Plant factory leaf vegetable growth state sensing and adaptive light supplementing regulation method

Abstract

This invention relates to the field of intelligent equipment and environmental control technology for facility agriculture, and discloses a method for sensing the growth status and adaptive supplemental lighting control of leafy vegetables in plant factories. The method includes the following steps: Step S1: Multimodal data acquisition; Step S2: Multimodal data fusion and growth status assessment; Step S3: Intelligent identification of growth stages and prediction of growth trends; Step S4: Adaptive optimization of multi-dimensional supplemental lighting parameters, etc. This method for sensing the growth status and adaptive supplemental lighting control of leafy vegetables in plant factories, through the fusion of hyperspectral, RGB-D, and chlorophyll fluorescence multimodal data plus attention mechanism feature weighting, can achieve comprehensive perception of leafy vegetable morphology, physiology, and spectral information, significantly improving the accuracy of growth status assessment. Through growth stage identification, LSTM trend prediction, and improved particle swarm optimization, it can achieve synchronous adaptive control of spectrum, light intensity, and photoperiod, while simultaneously improving leafy vegetable yield and quality, and significantly reducing supplemental lighting energy consumption, making the control more intelligent and energy-efficient.

Description

Technical Field

[0001] This invention relates to the field of intelligent equipment and environmental control technology for facility agriculture, specifically a method for sensing the growth status of leafy vegetables in plant factories and adaptive supplemental lighting control. Background Technology

[0002] Plant factories, as a highly efficient, controllable, intensive, and environmentally closed modern agricultural production method, achieve high-density, high-quality, year-round, safe, and stable factory production of leafy vegetables by precisely controlling all environmental factors such as light, temperature, humidity, CO2 concentration, and nutrient solution. This is achieved through artificial control of all environmental factors, including light, temperature, humidity, CO2 concentration, and nutrient solution. Among all environmental control factors in plant factories, light is the most frequently controlled, has the most significant impact, and accounts for the highest proportion of energy consumption. It directly determines the photosynthetic electron transfer efficiency, chlorophyll synthesis level, carbon and nitrogen metabolism intensity, biomass accumulation rate, nutritional quality formation, and overall yield composition of leafy vegetables. At the same time, the energy consumption of the supplemental lighting system usually accounts for more than 60% of the total operating energy consumption of a plant factory. The rationality of light parameters and the intelligence of the control mode directly affect the uniformity of leafy vegetable growth, marketability, production cycle, and overall operating costs. It is the core technical link that determines whether a plant factory can operate efficiently, energy-savingly, and stably.

[0003] Existing methods for supplemental lighting control in plant factories generally suffer from numerous significant drawbacks. Their overall level of intelligence and precision is low, and their methods for sensing growth status are relatively simplistic, mostly relying on environmental parameter monitoring or ordinary monocular visual detection. This fails to simultaneously acquire multi-dimensional growth information such as leafy vegetable morphology, physiology, and spectrum, resulting in low sensing accuracy and significant lag. Supplemental lighting strategies often employ traditional models with fixed light intensity, fixed spectrum, and timed switching, failing to dynamically adapt to the real-time growth status and physiological needs of leafy vegetables at different growth stages. This easily leads to insufficient or excessive supplemental lighting, causing energy waste, high operating energy consumption, and significant fluctuations in leafy vegetable yield and quality, resulting in instability. Furthermore, these systems generally lack growth trend prediction. The existing light control mechanisms are primarily passive, failing to provide forward-looking and predictive light optimization. Furthermore, most systems lack real-time identification of abnormal growth states and emergency supplemental lighting intervention, making it difficult to detect and address common anomalies such as nitrogen deficiency, excessive vegetative growth, and early-stage diseases, which can easily lead to widespread poor growth and even yield reduction. The intelligent algorithm models used are generally large and computationally expensive, making it difficult to achieve real-time, efficient operation on edge computing devices, resulting in high costs for large-scale deployment and promotion. All these issues contribute to the overall low energy efficiency of plant factory supplemental lighting systems, poor uniformity in leafy vegetable growth, and insufficient intelligent control throughout the production process, severely hindering the efficient, stable, and low-cost operation and large-scale application of plant factories. Summary of the Invention

[0004] (a) Technical problems to be solved To address the shortcomings of existing technologies, this invention provides a method for sensing the growth status of leafy vegetables and adaptive supplemental lighting in plant factories. It has the advantages of multimodal sensing, full-cycle regulation, predictive optimization, anomaly intervention, and lightweight operation. It solves the problems of traditional plant factories, such as single supplemental lighting sensing, rigid regulation, delayed response, high energy consumption, untimely anomaly handling, and difficulty in deploying intelligent systems.

[0005] (II) Technical Solution To achieve the aforementioned objectives of multimodal sensing, full-cycle regulation, predictive optimization, anomaly intervention, and lightweight operation, this invention provides the following technical solution: a method for sensing the growth status of leafy vegetables in a plant factory and adaptive supplemental lighting regulation, comprising the following steps: Step S1: Multimodal data acquisition; The system simultaneously acquires hyperspectral images, three-dimensional morphological images, chlorophyll fluorescence parameters, and environmental temperature, humidity, and CO2 concentration data of leafy vegetables using a hyperspectral imager, RGB-D camera, chlorophyll fluorescence meter, and environmental sensors. Step S2: Multimodal data fusion and growth status assessment; A multimodal feature fusion network driven by an attention mechanism is used to fuse the collected hyperspectral features, morphological features and physiological features, output key growth indicators of leafy vegetables and generate a comprehensive score of leafy vegetable growth status. Step S3: Intelligent identification of growth stages and prediction of growth trends; The fused growth features are input into a lightweight Transformer classification model to automatically identify the growth stage of leafy vegetables; at the same time, a leafy vegetable growth prediction model is established based on a long short-term memory network to predict the growth trend and photosynthetic demand changes of leafy vegetables in the next 24 hours. Step S4: Adaptive optimization of multi-dimensional supplementary lighting parameters; Based on the identified growth stage, real-time growth status score, and growth trend prediction results, combined with environmental factor data, an improved particle swarm optimization algorithm is used to simultaneously optimize the parameters of the three dimensions of supplementary lighting: spectral composition, light intensity, and photoperiod, to generate the optimal supplementary lighting control scheme. Step S5: Execution and closed-loop feedback of the supplementary lighting system; The optimal supplemental lighting control scheme is sent to the LED supplemental lighting array, and independent and precise control of different spectral channels is achieved through PWM dimming technology; at the same time, the growth status data of leafy vegetables is continuously collected, the control effect is evaluated in real time, and the supplemental lighting parameters are dynamically adjusted to form a closed-loop feedback control system. Step S6: Identification of abnormal growth states and emergency control; Based on the growth indicators output in step S2, abnormal growth status of leafy vegetables is identified and corresponding emergency supplemental lighting strategies are executed. The corresponding emergency strategies for abnormal triggering are: increasing blue light by 5%-10% and enhancing light intensity by 20% for nitrogen deficiency; increasing blue light by 10%-15% and reducing far-red light to below 2% for excessive growth; and increasing ultraviolet light to 5%-8% and shortening the photoperiod by 2 hours for early diseases. At the same time, the system pushes early warning information and generates anomaly handling logs.

[0006] Preferably, in step S1, the hyperspectral imager has a spectral range of 400-1000nm and a spectral resolution of 2nm, and collects data every 30 minutes; the RGB-D camera collects data every hour; the chlorophyll fluorometer collects the Fv / Fm value and ΦPSII value of the leaves every 2 hours; and the environmental sensor collects environmental data every 5 minutes.

[0007] Preferably, in step S2, the multimodal feature fusion network extracts the spectral features of the hyperspectral image and the morphological features of the RGB-D image through a convolutional neural network, while a fully connected network extracts the numerical features of chlorophyll fluorescence parameters and environmental data. Then, channel attention mechanism and spatial attention mechanism are introduced to adaptively weight and fuse the features of different modalities. Finally, six key growth indicators are output through a regression layer.

[0008] Preferably, in step S3, the lightweight Transformer classification model adopts the MobileViT architecture, with fewer than 5M model parameters and an inference speed greater than 30fps. The long short-term memory network growth prediction model takes the growth data and environmental data of the past 72 hours as input and outputs the predicted values ​​of leafy vegetable growth indicators every 6 hours for the next 24 hours.

[0009] Preferably, the spectral optimization objective for different growth stages in step S4 is: Germination period: Blue light accounts for 45%-55%, red light accounts for 40%-50%, and far-red light accounts for 5%-10%; Seedling stage: Blue light accounts for 30%-40%, red light accounts for 55%-65%, and far-red light accounts for 5%-10%; Lotus stage: Blue light accounts for 20%-30%, red light accounts for 65%-75%, and far-red light accounts for 5%-10%; Mature stage: Blue light accounts for 35%-45%, red light accounts for 50%-60%, and ultraviolet light accounts for 2%-5%.

[0010] Preferably, the improved particle swarm optimization algorithm in step S4 introduces adaptive inertia weights and learning factors, and also incorporates energy consumption constraints. The optimization objective function is:

[0011] in, For the expected yield of leafy vegetables, Rate the quality of leafy vegetables. To compensate for the energy consumption of the supplemental lighting system, These are the weighting coefficients.

[0012] Preferably, in step S5, the LED supplementary lighting array consists of four independently controlled spectral channels: ultraviolet, blue, red, and far-red. Each channel has a dimming accuracy of 0.1% and a light intensity adjustment range of 0-1000 μmol / (m²). 2 •s) Preferably, the closed-loop feedback control in S5 adopts a three-level gradient adjustment mechanism, including: A rapid assessment is conducted every 2 hours. If the overall growth status score deviates from the target value by within ±5 points, only the supplementary light intensity is slightly adjusted, with the adjustment not exceeding 10% of the current light intensity. A mid-term evaluation should be conducted every 12 hours. If the deviation is between ±5 and 15 minutes, the composition and intensity of the supplementary light spectrum should be adjusted simultaneously, with the adjustment range not exceeding 15% of the current spectrum ratio and 20% of the current light intensity. A comprehensive evaluation is conducted every 24 hours. If the deviation exceeds ±15 points, the particle swarm optimization algorithm is rerun to simultaneously optimize the parameters in the three dimensions of spectral composition, light intensity, and photoperiod.

[0013] Preferably, in step S1, the multimodal data acquisition adopts a single-plant precise perception mode. The individual boundary of each leafy vegetable is identified by the instance segmentation algorithm of the RGB-D camera. The morphological features and spectral features of the single leafy vegetable are extracted respectively to generate a single-plant growth status score. Differentiated supplemental lighting control is performed on single leafy vegetables in different growth states.

[0014] (III) Beneficial Effects Compared with existing technologies, this invention provides a method for sensing the growth status of leafy vegetables in plant factories and for adaptive supplemental lighting regulation, which has the following beneficial effects: 1. The plant factory leafy vegetable growth status perception and adaptive supplemental lighting regulation method, through the fusion of hyperspectral, RGB-D, and chlorophyll fluorescence multimodal data and attention mechanism feature weighting, can achieve comprehensive perception of leafy vegetable morphology, physiology, and spectral information, and significantly improve the accuracy of growth status assessment.

[0015] 2. The plant factory's leafy vegetable growth status perception and adaptive supplemental lighting regulation method, through growth stage identification + LSTM trend prediction + improved particle swarm optimization, can achieve synchronous adaptive regulation of spectrum, light intensity, and photoperiod. At the same time, it improves leafy vegetable yield and quality, significantly reduces supplemental lighting energy consumption, and makes regulation more intelligent and energy-efficient.

[0016] 3. The plant factory's leafy vegetable growth status perception and adaptive supplemental lighting control method, through a three-level gradient closed-loop feedback + abnormal growth identification and emergency supplemental lighting, can quickly respond to growth deviations and automatically correct them. At the same time, it can provide real-time early warning and intervention for nutrient deficiency, excessive growth, and early disease abnormalities, which greatly improves the production stability and yield of plant factories.

[0017] 4. The plant factory leafy vegetable growth status perception and adaptive supplemental lighting control method adopts the lightweight MobileViT model, which can run in real time at the edge without relying on cloud computing power. It has low deployment cost and fast response speed, making it more suitable for large-scale application in plant factories. Attached Figure Description

[0018] Figure 1 This is a schematic block diagram of the method structure of the present invention; Figure 2 This is a schematic flowchart illustrating the operation of the method of the present invention; Figure 3 This is a schematic block diagram of the multimodal data acquisition component in the method of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Please see Figure 1-3 A method for sensing the growth status of leafy vegetables in a plant factory and for adaptive supplemental lighting regulation includes the following steps: Step S1: Multimodal data acquisition; The system simultaneously acquires hyperspectral images, three-dimensional morphological images, chlorophyll fluorescence parameters, and environmental temperature, humidity, and CO2 concentration data of leafy vegetables using a hyperspectral imager, RGB-D camera, chlorophyll fluorescence meter, and environmental sensors. Step S2: Multimodal data fusion and growth status assessment; A multimodal feature fusion network driven by an attention mechanism is used to fuse the collected hyperspectral features, morphological features and physiological features, output key growth indicators of leafy vegetables and generate a comprehensive score of leafy vegetable growth status. Step S3: Intelligent identification of growth stages and prediction of growth trends; The fused growth features are input into a lightweight Transformer classification model to automatically identify the growth stage of leafy vegetables; at the same time, a leafy vegetable growth prediction model is established based on a long short-term memory network to predict the growth trend and photosynthetic demand changes of leafy vegetables in the next 24 hours. Step S4: Adaptive optimization of multi-dimensional supplementary lighting parameters; Based on the identified growth stage, real-time growth status score, and growth trend prediction results, combined with environmental factor data, an improved particle swarm optimization algorithm is used to simultaneously optimize the parameters of the three dimensions of supplementary lighting: spectral composition, light intensity, and photoperiod, to generate the optimal supplementary lighting control scheme. Step S5: Execution and closed-loop feedback of the supplementary lighting system; The optimal supplemental lighting control scheme is sent to the LED supplemental lighting array, and independent and precise control of different spectral channels is achieved through PWM dimming technology; at the same time, the growth status data of leafy vegetables is continuously collected, the control effect is evaluated in real time, and the supplemental lighting parameters are dynamically adjusted to form a closed-loop feedback control system. Step S6: Identification of abnormal growth states and emergency control; Based on the growth indicators output in step S2, abnormal growth status of leafy vegetables is identified and corresponding emergency supplemental lighting strategies are executed. The corresponding emergency strategies for abnormal triggering are: increasing blue light by 5%-10% and enhancing light intensity by 20% for nitrogen deficiency; increasing blue light by 10%-15% and reducing far-red light to below 2% for excessive growth; and increasing ultraviolet light to 5%-8% and shortening the photoperiod by 2 hours for early diseases. At the same time, the system pushes early warning information and generates anomaly handling logs.

[0021] In the implementation of the case, in step S1, the hyperspectral imager has a spectral range of 400-1000nm and a spectral resolution of 2nm, and collects data every 30 minutes. The RGB-D camera collects data every hour. The chlorophyll fluorometer collects the Fv / Fm value and ΦPSII value of the leaves every 2 hours. The environmental sensor collects environmental data every 5 minutes. The acquisition frequency and equipment parameters are the optimal combination determined through multiple experiments. This combination can reduce redundant data transmission and processing pressure, and lower the overall power consumption and hardware load of the system, while ensuring the real-time and completeness of growth status perception. Specifically, the hyperspectral imager covers the visible to near-infrared bands, effectively retrieving key physiological indicators such as leaf chlorophyll content, water content, nitrogen level, and soluble solids; the RGB-D camera simultaneously outputs two-dimensional color images and three-dimensional point cloud data for accurate calculation of morphological parameters such as plant height, leaf area, leaf spread, and plant volume; the chlorophyll fluorometer uses non-contact in-situ measurement, with Fv / Fm representing the potential maximum photosynthetic efficiency of the leaf and ΦPSII representing the actual photosynthetic efficiency of the leaf, directly reflecting the health status of leafy vegetables; and environmental sensors collect real-time data on temperature, humidity, and CO2 concentration within the cultivation area, providing stable environmental boundary conditions for subsequent supplemental lighting strategy optimization.

[0022] In the implementation of the case, in step S2, the multimodal feature fusion network extracts the spectral features of the hyperspectral image and the morphological features of the RGB-D image through the convolutional neural network, while the fully connected network extracts the numerical features of chlorophyll fluorescence parameters and environmental data. Then, channel attention mechanism and spatial attention mechanism are introduced to adaptively weight and fuse the features of different modalities. Finally, six key growth indicators are output through the regression layer. Among them, the introduction of the attention mechanism enables the model to automatically focus on features that have a more significant impact on the growth state, suppress background noise, invalid regions and redundant information interference, and significantly improve the prediction accuracy of growth indicators and the reliability of the overall score. Specifically, the six key growth indicators are, in order: plant height, leaf area, aboveground fresh weight, relative chlorophyll content, net photosynthetic rate, and plant nitrogen accumulation. The channel attention mechanism is used to automatically enhance the weights of feature channels that are strongly correlated with physiological state in hyperspectral and fluorescence data. The spatial attention mechanism is used to accurately locate the effective area of ​​the leaf and eliminate background interference such as substrate, cultivation rack, and shadow. The prediction error of the integrated growth status score after fusion is controlled within 3%, which can truly and stably reflect the actual growth level of leafy vegetables.

[0023] In the case implementation, the lightweight Transformer classification model in step S3 adopts the MobileViT architecture, with a model parameter count of less than 5M and an inference speed of more than 30fps. The long short-term memory network growth prediction model takes the growth data and environmental data of the past 72 hours as input and outputs the predicted values ​​of leafy vegetable growth indicators every 6 hours in the next 24 hours. The lightweight architecture can run independently on the edge computing module without cloud dependency, meeting the real-time control needs of plant factories; the LSTM network is good at processing time-series data and can accurately capture the dynamic changes between leafy vegetable growth and environmental factors. Specifically, MobileViT combines the advantages of local feature extraction from convolutional neural networks with the global modeling capabilities of Transformer to quickly and accurately distinguish the four key growth stages of leafy vegetables: germination, seedling, rosette, and maturity. The LSTM model takes growth indicators, light environment parameters, and environmental factors over a continuous 72-hour period as time-series inputs and outputs predicted growth trends and photosynthetic requirements every 6 hours for the next 24 hours, with an overall prediction accuracy of over 92%, providing a basis for proactive supplemental lighting.

[0024] In the case implementation, the spectral optimization objectives for different growth stages in step S4 are as follows: Germination period: Blue light accounts for 45%-55%, red light accounts for 40%-50%, and far-red light accounts for 5%-10%; Seedling stage: Blue light accounts for 30%-40%, red light accounts for 55%-65%, and far-red light accounts for 5%-10%; Lotus stage: Blue light accounts for 20%-30%, red light accounts for 65%-75%, and far-red light accounts for 5%-10%; Mature stage: Blue light accounts for 35%-45%, red light accounts for 50%-60%, and ultraviolet light accounts for 2%-5%; The spectral ratios for each stage are designed differently based on the photosynthetic characteristics, morphological development requirements, and nutrient accumulation patterns of leafy vegetables at different growth stages, to achieve optimal light environment matching throughout the entire cycle.

[0025] Specifically, increasing the proportion of blue light during the germination period helps break seed dormancy, promotes radicle development, and strengthens seedlings; a moderate proportion of blue light during the seedling stage promotes leaf expansion, chlorophyll synthesis, and root growth; increasing the proportion of red light during the rosette stage enhances photosynthetic efficiency and accelerates biomass accumulation; and adding an appropriate amount of ultraviolet light during the maturity stage can increase vitamin C and soluble sugar content, reduce nitrate accumulation, and significantly improve the quality and marketability of leafy vegetables.

[0026] In the case implementation, the improved particle swarm optimization algorithm in step S4 introduces adaptive inertia weights and a learning factor, while also incorporating energy consumption constraints. The optimization objective function is:

[0027] in, For the expected yield of leafy vegetables, Rate the quality of leafy vegetables. To compensate for the energy consumption of the supplemental lighting system, These are the weighting coefficients; Specifically, adaptive weights can improve the algorithm's global optimization ability and convergence speed, avoiding premature convergence; energy consumption constraints achieve synergistic optimization of high output, high quality, and low consumption, and the weight coefficients can be flexibly set according to production targets, enabling a production-first mode. =0.5、 =0.3、 =0.2; Quality-first mode =0.3、 =0.5、 =0.2; Equilibrium mode =0.4、 =0.4、 =0.2; the algorithm can complete the optimization of three-dimensional parameters of spectrum, light intensity, and light period within 1 second.

[0028] In the implementation of this case, the LED supplementary lighting array in step S5 consists of four independently controlled spectral channels: ultraviolet, blue, red, and far-red. The dimming accuracy of each channel is 0.1%, and the light intensity adjustment range is 0-1000 μmol / (m²). 2 •s); Among them, the four-channel independent adjustable structure can realize any target spectrum combination, and the high-precision dimming ensures stable and uniform light output without obvious fluctuations and flicker, thus avoiding interference with the photosynthetic system of leafy vegetables. Specifically, the ultraviolet light channel has a wavelength range of 380-400nm, the blue light channel has a wavelength range of 430-470nm, the red light channel has a wavelength range of 630-670nm, and the far-red light channel has a wavelength range of 720-740nm, all of which are sensitive wavelengths for leafy vegetables' photosynthesis. The PWM digital dimming technology has a short response time and high control precision, which can meet the precise lighting needs of leafy vegetables throughout their entire growth period and in multiple scenarios.

[0029] In the case implementation, the closed-loop feedback control in S5 adopts a three-level gradient adjustment mechanism, including: A rapid assessment is conducted every 2 hours. If the overall growth status score deviates from the target value by within ±5 points, only the supplementary light intensity is slightly adjusted, with the adjustment not exceeding 10% of the current light intensity. A mid-term evaluation should be conducted every 12 hours. If the deviation is between ±5 and 15 minutes, the composition and intensity of the supplementary light spectrum should be adjusted simultaneously, with the adjustment range not exceeding 15% of the current spectrum ratio and 20% of the current light intensity. A comprehensive evaluation is conducted every 24 hours. If the deviation exceeds ±15 points, the particle swarm optimization algorithm is rerun to simultaneously optimize the parameters in the three dimensions of spectral composition, light intensity, and light period. Among them, the three-level gradient regulation mechanism takes into account both system response speed and operational stability, avoids fluctuations in leafy vegetable growth caused by frequent and drastic parameter adjustments, and achieves stable, efficient, and continuous optimal regulation. Specifically, rapid assessment is used for real-time correction of minor deviations; mid-term assessment is used for coordinated adjustment of moderate deviations; and comprehensive assessment is used for global re-optimization when the growth status deviates significantly from the target, ensuring that leafy vegetables are always in the best light environment, and significantly improving growth consistency and yield stability.

[0030] In the implementation of the case, the multimodal data acquisition in step S1 adopts the single-plant precise perception mode. The instance segmentation algorithm of the RGB-D camera is used to identify the individual boundary of each leafy vegetable, extract the morphological features and spectral features of the single leafy vegetable, generate a single plant growth status score, and perform differentiated supplemental lighting control on single leafy vegetables in different growth statuses. Among them, the precise perception at the single-plant level breaks through the limitations of the traditional average perception of a group; Specifically, the instance segmentation algorithm can locate, segment, and evaluate each plant individually. For plants with weak growth, it automatically increases the light intensity and blue light ratio to promote rapid recovery. For plants with excessive growth, it appropriately reduces the light intensity and total energy consumption to achieve refined management. Overall, it can improve the uniformity of the cultivation population and increase the yield rate by more than 15%, resulting in a significant improvement in economic benefits.

[0031] In summary, this plant factory method for sensing and adaptive supplemental lighting of leafy vegetable growth status, through the fusion of hyperspectral, RGB-D, and chlorophyll fluorescence multimodal data and attention mechanism feature weighting, can achieve comprehensive sensing of leafy vegetable morphology, physiology, and spectral information, significantly improving the accuracy of growth status assessment. By identifying growth stages, using LSTM trend prediction, and improving particle swarm optimization, it can achieve synchronous adaptive control of spectrum, light intensity, and photoperiod. At the same time, it improves leafy vegetable yield and quality, significantly reduces supplemental lighting energy consumption, and makes the control more intelligent and energy-efficient. It solves the problems of traditional supplemental lighting modes, such as single sensing, lagging control, fixed and rigid light parameters, and inability to match the real-time growth needs of leafy vegetables.

[0032] Furthermore, through a three-level gradient closed-loop feedback system, abnormal growth identification, and emergency supplemental lighting, it can quickly respond to growth deviations and automatically correct them. At the same time, it provides real-time early warning and intervention for nutrient deficiencies, excessive growth, and early disease anomalies, significantly improving the production stability and yield rate of plant factories. By adopting the lightweight MobileViT model, it can run in real time at the edge without relying on cloud computing power. It has low deployment costs and fast response speed, making it more suitable for large-scale application in plant factories. It solves the problems of existing intelligent control systems, such as high computing power requirements, high deployment costs, difficulty in real-time stable operation in the factory, and inability to quickly deal with abnormal growth.

[0033] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0034] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for sensing the growth status of leafy vegetables and adaptively regulating supplemental lighting in plant factories, characterized by: Includes the following steps: Step S1: Multimodal data acquisition; The system simultaneously acquires hyperspectral images, three-dimensional morphological images, chlorophyll fluorescence parameters, and environmental temperature, humidity, and CO2 concentration data of leafy vegetables using a hyperspectral imager, RGB-D camera, chlorophyll fluorescence meter, and environmental sensors. Step S2: Multimodal data fusion and growth status assessment; A multimodal feature fusion network driven by an attention mechanism is used to fuse the collected hyperspectral features, morphological features and physiological features, output key growth indicators of leafy vegetables and generate a comprehensive score of leafy vegetable growth status. Step S3: Intelligent identification of growth stages and prediction of growth trends; The fused growth features are input into a lightweight Transformer classification model to automatically identify the growth stage of leafy vegetables; at the same time, a leafy vegetable growth prediction model is established based on a long short-term memory network to predict the growth trend and photosynthetic demand changes of leafy vegetables in the next 24 hours. Step S4: Adaptive optimization of multi-dimensional supplementary lighting parameters; Based on the identified growth stage, real-time growth status score, and growth trend prediction results, combined with environmental factor data, an improved particle swarm optimization algorithm is used to simultaneously optimize the parameters of the three dimensions of supplementary lighting: spectral composition, light intensity, and photoperiod, to generate the optimal supplementary lighting control scheme. Step S5: Execution and closed-loop feedback of the supplementary lighting system; The optimal supplemental lighting control scheme is sent to the LED supplemental lighting array, and independent and precise control of different spectral channels is achieved through PWM dimming technology; at the same time, the growth status data of leafy vegetables is continuously collected, the control effect is evaluated in real time, and the supplemental lighting parameters are dynamically adjusted to form a closed-loop feedback control system. Step S6: Identification of abnormal growth states and emergency control; Based on the growth indicators output in step S2, abnormal growth status of leafy vegetables is identified and corresponding emergency supplemental lighting strategies are executed. The corresponding emergency strategies for abnormal triggering are: increasing blue light by 5%-10% and enhancing light intensity by 20% for nitrogen deficiency; increasing blue light by 10%-15% and reducing far-red light to below 2% for excessive growth; and increasing ultraviolet light to 5%-8% and shortening the photoperiod by 2 hours for early diseases. At the same time, the system pushes early warning information and generates anomaly handling logs.

2. The method for sensing the growth status of leafy vegetables in a plant factory and adaptively regulating supplemental lighting according to claim 1, characterized in that: In step S1, the hyperspectral imager has a spectral range of 400-1000nm and a spectral resolution of 2nm, and collects data every 30 minutes. The RGB-D camera collects data every hour. The chlorophyll fluorometer collects the Fv / Fm value and ΦPSII value of the leaves every 2 hours. The environmental sensor collects environmental data every 5 minutes.

3. The method for sensing the growth status of leafy vegetables in a plant factory and adaptively regulating supplemental lighting according to claim 1, characterized in that: In step S2, the multimodal feature fusion network extracts the spectral features of the hyperspectral image and the morphological features of the RGB-D image through a convolutional neural network. At the same time, the fully connected network extracts the numerical features of chlorophyll fluorescence parameters and environmental data. Then, channel attention mechanism and spatial attention mechanism are introduced to adaptively weight and fuse the features of different modalities. Finally, six key growth indicators are output through a regression layer.

4. The method for sensing the growth status of leafy vegetables in a plant factory and adaptively regulating supplemental lighting according to claim 1, characterized in that: In step S3, the lightweight Transformer classification model adopts the MobileViT architecture, with fewer than 5M model parameters and an inference speed greater than 30fps. The long short-term memory network growth prediction model takes the growth data and environmental data of the past 72 hours as input and outputs the predicted values ​​of leafy vegetable growth indicators every 6 hours for the next 24 hours.

5. The method for sensing and adaptively adjusting the growth status of leafy vegetables in a plant factory according to claim 1, characterized in that: The spectral optimization objectives for different growth stages in step S4 are: Germination period: Blue light accounts for 45%-55%, red light accounts for 40%-50%, and far-red light accounts for 5%-10%; Seedling stage: Blue light accounts for 30%-40%, red light accounts for 55%-65%, and far-red light accounts for 5%-10%; Lotus stage: Blue light accounts for 20%-30%, red light accounts for 65%-75%, and far-red light accounts for 5%-10%; Mature stage: Blue light accounts for 35%-45%, red light accounts for 50%-60%, and ultraviolet light accounts for 2%-5%.

6. The method for sensing the growth status of leafy vegetables in a plant factory and adaptively regulating supplemental lighting according to claim 1, characterized in that: The improved particle swarm optimization algorithm in step S4 introduces adaptive inertia weights and learning factors, and also incorporates energy consumption constraints. The optimization objective function is: in, For the expected yield of leafy vegetables, Rate the quality of leafy vegetables. To compensate for the energy consumption of the supplemental lighting system, These are the weighting coefficients.

7. The method for sensing the growth status of leafy vegetables in a plant factory and adaptively regulating supplemental lighting according to claim 1, characterized in that: In step S5, the LED supplementary lighting array consists of four independently controlled spectral channels: ultraviolet, blue, red, and far-red. Each channel has a dimming accuracy of 0.1% and a light intensity adjustment range of [missing information]. .

8. The method for sensing and adaptively adjusting the growth status of leafy vegetables in a plant factory according to claim 1, characterized in that: The closed-loop feedback control in S5 adopts a three-level gradient adjustment mechanism, including: A rapid assessment is conducted every 2 hours. If the overall growth status score deviates from the target value by within ±5 points, only the supplementary light intensity is slightly adjusted, with the adjustment not exceeding 10% of the current light intensity. A mid-term evaluation should be conducted every 12 hours. If the deviation is between ±5 and 15 minutes, the composition and intensity of the supplementary light spectrum should be adjusted simultaneously, with the adjustment range not exceeding 15% of the current spectrum ratio and 20% of the current light intensity. A comprehensive evaluation is conducted every 24 hours. If the deviation exceeds ±15 points, the particle swarm optimization algorithm is rerun to simultaneously optimize the parameters in the three dimensions of spectral composition, light intensity, and photoperiod.

9. The method for sensing the growth status of leafy vegetables in a plant factory and adaptively regulating supplemental lighting according to claim 1, characterized in that: In step S1, the multimodal data acquisition adopts a single-plant precise perception mode. The instance segmentation algorithm of the RGB-D camera is used to identify the individual boundary of each leafy vegetable, extract the morphological and spectral features of the single leafy vegetable, generate a single plant growth status score, and perform differentiated supplemental lighting control on single leafy vegetables in different growth states.

Images