploid parents in physiological metabolism, morphogenesis, and environmental adaptability. Their cultivation requires more refined and dynamic environmental control strategies to stabilize induction success rates and promote the expression of desirable traits.

[0004] Existing environmental control systems mostly operate based on preset fixed thresholds or simple linear feedback control logic. These systems struggle to effectively capture and respond to the nonlinear, time-varying dynamic response characteristics of key physiological parameters such as photosynthetic rate, transpiration, and nutrient absorption in polyploid plants at different stages, including the induction period, seedling establishment period, and rapid growth period. Fixed threshold control easily leads to frequent fluctuations in environmental parameters around the set point or long-term deviations from the plant's actual needs, resulting not only in energy waste but also potentially inducing physiological stress in the plant, affecting polyploid induction efficiency and seedling vigor. Furthermore, traditional systems lack the ability to non-destructively and in real-time perceive plant phenotypes and physiological states, failing to establish a closed-loop correlation between environmental control decisions and the plant's actual growth status, resulting in delayed and indiscriminate control behavior. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent sensing-based environment control system for honeysuckle polyploid cultivation, in order to solve the technical contradiction that existing environment control systems based on fixed thresholds cannot adapt to the nonlinear and time-varying dynamic physiological responses of polyploid plants, resulting in lagging regulation, high energy consumption, and the potential to induce physiological stress in plants.

[0006] This invention provides an intelligent sensing-based environment control system for honeysuckle polyploid cultivation. Deployed within a closed or semi-closed plant cultivation facility, the system's core lies in constructing a closed-loop intelligent control system that uses the plant's real-time physiological and ecological state as a feedback source, dynamic growth stages as a decision-making framework, and multimodal environmental actuators as control methods. The system includes a multi-source fusion sensing module, a physiological state interpretation and stage identification module, a dynamic environment strategy generation module, and a collaborative execution control module.

[0007] The multi-source fusion sensing module is used for synchronous, high-frequency acquisition of environmental physicochemical factors and plant phenotypic and physiological signals within the cultivation unit. This module integrates multiple functional sensing sub-units. The environmental physicochemical factor sensing sub-unit includes air temperature and humidity sensors, carbon dioxide concentration sensors, light intensity and spectrum sensors, nutrient solution conductivity sensors, pH sensors, and dissolved oxygen sensors. The plant-specific sensing sub-unit includes a high-resolution multispectral imaging unit, a leaf surface temperature infrared thermometer, and a stem flow meter. The high-resolution multispectral imaging unit acquires top-view and side-view images of all plants within the cultivation area at a preset acquisition cycle, with image data covering red, green, blue, near-infrared, and specific red-edge bands. The leaf surface temperature infrared thermometer synchronously measures the average leaf temperature of the sample plant canopy in a non-contact manner. The stem flow meter, based on the principle of heat dissipation, is installed at the base of the main stem of a representative plant to continuously monitor the plant's transpiration rate. All sensors and acquisition units are connected to the central processing unit via a fieldbus network, and a timestamp synchronization mechanism ensures the spatiotemporal alignment of multi-source data.

[0008] The physiological state interpretation and stage identification module is connected to the multi-source fusion sensing module to process and analyze the collected raw data to interpret the real-time physiological state of the plants and determine their cultivation stage. This module first preprocesses the multispectral image data, including background segmentation, plant contour extraction, and noise filtering. Then, it calculates a series of vegetation indices, specifically the Normalized Difference Vegetation Index (NDVI), photochemical reflectance index, and modified chlorophyll absorption reflectance index. Simultaneously, based on stem flow meter data and synchronously collected air temperature, humidity, and light data, the module uses a simplified variant of the Penman-Montes equation to estimate the plant's transpiration rate in real time. Furthermore, the module incorporates a growth stage classification model based on a deep convolutional neural network. The model's input consists of a multidimensional time-series feature vector composed of vegetation index sequences extracted from three consecutive acquisition cycles, estimated transpiration rate sequences, and leaf temperature sequences. The model's output is the probability distribution of the current plant population being in one of three preset cultivation stages: induction stage, seedling establishment stage, or rapid growth stage. This module identifies the stage with the highest probability value as the current dominant growth stage and outputs the stage identifier along with the real-time calculated average vegetation index, average transpiration rate, and average leaf temperature.

[0009] The dynamic environmental strategy generation module is connected to the physiological state interpretation and stage identification module. It dynamically generates a set of optimal environmental factor target settings based on the identified growth stage and real-time physiological state parameters. This module internally stores three environmental strategy knowledge bases corresponding to the induction period, seedling establishment period, and rapid growth period, respectively. Each environmental strategy knowledge base does not store fixed settings, but rather a series of dynamic response functions with key physiological state parameters as independent variables and environmental factor target values ​​as dependent variables. For the induction period, the target light intensity setting is a negative correlation function of the real-time photochemical reflectance index, and the target diurnal temperature range setting is a first-order function of the real-time transpiration rate change rate. For the seedling establishment period, the target air humidity setting is a proportional integral function of the real-time leaf temperature and air temperature difference, and the target nutrient solution conductivity setting is a lookup table function of the ratio of the modified chlorophyll absorption reflectance index to the normalized difference vegetation index. For the rapid growth period, the target carbon dioxide concentration setting is a saturation response function of the real-time estimated net photosynthetic rate, and the target photoperiod setting is a linear function of the plant canopy coverage. This module receives the stage identifier and physiological parameters output by the physiological state interpretation and stage identification module, calls the corresponding dynamic response function from the corresponding environmental strategy knowledge base for calculation, and outputs in real time a complete set of environmental factor target settings including temperature, humidity, light intensity and spectrum, light cycle, carbon dioxide concentration, nutrient solution conductivity, and pH.

[0010] The collaborative execution control module connects to environmental sensors in the dynamic environment strategy generation module and the multi-source fusion sensing module. It coordinates and controls multiple environmental actuators within the cultivation facility based on the generated set of environmental factor target values, achieving precise regulation. This module employs a hierarchical predictive control architecture. The upper layer is a coordinating optimizer, operating on a minute-level cycle. It receives the target set of values ​​from the dynamic environment strategy generation module and comprehensively considers the coupling relationships between various environmental factors and the energy consumption characteristics of the actuators. By solving a constrained multi-objective optimization problem, it calculates the optimal control sequence for each actuator within a 15-minute time window. The objective function of the multi-objective optimization problem aims to minimize the combined deviation between the measured environmental values ​​and the target values, as well as the total system energy consumption. Constraints include the actuator's power limit, action rate limit, and physiological safety thresholds for environmental factor changes. The lower layer is a distributed high-speed closed-loop controller, operating on a second-level cycle, including a temperature controller, humidity controller, intelligent lighting controller, carbon dioxide injection controller, and nutrient solution circulation controller. Each controller receives short-term setpoint instructions from the coordinator and generates drive signals based on real-time measurements from corresponding sensors using a proportional-integral-derivative control algorithm. These signals control the heating and cooling units, humidification and dehumidification units, tunable spectrum LED array, carbon dioxide solenoid valve, nutrient solution pump, and acid-base regulating pump, enabling each environmental parameter to quickly and smoothly track its dynamic target value.

[0011] In one embodiment of the present invention, the high-resolution multispectral imaging unit in the multi-source fusion sensing module employs a combination of supergreen index and adaptive thresholding for image background segmentation. First, the supergreen index value of each pixel in the image is calculated. Then, the Otsu method is used to automatically calculate the segmentation threshold on the supergreen index image, distinguishing pixels as plants and background. For segmentation holes caused by occlusion or uneven lighting, a filling algorithm based on morphological closing operations is used to repair them, ensuring the integrity of the plant outline.

[0012] As one embodiment of the present invention, the training process of the growth stage classification model based on a deep convolutional neural network in the physiological state interpretation and stage identification module is as follows: First, in historical cultivation experiments, a large number of multispectral image sequences, stem flow data sequences, and leaf temperature data sequences covering different growth stages are collected simultaneously, and experts label the corresponding growth stages according to plant morphology and physiological indicators. Subsequently, multidimensional time series feature vectors consistent with the current module input are extracted from these sequence data to form a training dataset. This dataset is divided into a training set, a validation set, and a test set. The training set is used to train a network model containing two one-dimensional convolutional layers, two pooling layers, and three fully connected layers. During the training process, the cross-entropy loss function and adaptive moment estimation algorithm are used for optimization, and the model performance is monitored through the validation set to prevent overfitting. Finally, after achieving a stage identification accuracy of no less than 95% on the test set, the model parameters are fixed and deployed in the module.

[0013] In one embodiment of the present invention, the parameters of the dynamic response function stored in the dynamic environment strategy generation module are calibrated by combining historical data-driven and physiological mechanism models. Specifically, within a controlled environment chamber, a series of stepwise or sinusoidal environmental factor treatments are applied to honeysuckle polyploid plants at different growth stages, while continuously monitoring the plants' physiological response indicators. Using the collected environmental stimulus and physiological response datasets, a nonlinear regression analysis method is employed to fit a function form and parameters that best describe the quantitative relationship between physiological parameters and environmental target values. For some complex relationships, a nonlinear mapping based on a radial basis function neural network is used for approximation, and the trained network weights are stored as a lookup table.

[0014] In one embodiment of the present invention, the coordinated optimizer in the cooperative execution control module employs a non-dominated sorting genetic algorithm to solve constrained multi-objective optimization problems. This algorithm encodes the control action sequence of each actuator over the next 15 minutes as a chromosome, and iteratively evolves the population through selection, crossover, and mutation operations. Each generation of individuals is evaluated for its corresponding environmental tracking error and energy consumption by simulating actuator models and environmental response models. The algorithm ultimately outputs a Pareto optimal solution set, from which a decision rule selects a solution that achieves a balance between environmental tracking accuracy and energy consumption as the optimal control sequence and sends it to the lower-level controller.

[0015] In one embodiment of the invention, the system further includes a strategy self-evolution module connected to the physiological state interpretation and stage identification module and the dynamic environment strategy generation module. The strategy self-evolution module continuously collects full-cycle data for each cultivation batch from start to finish, including ultimate quality indicators such as the final polyploid induction success rate, plant biomass, and key active ingredient content, as well as environmental regulation records and physiological state trajectories throughout the growth process. This module employs a reinforcement learning framework, treating the dynamic environment strategy generation module as an agent and the cultivation environment and plants as the environment. The agent's actions involve adjusting the parameters of the dynamic response function in the environmental strategy knowledge base, with the environment's reward signal based on a comprehensive evaluation of the ultimate quality indicators. Through offline training, the strategy self-evolution module continuously optimizes the parameters of the dynamic response function, enabling the system to adaptively improve long-term cultivation efficiency.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention constructs a closed-loop intelligent control system that uses the real-time physiological and ecological state of plants as a direct feedback source, completely changing the passive mode of traditional environmental control based on fixed set points or simple feedback. The system achieves non-destructive and continuous monitoring of plant phenotype and physiological state through multi-source fusion sensing, and accurately identifies its dynamic growth stage using a deep learning model, so that environmental control decisions are based on the actual needs of the plant.

[0017] 2. The dynamic environment strategy generation module defines the target values ​​of environmental factors as functions of key physiological parameters, realizing real-time matching between the regulation strategy and the nonlinear physiological response of the plant. This ensures that the environmental conditions are always close to the optimal demand range of the plant at different stages, thereby significantly improving the success rate of honeysuckle polyploid induction and seedling vigor, while avoiding physiological stress caused by unsuitable environment.

[0018] 3. The collaborative execution control module adopts hierarchical predictive control, which optimizes and coordinates the actions of each actuator through multi-objective optimization. While accurately tracking dynamic targets, it effectively reduces system energy consumption, achieving a balance between high efficiency and energy saving.

[0019] 4. The introduction of the strategy self-evolution module endows the system with the ability to self-optimize in the long term. It can continuously fine-tune and control the strategy based on historical cultivation results data, so that the system performance can be continuously improved with the accumulation of running time, and has significant technological advancement and application value. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention; Figure 2 This is a schematic diagram of the core principle framework of the physiological state interpretation and stage identification module in this invention; Figure 3 This is a logical flow diagram of the dynamic environment strategy generation module in this invention; Figure 4 This is a schematic diagram of the multi-level interaction relationship and data flow of the collaborative execution control module in this invention; Figure 5 This is a schematic diagram of the reinforcement learning optimization framework of the policy self-evolution module in this invention. Detailed Implementation

[0021] Example 1: The overall technical architecture of the honeysuckle polyploid cultivation environment control system based on intelligent sensing proposed in this invention is shown in the attached figure. Figures 1 to 5 As shown, this system is deployed within a closed or semi-closed plant cultivation facility, constructing a closed-loop intelligent control system that uses the real-time physiological and ecological state of the plant as a feedback source, dynamic growth stages as a decision-making framework, and multimodal environmental actuators as the control means. The system comprises a multi-source fusion sensing module, a physiological state interpretation and stage identification module, a dynamic environment strategy generation module, a collaborative execution control module, and a strategy self-evolution module. These modules interact with each other via a high-speed data bus, achieving low-latency and high-reliability information exchange, ensuring the entire control process possesses high real-time performance, accuracy, and adaptability.

[0022] The multi-source fusion sensing module serves as the system's sensing front end, responsible for the synchronous and high-frequency acquisition of environmental physicochemical factors and plant phenotypic and physiological signals within the cultivation unit. This module integrates multiple functional sensing sub-units, all of which are connected to the central processing unit via an industrial-grade fieldbus network. A hardware-level timestamp synchronization mechanism ensures strict temporal alignment of data from different sources, with an error not exceeding 50 milliseconds. The environmental physicochemical factor sensing sub-units include air temperature and humidity sensors, carbon dioxide concentration sensors, light intensity and spectrum sensors, nutrient solution conductivity sensors, pH sensors, and dissolved oxygen sensors. Among them, the air temperature and humidity sensor uses a combination of a capacitive humidity sensing element and a platinum resistance temperature probe; the carbon dioxide concentration sensor is based on the principle of non-dispersive infrared absorption, with a range of 0 to 2000 ppm and a resolution of 1 ppm; the light intensity and spectrum sensor is equipped with a silicon photodiode array, which can simultaneously detect red light (620 to 750 nm), green light (495 to 570 nm), blue light (450 to 495 nm), near-infrared light (780 to 1100 nm), and a specific red-edge band (700 to 730 nm), with a light intensity measurement range of 0 to 2000 micromoles per square meter per second and a spectral resolution better than 5 nm; the nutrient solution conductivity sensor uses a four-electrode AC excitation method, with a range of 0.1 to 10 millisiemens per centimeter; the pH sensor is a glass composite electrode, with a measurement range of pH 2 to 12; and the dissolved oxygen sensor is based on the principle of fluorescence quenching, with a range of 0 to 20 milligrams per liter and a response time of less than 30 seconds.

[0023] The plant-based sensing subunit includes a high-resolution multispectral imaging unit, a leaf surface temperature infrared thermometer, and a stem flow meter. The high-resolution multispectral imaging unit consists of two independent imaging systems: a top-view imaging system installed 2 meters directly above the cultivation rack, and a side-view imaging system spaced along the cultivation row direction at a height of 1.2 meters and an angle of 45 degrees. Both systems are equipped with cooled CMOS image sensors with a pixel resolution of no less than 4096×3072 and a frame rate of up to 5 frames per second. The imaging unit is synchronously triggered at a preset acquisition cycle to acquire top-view and side-view images of all plants within the cultivation area. The image data is compressed in real time by an embedded image processor and then transmitted to the central processing unit. The leaf surface temperature infrared thermometer uses a long-wave infrared thermal imager with a spectral response range of 8 to 14 micrometers and a spatial resolution of 1.5 milliradians. It can simultaneously measure the average leaf temperature of the sample plant canopy under non-contact conditions, with a measurement range of -20 to 80 degrees Celsius and an accuracy of ±0.5 degrees Celsius. The stem flow meter is based on the principle of heat dissipation. Its probes are installed at a standard interval at the base of the main stems of three representative honeysuckle polyploid plants, 10 cm above the ground. The probes contain a pair of heating and reference thermocouples. The flow rate is calculated by measuring the temperature difference before and after heating. The sampling frequency is once per second, and the output signal is uploaded in digital form after analog-to-digital conversion.

[0024] The raw data collected by the multi-source fusion sensing module first undergoes data cleaning and format standardization. For image data, background segmentation, plant contour extraction, and noise filtering are performed before entering the physiological state interpretation and stage recognition module. Background segmentation uses a combination of the supergreen index and adaptive thresholding. The supergreen index is defined as 2G-RB, where R, G, and B are the normalized pixel values ​​of the red, green, and blue channels, respectively. First, the supergreen index value of each pixel in the image is calculated to generate a single-channel supergreen index image. Then, Otsu's method is used to automatically calculate the optimal segmentation threshold on the image to distinguish the pixels as plant foreground and background. For segmentation voids caused by leaf occlusion or uneven local lighting, a filling algorithm based on morphological closing operations is used for repair: first, a dilation operation with a radius of 5 pixels is performed on the binary image, followed by an erosion operation with the same radius, thereby effectively connecting broken edges and filling internal holes, ultimately obtaining a complete and continuous plant contour mask. Based on this mask, the system extracts geometric features such as plant projection area, canopy coverage, and leaf tilt angle distribution, and calculates vegetation indices by combining multispectral reflectance.

[0025] The core principle framework of the physiological state interpretation and stage recognition module is attached. Figure 2 As shown. This module receives a multidimensional data stream from the multi-source fusion sensing module and first calculates the vegetation index from the multispectral image. The formula for calculating the Normalized Difference Vegetation Index (NDVI) is as follows: ,in For near-infrared reflectivity, The reflectance is in the red light band; the formula for calculating the photochemical reflectance index (PRI) is: It is used to characterize the photochemical efficiency of photosynthetic system II; the modified chlorophyll absorbance reflectance index (MCARI) is calculated as follows: ,in For red-edge band reflectivity, The green light reflectance index is highly sensitive to changes in chlorophyll content. All the above vegetation indices were calculated as the arithmetic mean of all plants within each collection period, forming a time series.

[0026] Simultaneously, based on stem flow meter data and synchronously collected air temperature, humidity, and light data, this module uses a simplified variant of the Penman-Montes equation to estimate the plant's transpiration rate in real time. This simplified model is expressed as:

[0027] in, This refers to the transpiration rate; The slope of the saturated water vapor pressure curve; Net radiation; Soil heat flux; air density; The specific heat of air at constant pressure; It is the saturated vapor pressure; This is the actual water vapor pressure; For canopy aerodynamic drag; The latent heat of vaporization of water; This is the hygrometer constant. The system calculates the above parameters in real time based on real-time light intensity, air temperature, relative humidity, and wind speed (provided by a miniature anemometer), thereby obtaining an estimated evapotranspiration rate updated every 10 minutes.

[0028] Furthermore, this module incorporates a growth stage classification model based on a deep convolutional neural network. The model's input consists of a multi-dimensional time-series feature vector extracted over three consecutive acquisition cycles (i.e., 30-minute time windows), with nine dimensions: NDVI, PRI, MCARI, transpiration rate, leaf temperature, leaf-air temperature difference, canopy coverage, daily cumulative photosynthetically active radiation, and stem flow rate change. These nine-dimensional vectors are stacked chronologically to form a 3×9 two-dimensional tensor as the network input. The network structure includes two one-dimensional convolutional layers (kernel sizes of 5 and 3, with 32 and 64 output channels, respectively), two max-pooling layers (pooling window size of 2), and three fully connected layers (with 128, 64, and 3 nodes, respectively). The output layer uses the Softmax activation function to output the probability distribution of the current plant population being in one of three preset cultivation stages: induction, seedling establishment, or rapid growth. The model was trained before deployment. Training data consisted of over 5000 sets of multispectral image sequences, stem flow data sequences, and leaf temperature data sequences collected synchronously during historical cultivation experiments. Plant physiology experts labeled the data with stage tags based on plant morphology (e.g., number of new shoots, leaf expansion, root development status) and physiological indicators (e.g., chlorophyll fluorescence Fv / Fm, proline content). During training, a cross-entropy loss function and an adaptive moment estimation algorithm (Adam optimizer, learning rate 0.001) were used, and an early stopping mechanism was employed to monitor performance on the validation set. Ultimately, a stage recognition accuracy of 96.2% was achieved on the independent test set. After the model parameters were solidified, they were embedded in the module's dedicated inference engine, resulting in an inference latency of less than 200 milliseconds.

[0029] The physiological state interpretation and stage identification module determines the stage with the highest probability value as the current dominant growth stage, and outputs the stage identifier along with the real-time calculated average vegetation index, average transpiration rate and average leaf temperature to the dynamic environment strategy generation module.

[0030] The logical flow of the dynamic environment policy generation module is shown in the appendix. Figure 3As shown, this module internally stores three environmental strategy knowledge bases corresponding to the induction period, seedling establishment period, and rapid growth period, respectively. Each knowledge base does not store fixed set values, but rather a series of dynamic response functions with key physiological state parameters as independent variables and environmental factor target values ​​as dependent variables. The parameters of these functions are calibrated through a combination of historical data-driven and physiological mechanism models: In a controlled environment chamber, stepwise or sinusoidal environmental factor perturbations are applied to honeysuckle polyploid plants at different growth stages (e.g., light intensity gradually increases from 200 to 1000 micromoles per square meter per second, with a step size of 200, each step lasting 2 hours), while simultaneously recording plant response indicators such as NDVI, PRI, and transpiration rate at 5-minute intervals. Using thousands of collected stimulus-response data pairs, nonlinear least squares regression is used to fit the optimal function form. For complex nonlinear relationships (such as the relationship between nutrient solution conductivity and the vegetation index ratio), a radial basis function neural network is used for approximation, with 20 nodes in the hidden layer of the network. After training, the weights and center point parameters are stored as a lookup table.

[0031] For the induction period, the target light intensity setpoint Defined as real-time photochemical reflectance index Negative correlation function: in =800, =15, =100, in micromoles per square meter per second. This function reflects that in the early stages of induction, when PRI is low (indicating suppressed photosynthetic efficiency), the system automatically reduces light intensity to alleviate photoinhibition; as PRI recovers, light intensity is gradually increased to promote cell division. Target diurnal temperature range setpoint. The rate of change of real-time transpiration rate First-order function:

[0032] in =−2.5 degrees Celsius per, =8 degrees Celsius. When the transpiration rate increases rapidly, the system reduces the diurnal temperature range to maintain water balance.

[0033] For the seedling establishment period, the target air humidity setpoint The real-time difference between leaf temperature and air temperature Proportional integral function: ,in =3.2% per degree Celsius, =0.1% per degree Celsius per minute, =70%. This strategy aims to increase humidity to reduce water loss when leaf temperature is significantly lower than air temperature (indicating excessive transpiration). Target nutrient solution conductivity setpoint. To improve the ratio of chlorophyll absorption reflectance index to normalized difference vegetation index The lookup table function. The system pre-builds... and The mapping table: when R < 0.8, =1.2 millisiemens per centimeter; when 0.8 ≤ R < 1.2, =1.5; when R≥1.2, =1.8. This ratio reflects nitrogen nutrition status; a higher ratio indicates more active chlorophyll synthesis and a greater need for nutrient supply.

[0034] For the rapid growth period, the target carbon dioxide concentration setpoint Net photosynthetic rate for real-time estimation Saturated response function: in =1200ppm, =300ppm, This represents the current carbon dioxide concentration. This function ensures that the carbon dioxide concentration remains high to maximize carbon assimilation efficiency when the photosynthetic rate is near saturation. Target photoperiod setpoint. Canopy coverage of the plant population Linear functions:

[0035] in Normalized to the 0 to 1 range, the light cycle is dynamically adjusted between 14 and 20 hours. The higher the coverage, the longer the light is provided to support vigorous growth.

[0036] The dynamic environment strategy generation module receives the stage identifier and physiological parameters output by the physiological state interpretation and stage identification module, calls the corresponding dynamic response function from the corresponding environmental strategy knowledge base for calculation, updates every 10 minutes, and outputs in real time a complete set of environmental factor target settings including temperature (day / night setpoint), humidity, light intensity and spectral ratio (based on red:blue:far-red=6:3:1, fine-tuned according to PRI), light cycle, carbon dioxide concentration, nutrient solution conductivity, and pH (maintained at 5.8±0.2).

[0037] The multi-level interaction relationships and data flow of the collaborative execution control module are shown in the appendix. Figure 4As shown. This module adopts a hierarchical predictive control architecture, with a coordinated optimizer at the upper layer and a distributed high-speed closed-loop controller at the lower layer. The coordinated optimizer runs on a 5-minute cycle, receiving a set of target setpoints from the dynamic environment strategy generation module. It comprehensively considers the strong coupling relationships between various environmental factors (e.g., humidification lowers the temperature, supplemental lighting raises the temperature and evaporation) and the energy consumption characteristics of the actuators (e.g., high energy consumption during chiller start-up and shutdown, linear energy consumption during LED dimming). By solving a constrained multi-objective optimization problem, it calculates the optimal control sequence for each actuator within a 15-minute time window. The objective function of this multi-objective optimization problem is:

[0038] in, Let be the measured value of the i-th environmental factor (out of 7). Its dynamic target value, The total instantaneous power of the system. =0.7、 =0.3 is the weighting coefficient, reflecting the optimization orientation of prioritizing environmental tracking accuracy and secondarily considering energy consumption. The constraints include: heating power limit of 5 kW, cooling power limit of 8 kW, humidification rate limit of 20 grams of water per minute, LED total luminous flux change rate not exceeding 10% per minute, carbon dioxide injection rate not exceeding 500 ml per minute, and nutrient solution pH adjustment pump flow rate not exceeding 2 liters per minute; at the same time, the rate of change of environmental factors is limited by physiological safety thresholds, such as temperature change rate not exceeding 2 degrees Celsius per hour and humidity change rate not exceeding 10% per hour.

[0039] This optimization problem is solved using the Non-Dominated Sorting Genetic Algorithm (NSGA-II). The algorithm encodes the control action sequence of each actuator over the next 15 minutes (15 control points in 1-minute increments) into chromosomes, with a population size of 100 and a maximum generation count of 50. Each generation is simulated using a simplified environmental dynamics model (based on a first-order inertial link and coupling gain matrix), and its corresponding environmental tracking error and energy consumption are evaluated. The algorithm ultimately outputs a Pareto optimal solution set, from which a solution balancing environmental tracking accuracy and energy consumption is selected as the optimal control sequence by a decision rule (selecting the solution with the highest overall evaluation value, where evaluation value = 0.6 × (1 - normalized error) + 0.4 × (1 - normalized energy consumption)). This optimal sequence is then sent to the next-level controller.

[0040] The lower layer is a distributed high-speed closed-loop controller, operating on a 1-second cycle, including a temperature controller, humidity controller, intelligent lighting controller, carbon dioxide injection controller, and nutrient solution circulation controller. Each controller receives short-term setpoint instructions from the coordinator (updated every 5 minutes, but segmented constant values ​​within a 15-minute window) and generates drive signals based on real-time measurements from corresponding sensors using a proportional-integral-derivative (PID) control algorithm. The temperature controller outputs signals to control the electric heating wire and the variable-frequency compressor refrigeration unit; the humidity controller drives the ultrasonic humidifier and the rotary dehumidifier; the intelligent lighting controller adjusts the current of the red, blue, and far-infrared LED arrays through pulse width modulation signals to achieve independent control of the spectrum and light intensity; the carbon dioxide injection controller controls the duty cycle of the solenoid valve to regulate the release rate of the food-grade carbon dioxide cylinder; and the nutrient solution circulation controller coordinates the start / stop and flow rate of the nitrogen, phosphorus, potassium, calcium, magnesium, and sulfur pumps and the phosphate / potassium hydroxide acid-base regulating pump to ensure that the conductivity and pH remain stable within ±2% of the target value. All controllers are equipped with anti-integral saturation and setpoint filtering functions to ensure that the system can still smoothly transition when the setpoint changes abruptly.

[0041] The reinforcement learning optimization framework for the policy self-evolution module is attached. Figure 5 As shown. This module connects to the physiological state interpretation and stage identification module and the dynamic environment strategy generation module, continuously collecting full-cycle data for each breeding batch from start to finish. The collected data includes: ultimate quality indicators (polyploid induction success rate, plant fresh weight, dry matter accumulation, chlorogenic acid and luteolin content), environmental regulation records (actuator action logs, historical environmental parameter trajectories), and physiological state trajectories (time series of NDVI, PRI, transpiration rate, etc.). This module employs an offline deep Q-network (DQN) reinforcement learning framework, incorporating the dynamic response function parameters (such as...) from the dynamic environment strategy generation module... The agent's action space is defined by 24 adjustable parameters (including the cultivation environment and the plant itself); the reward signal is considered the external environment. Defined as the weighted sum of ultimate quality metrics:

[0042] in The success rate of induction is 0 to 1. The percentage increase in biomass relative to the baseline batch (from -1 to 1). The percentage increase in active ingredient content is represented by a range of -1 to 1. The agent is trained by replaying historical batches of experience tuples (state, action, reward, next state), using double-Q learning and priority experience replay to improve stability. Every 10 breeding batches, the policy self-evolution module updates the dynamic response function parameters and deploys the new policy to the dynamic environment policy generation module, enabling continuous evolution of the regulatory policy.

[0043] This embodiment achieves precise, dynamic, and adaptive control of the cultivation environment for honeysuckle polyploids through the close collaboration of the aforementioned modules. In actual operation, the system can adjust environmental settings in real time according to the actual physiological needs of the plants, avoiding over-intervention or delayed response caused by traditional fixed threshold control. For example, if the PRI level is detected to be consistently below 0.02 for 30 minutes during the induction period, the system will automatically reduce the light intensity from 600 to 300 micromoles per square meter per second and decrease the diurnal temperature range from 8 degrees Celsius to 5 degrees Celsius, effectively alleviating photoinhibition and water stress. After entering the rapid growth phase, if the canopy coverage reaches 0.75, the system will extend the photoperiod from 16 hours to 18.5 hours and increase the carbon dioxide concentration to 1000 ppm, significantly promoting biomass accumulation.

[0044] Example 2: Building upon Example 1, this example enhances the high-resolution multispectral imaging unit in the multi-source fusion sensing module by introducing 3D point cloud reconstruction technology to improve the extraction accuracy of plant phenotypic parameters. Specifically, in addition to the existing top-view and side-view imaging systems, the high-resolution multispectral imaging unit adds a structured light projection device. This device consists of a high-brightness blue laser line generator and a high-speed global shutter camera, mounted coaxially with the side-view imaging system. Simultaneously with each multispectral image acquisition, the structured light projection device projects a vertical laser plane onto the plant canopy, and the high-speed camera captures laser stripe deformation images at a rate of 30 frames per second. The central processing unit utilizes the triangulation principle, combined with the known projection-imaging baseline distance and camera intrinsic parameters, to perform sub-pixel-level centerline extraction and 3D coordinate inversion on each frame of stripe image. Finally, it fuses multi-view point cloud data to generate a high-density 3D point cloud model of the plant, with a point cloud density of no less than 500 points per cubic centimeter.

[0045] Based on this 3D point cloud model, the system can accurately calculate advanced phenotypic parameters such as the plant's true leaf area index, leaf tilt angle distribution, canopy volume, and spatial distribution uniformity. These parameters are introduced into the physiological state interpretation and stage recognition module as additional input features for growth stage classification. For example, the temporal rate of change of leaf area index is used to distinguish the transitional stage between the seedling establishment period and the rapid growth period: when the daily growth rate of leaf area index reaches 0.15 or higher and lasts for 2 consecutive days, the system determines that the plant has entered the rapid growth period 0.5 days earlier, thus initiating a high-light, high-carbon dioxide strategy earlier. In addition, the canopy spatial distribution uniformity index (defined as the ratio of the standard deviation to the mean of the leaf area index in each sub-region) is used to dynamically adjust the local supplemental lighting strategy. If the uniformity index of a certain region exceeds 0.3, the collaborative execution control module will instruct the addressable LEDs above that region to increase the light intensity by 10% to 20%, achieving active balance of light within the canopy and avoiding localized etiolation or weak light suppression.

[0046] Meanwhile, this embodiment refines the reward function of the strategy self-evolution module by introducing an energy efficiency factor. The new reward signal is defined as follows:

[0047] in Energy efficiency is calculated as (biomass increment + active ingredient increment) / total energy consumption, normalized to the range of -1 to 1. This adjustment encourages reinforcement learning agents to prioritize resource utilization efficiency while pursuing high yield and quality, further optimizing the long-term economic efficiency and sustainability of the system.

Claims

1. A honeysuckle polyploid cultivation environment control system based on intelligent sensing, characterized in that, include: The multi-source fusion sensing module is used to simultaneously collect environmental physicochemical factors and plant phenotypic and physiological signals within the cultivation unit; The physiological state interpretation and stage identification module is connected to the multi-source fusion sensing module and is used to process and analyze the collected raw data to interpret the real-time physiological state of the plant and determine its cultivation stage. The dynamic environment strategy generation module is connected to the physiological state interpretation and stage identification module, and is used to dynamically generate a set of optimal environmental factor target settings based on the identified growth stage and real-time physiological state parameters. The collaborative execution control module is connected to the environmental sensors in the dynamic environment strategy generation module and the multi-source fusion perception module, and is used to coordinate and control multiple environmental actuators in the cultivation facility according to the generated set of environmental factor target settings.

2. The honeysuckle polyploid cultivation environment control system based on intelligent sensing according to claim 1, characterized in that, The multi-source fusion sensing module includes an environmental physicochemical factor sensing subunit and a plant body sensing subunit. The environmental physicochemical factor sensing subunit includes an air temperature and humidity sensor, a carbon dioxide concentration sensor, a light intensity and spectrum sensor, a nutrient solution conductivity sensor, a pH sensor, and a dissolved oxygen sensor. The plant body sensing subunit includes a high-resolution multispectral imaging unit, a leaf surface temperature infrared thermometer, and a stem flow meter. The high-resolution multispectral imaging unit acquires top-view and side-view images of all plants in the cultivation area at a preset acquisition cycle. The image data covers red light, green light, blue light, near-infrared light, and specific red-edge bands. The leaf surface temperature infrared thermometer synchronously measures the average leaf temperature of the sample plant canopy in a non-contact manner. The stem flow meter is installed at the base of the main stem of a representative plant based on the principle of heat dissipation to continuously monitor the transpiration flow rate of the plant. All sensors and acquisition units are connected to the central processing unit through a fieldbus network, and the spatiotemporal alignment of multi-source data is ensured through a timestamp synchronization mechanism.

3. The honeysuckle polyploid cultivation environment control system based on intelligent sensing according to claim 2, characterized in that, The physiological state interpretation and stage recognition module first preprocesses the multispectral image data, including background segmentation, plant contour extraction, and noise filtering. Then, it calculates a series of vegetation indices, specifically the normalized difference vegetation index, photochemical reflectance index, and modified chlorophyll absorption reflectance index. Simultaneously, based on stem flow meter data and synchronously acquired air temperature, humidity, and light data, the module uses a simplified variant of the Penman-Montes equation to estimate the plant's transpiration rate in real time. Furthermore, the physiological state interpretation and stage recognition module incorporates a deep convolutional neural network... The growth stage classification model of the network takes as input a multidimensional time series feature vector composed of vegetation index sequences, estimated transpiration rate sequences, and leaf temperature sequences extracted within three consecutive collection cycles. The output of the growth stage classification model is the probability distribution of the current plant population being in one of three preset cultivation stages: induction period, seedling establishment period, or rapid growth period. The physiological state interpretation and stage identification module determines the stage with the highest probability value as the current dominant growth stage and outputs the stage identifier along with the real-time calculated average vegetation index, average transpiration rate, and average leaf temperature.

4. The honeysuckle polyploid cultivation environment control system based on intelligent sensing according to claim 3, characterized in that, The dynamic environmental strategy generation module internally stores three environmental strategy knowledge bases corresponding to the induction period, seedling establishment period, and rapid growth period, respectively. Each environmental strategy knowledge base stores a series of dynamic response functions with key physiological state parameters as independent variables and environmental factor target values ​​as dependent variables. For the induction period, the target light intensity is set as a negative correlation function of the real-time photochemical reflectance index, and the target diurnal temperature difference is set as a first-order function of the real-time transpiration rate change rate. For the seedling establishment period, the target air humidity is set as a proportional integral function of the real-time leaf temperature and air temperature difference, and the target nutrient solution conductivity is set as a function of the improved chlorophyll absorption reflectance index and... The lookup table function for the normalized difference vegetation index ratio is used. For the rapid growth period, the target carbon dioxide concentration is set as the saturation response function of the net photosynthetic rate estimated in real time, and the target photoperiod is set as a linear function of the canopy coverage of the plant population. The dynamic environmental strategy generation module receives the stage identifier and physiological parameters output by the physiological state interpretation and stage identification module, calls the corresponding dynamic response function from the corresponding environmental strategy knowledge base for calculation, and outputs in real time a complete set of target values ​​for environmental factors including temperature, humidity, light intensity and spectrum, photoperiod, carbon dioxide concentration, nutrient solution conductivity, and pH.

5. The honeysuckle polyploid cultivation environment control system based on intelligent sensing according to claim 4, characterized in that, The collaborative execution control module adopts a hierarchical predictive control architecture. The upper layer is a coordinating optimizer, which operates on a minute-level cycle. It receives the target setpoints from the dynamic environment strategy generation module and comprehensively considers the coupling relationship between various environmental factors and the energy consumption characteristics of the actuators. By solving a constrained multi-objective optimization problem, it calculates the optimal control sequence for each actuator within a 15-minute time window. The objective function of the multi-objective optimization problem aims to minimize the comprehensive deviation between the measured environmental values ​​and the target values, as well as the total energy consumption of the system. The constraints include the upper limit of the actuator's power, the limit of the action rate, and the physiological safety threshold of environmental factor changes. The lower layer is a distributed high-speed closed-loop controller, which operates on a second-level cycle. It includes a temperature controller, a humidity controller, an intelligent lighting controller, a carbon dioxide injection controller, and a nutrient solution circulation controller. Each controller receives short-term setpoint instructions from the coordinating optimizer and, based on the real-time measurement values ​​fed back by the corresponding sensors, uses a proportional-integral-derivative control algorithm to generate drive signals to control the heating and cooling units, the humidification and dehumidification units, the adjustable spectrum light-emitting diode array, the carbon dioxide solenoid valve, and the nutrient solution component pump and acid-base adjustment pump, respectively, so that each environmental parameter tracks its dynamic target value.

6. The honeysuckle polyploid cultivation environment control system based on intelligent sensing according to claim 5, characterized in that, The high-resolution multispectral imaging unit uses a combination of supergreen index and adaptive thresholding for image background segmentation. The specific process is as follows: calculate the supergreen index value of each pixel in the image, and then use the Otsu method to automatically calculate the segmentation threshold on the supergreen index image to distinguish the pixels into plants and background; for segmentation holes caused by occlusion or uneven lighting, a filling algorithm based on morphological closing operation is used for repair.

7. The honeysuckle polyploid cultivation environment control system based on intelligent sensing according to claim 6, characterized in that, The training process of the growth stage classification model based on deep convolutional neural networks is as follows: In historical cultivation experiments, a large number of multispectral image sequences, stem flow data sequences, and leaf temperature data sequences covering different growth stages are collected simultaneously, and experts label the corresponding growth stages according to plant morphology and physiological indicators; multidimensional time series feature vectors consistent with the current module input are extracted from these sequence data to form a training dataset; the training dataset is divided into a training set, a validation set, and a test set; a network model containing two one-dimensional convolutional layers, two pooling layers, and three fully connected layers is trained using the training set. During the training process, the cross-entropy loss function and adaptive moment estimation algorithm are used for optimization, and the model performance is monitored through the validation set to prevent overfitting; after achieving a stage recognition accuracy of no less than 95% on the test set, the model parameters are solidified and deployed in the physiological state interpretation and stage recognition module.

8. The honeysuckle polyploid cultivation environment control system based on intelligent sensing according to claim 7, characterized in that, The parameters of the dynamic response function stored in the dynamic environment strategy generation module are calibrated by combining historical data-driven and physiological mechanism models. Specifically, in a controlled environment chamber, a series of stepwise or sinusoidal environmental factor treatments are applied to honeysuckle polyploid plants at different growth stages, while the physiological response indicators of the plants are continuously monitored. Using the collected environmental stimulus and physiological response datasets, a nonlinear regression analysis method is used to fit the function form and parameters that best describe the quantitative relationship between physiological parameters and environmental target values. For some complex relationships, a nonlinear mapping based on radial basis function neural networks is used for approximation, and the trained network weights are stored as a lookup table.

9. The honeysuckle polyploid cultivation environment control system based on intelligent sensing according to claim 8, characterized in that, The coordinated optimizer in the cooperative execution control module uses a non-dominated sorting genetic algorithm to solve constrained multi-objective optimization problems. The specific process is as follows: the control action sequence of each actuator in the next 15 minutes is encoded as a chromosome, and the population is iteratively evolved through selection, crossover, and mutation operations; each generation of individuals is evaluated for its corresponding environmental tracking error and energy consumption by simulating the actuator model and the environmental response model; the algorithm finally outputs a Pareto optimal solution set, and the decision rule selects a solution that achieves a balance between environmental tracking accuracy and energy consumption as the optimal control sequence and sends it to the lower-level controller.

10. The honeysuckle polyploid cultivation environment control system based on intelligent sensing according to claim 9, characterized in that, It also includes a strategy self-evolution module, which is connected to the physiological state interpretation and stage identification module and the dynamic environment strategy generation module. The strategy self-evolution module continuously collects full-cycle data for each cultivation batch from start to finish, including the final polyploid induction success rate, plant biomass, key active ingredient content and other ultimate quality indicators, as well as environmental regulation records and physiological state trajectories throughout the growth process. The strategy self-evolution module adopts a reinforcement learning framework, treating the dynamic environment strategy generation module as an agent, and the cultivation environment and plants as the environment. The agent's actions are to adjust the parameters of the dynamic response function in the environmental strategy knowledge base, and the reward signal of the environment is based on the comprehensive evaluation of the ultimate quality indicators. Through offline training, the strategy self-evolution module continuously optimizes the parameters of the dynamic response function.

Images