LED dynamic light formula regulation method and system suitable for industrial cultivation of edible fungi

Abstract

The application discloses a kind of LED dynamic light formula regulation and control method and system suitable for edible mushroom factory cultivation.Method includes:S1 global environment perception and heterogeneous data acquisition, wherein acquisition frequency follows light efficiency quality dynamic adjustment;S2 multi-modal data fusion predicts bacterial population growth trend and confidence score, and confidence reversely drives S1 sampling phase correction;S3 adopts fuzzy membership degree decision light formula, and pre-trigger instruction is carried based on spatial difference gradual change time length coefficient;S4 five-channel LED light mixing drive, support cross-stage backtracking correction;S5 closed-loop feedback correction, and establish cross-batch history database realizes gradual learning.System includes perception layer, control layer, execution layer and feedback layer.The application realizes the full-chain intelligentization of perception-prediction-decision-execution-evaluation-learning by multiple linkage synergistic relationship between the five steps, greatly improves the yield, quality and self-adaptive capacity of edible mushroom factory cultivation.

Description

Technical Field

[0001] This application relates to the field of industrialized cultivation of edible fungi and intelligent optoelectronic control technology, specifically to an LED dynamic light formula regulation method and system suitable for industrialized cultivation of edible fungi. Background Technology

[0002] Edible fungi, as an important agricultural product, have seen a continuous expansion in the scale of factory cultivation in recent years. In existing factory production processes, light is one of the key environmental factors affecting the growth and development of edible fungi. Different wavelengths of light have differentiated biological effects on mycelial growth, primordium formation, and fruiting body development. Currently, LED supplemental lighting technology is widely used in edible fungi cultivation environments due to its advantages such as low energy consumption, long lifespan, and adjustable spectrum. Existing LED supplemental lighting systems are typically equipped with photosensors, temperature and humidity sensors, and imaging equipment, supplementing the cultivation environment with light through preset programs or fixed spectral modes. These systems can collect environmental data and drive the LED light source to output a specific spectrum; some systems can also switch the light cycle according to a preset schedule to achieve basic supplemental lighting functions.

[0003] However, in existing technologies, the light regulation process often lacks precise perception and coordinated response to the real-time growth status of the microbial community, and there is a lack of temporal phase coordination between spectral regulation execution and image data acquisition. This makes image-based growth status determination susceptible to inconsistencies caused by light source switching, making it difficult to achieve dynamic and refined regulation for different growth stages and spatial differences, thus affecting the stability of edible fungi yield and quality. Summary of the Invention

[0004] This application provides an LED dynamic light formula control method and system suitable for industrialized cultivation of edible fungi, which can solve the technical problems of extensive light control, lack of real-time growth status perception and linkage response of fungi in the prior art, and lack of temporal coordination between spectral control and image acquisition.

[0005] To achieve the above objectives, this application provides the following technical solution: The first aspect of this application provides a method for controlling LED dynamic light formulation suitable for industrialized cultivation of edible fungi, comprising the following steps: Step S1: Collect multispectral image data of light intensity, temperature and humidity, CO2 concentration and edible fungi in each cultivation unit in the cultivation environment, and synchronize the acquisition of multispectral images with the adjustment of LED light source in time phase. Step S2: Input the collected multi-type data into the multimodal time series fusion model, and output the prediction results of the microbial community growth status of each cultivation unit and the confidence score of the current growth stage; Step S3: Based on the confidence score, the current growth stage or stage transition interval is determined by the fuzzy membership method, and the corresponding light formula parameters are output; when any cultivation unit is detected to have entered the preset key growth stage, a linkage pre-trigger command is issued to other units in the same stage within the cultivation layer. Step S4: Decompose the light formula parameters into driving signals for multiple independent spectral channels, drive the LED light source to output the corresponding spectrum, and when a linkage pre-trigger command is received, control multiple cultivation units to switch or transition the light formula simultaneously in a synchronous manner. Step S5: Collect the spectral parameters of the actual output of the LED light source, calculate the difference between it and the target light formula, and perform feedback correction on the fuzzy membership parameter in step S3 based on the difference and the prediction result of the microbial community growth status. Among them, the above steps S1 to S5 are set with cross-step linkage and coordination relationship, including: at least one output result of the subsequent step is used as the adjustment input of the preceding step to form a two-way or cross-step closed-loop control link.

[0006] In an optional embodiment, the time phase synchronization alignment in step S1 is achieved by generating the exposure synchronization signal of the multispectral imager and the PWM drive signal of the LED light source by the same timing controller; and when the difference calculated in step S5 is lower than a preset threshold, the acquisition frequency of the multispectral image in step S1 is automatically increased.

[0007] In an optional embodiment, in step S2, when the confidence score is lower than the first confidence threshold multiple times consecutively, a fine-tuning correction is triggered on the sampling phase difference between image acquisition and LED driving in step S1 until the confidence score recovers to above the threshold.

[0008] In an optional embodiment, the linkage pre-triggering instruction in step S3 carries a gradual duration coefficient, which is dynamically calculated based on the standard deviation of the confidence scores of all linkage units in the same layer. The larger the standard deviation, the longer the gradual duration, so as to adapt to the growth differences between different units.

[0009] In an optional embodiment, the multiple independent spectral channels mentioned in step S4 include five channels: red light, blue light, green light, near-infrared light, and far-infrared light. Furthermore, when the difference reported in step S5 exceeds the second preset threshold, step S4 automatically retrieves the historical light formula parameters of the previous stage of the cultivation unit, performs weighted fusion with the current target light formula parameters, and re-executes the drive to form a cross-stage retrospective correction.

[0010] In an optional embodiment, step S5 further updates the difference and growth status prediction results to a cross-batch historical database, which is used to perform offline optimization of the light formula parameter library for the same variety and the same growth stage, so as to realize progressive learning between multiple cultivation batches.

[0011] The second aspect of this application provides an LED dynamic light formula control system suitable for the industrialized cultivation of edible fungi, used to implement the method described in any one of technical solutions 1 to 6 of this application, the system comprising: Sensing layer: includes a photosensitive sensor array, a temperature and humidity sensor group, a CO2 sensor group and a multispectral imager group deployed on each cultivation layer rack. The acquisition trigger signal of the multispectral imager group and the PWM drive signal of the LED light source are generated by the same timing synchronization controller. The control layer includes a first controller deployed at the edge and a second controller deployed on the cloud platform. The first controller is used to generate the drive signal and perform some online corrections in step S4, and the second controller is used to perform situation prediction in step S2 and optical formula decision-making in step S3. The first controller and the second controller use a breakpoint resume communication protocol. Execution layer: includes a five-channel independently dimmable LED array module, each channel equipped with an independent constant current drive circuit and current feedback loop; Feedback layer: includes a light field calibration module and a growth assessment module, which are used to correct the LED driver output and generate a growth status assessment report, respectively.

[0012] In one optional embodiment, the multispectral imager group includes at least four narrowband filters with center wavelengths set around 450nm, 540nm, 660nm and 730nm, respectively.

[0013] In one optional embodiment, the main wavelength of the red light of the five-channel LED array module is 660nm, the main wavelength of the blue light is 450nm, the main wavelength of the green light is 520nm, the main wavelength of the near-infrared light is 780nm, and the main wavelength of the far-infrared light is 730nm; the LED chips of each channel are arranged in a staggered triangular grid on the substrate, and a high reflectivity isolation grille is provided between the chips of different channels.

[0014] Beneficial effects This application provides a method and system for LED dynamic light formula regulation suitable for industrialized cultivation of edible fungi. This scheme collects light intensity, temperature, humidity, CO2 concentration, and multispectral image data of edible fungi in the cultivation environment, as well as data from each cultivation unit. The acquisition of multispectral images and the regulation of the LED light source are synchronized in time phase, ensuring that the image data accurately reflects the growth status under current light conditions and eliminating measurement errors caused by phase misalignment during light source switching. The collected data is then input into a multimodal time-series fusion model, which outputs predictions of the fungal community growth status of each cultivation unit and a confidence score for the current growth stage. Based on this confidence score, a fuzzy membership method is used to determine the current growth stage or transition interval and output the corresponding light formula parameters. This achieves a shift from experience-based regulation to data-driven intelligent predictive regulation, effectively avoiding light stress caused by sudden changes in growth stage judgment. Furthermore, when any cultivation unit is detected to have entered a preset critical growth stage... At the same time, a linkage pre-trigger command is issued to other units in the same stage within the cultivation layer, and the light formula parameters are decomposed into driving signals for multiple independent spectral channels. This synchronously controls multiple cultivation units to switch or transition light formulas simultaneously, thereby achieving coordinated regulation in the spatial dimension and ensuring the consistency of growth of cultivation units in the same layer. Subsequently, the spectral parameters of the actual output of the LED light source are collected, and the difference between them and the target light formula is calculated. Based on the difference and the predicted results of the microbial community growth status, the fuzzy membership parameters are corrected. Furthermore, a cross-step linkage and coordination relationship is established between each step, so that the output result of at least one subsequent step serves as the adjustment input of the preceding step, forming a two-way or cross-step closed-loop control link. Therefore, this effectively solves the problems of lack of real-time growth status perception and linkage response in the light regulation process of existing technologies, as well as the lack of temporal coordination between spectral regulation and image acquisition. This improves the regulation accuracy, system adaptability, and yield and quality stability of the final product in the industrialized cultivation of edible fungi.

[0015] In summary, this application constructs a closed-loop control system covering the entire process of perception, decision-making, execution, and feedback, and utilizes a deep linkage and collaboration mechanism across steps to achieve the fusion processing and dynamic optimization of multi-dimensional data. This not only ensures precise control within a single cultivation cycle but also possesses the ability to self-evolve across batches, forming a logically rigorous and stable intelligent light formula control system. Attached Figure Description

[0016] Figure 1 This application provides a flowchart of an LED dynamic light formula control method suitable for industrialized cultivation of edible fungi; Figure 2 The closed-loop feedback correction and cross-batch learning control block diagram provided in this application; Figure 3This application provides an architecture diagram of an LED dynamic light formula control system suitable for industrialized cultivation of edible fungi. Detailed Implementation

[0017] The present application will now be described in further detail with reference to embodiments. It is to be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the scope of the application.

[0018] Example 1 In existing industrialized cultivation of edible fungi, LED supplemental lighting systems mostly employ fixed spectral designs or preset program controls, failing to dynamically and precisely adjust the spectrum according to the specific light requirements of different edible fungi varieties and their different growth stages. Furthermore, the existing systems collect light and environmental data independently, lacking temporal and phase coordination, making image-based growth status assessments susceptible to inconsistencies caused by light source switching. In addition, light regulation is typically based on the overall cultivation environment, lacking the ability to identify and differentiate the growth differences of strains in different regions and at different stages, and lacking causal feedback and adaptive adjustment mechanisms between regulation steps, making multi-dimensional synergistic optimization difficult.

[0019] Based on the above issues, please refer to Figure 1 A flowchart of an LED dynamic light formula control method for industrialized cultivation of edible fungi, provided in this application embodiment, is shown. The method includes the following steps: Step 1: Collect multispectral image data of edible fungi in the cultivation environment, including light intensity, temperature and humidity, CO2 concentration, and each cultivation unit, and ensure that the acquisition of multispectral images and the control of LED light sources are synchronized in time phase. Among these parameters, light intensity, temperature and humidity, and CO2 concentration are fundamental parameters reflecting the physicochemical state of the cultivation environment. These are obtained in real time through photosensitive sensor arrays, temperature and humidity sensor groups, and CO2 sensor groups deployed on each cultivation layer shelf, and their function is to construct the environmental background field for edible fungi growth. Edible fungi multispectral image data refers to image sequences containing reflectance information in specific wavelength bands. These are acquired by periodically photographing each cultivation unit using a multispectral imager and are used to characterize the morphological characteristics and physiological state of the fungal community. Time-phase synchronization alignment means that the exposure time of the multispectral imager and the time when the LED light source outputs a stable spectrum strictly coincide on the time axis. This alignment is achieved by generating trigger signals for the imager and PWM drive signals for the LED light source separately using the same timing controller. Its function is to eliminate image spectral distortion caused by transients or duty cycle fluctuations during light source switching, ensuring that the acquired image data accurately reflects the state of the fungal community under the current light formula. For example, when the timing controller sends a high-level pulse, the LED light source immediately enters a steady-state output mode, and simultaneously the multispectral imager opens the shutter for exposure, with the phase difference between the two controlled within microseconds. If the light source driving frequency is detected to be 1kHz, the trigger signal delay time of the imager is set to an integer multiple of half a PWM cycle to avoid the voltage ramp-up phase. Through this time-phase synchronization alignment mechanism, the interference of ambient light noise on multispectral feature extraction can be significantly reduced, providing high signal-to-noise ratio input data for subsequent growth trend prediction.

[0020] Step 2: Input the collected data into the multimodal time series fusion model and output the prediction results of the microbial community growth status of each cultivation unit and the confidence score of the current growth stage; The "multi-class data" can refer to the collection of light intensity, temperature and humidity, CO2 concentration, and multispectral image data acquired above, which serves as the input vector for the multimodal temporal fusion model. The multimodal temporal fusion model can be a deep learning network capable of simultaneously processing temporal sensor data and spatial image features, specifically including an encoder-decoder structure. The encoder integrates a spatiotemporal attention mechanism for cross-modal feature alignment and fusion encoding of heterogeneous data. The microbial community growth prediction result refers to the quantitative prediction values ​​of key growth indicators such as microbial community density, primordium formation quantity, and average fruiting body diameter. This is obtained through regression analysis of the fused high-dimensional feature vector by the model decoder, and its function is to predict the future growth trend of edible fungi. The confidence score for the current growth stage refers to a probability value characterizing the reliability of the current prediction result, typically ranging from 0 to 1. It is calculated based on the matching degree between the feature distribution output by the model and the pre-training library, and its function is to serve as a weighting basis for subsequent light formulation decisions. For example, if the red light reflectance in the input data suddenly increases while the temperature remains constant, the model may output a prediction of the primordium formation stage and give a confidence score of 0.85; if there is a conflict between the sensor data and image features (e.g., the temperature shows a suitable temperature but the image shows sparse hyphae), the confidence score may drop to 0.4. Through deep fusion of multimodal data, this step can overcome the limitations of judging from a single data source and achieve intelligent perception and high-precision state assessment of complex growth environments.

[0021] Step 3: Based on the confidence score, the fuzzy membership method is used to determine the current growth stage or stage transition interval, and the corresponding light formula parameters are output; when any cultivation unit is detected to have entered the preset key growth stage, a linkage pre-trigger command is issued to other units in the same stage within the cultivation layer. The fuzzy membership method refers to an algorithm that uses fuzzy mathematics theory to handle the uncertainty of growth stage boundaries. It maps continuous confidence scores to discrete growth stage labels or continuous transition interval coefficients by constructing a sigmoid or triangular membership function. Its purpose is to avoid control abrupt changes caused by traditional threshold judgments and achieve a smooth transition in the light recipe. Light recipe parameters refer to a set of light intensity ratios, total illuminance, and photoperiod patterns for each spectral channel, such as red, blue, and green light. These parameters are retrieved from a pre-set light recipe parameter library or obtained through linear interpolation of parameters from adjacent stages, based on the determined growth stage, and are used to guide the output configuration of the LED light source. Pre-defined key growth stages refer to important developmental nodes that are sensitive to changes in illumination and require group coordination, such as the primordium differentiation stage or the fruiting body expansion stage. These are logical states pre-defined in the decision engine. The coordinated pre-trigger command can be a broadcast command carrying a synchronization switching signal. When the main control unit detects that a target unit has entered a critical stage, it sends this command to other units on the same shelf with similar confidence scores. Its function is to initiate a collaborative control process among multiple units, ensuring consistency of the light environment within the same spatial area. For example, when a unit's confidence score is 0.6, placing it in the fuzzy transition zone between the mycelial and primordium formation stages, the system calculates an interpolation coefficient of 0.6 based on the membership function and mixes the light formulas for the two stages in a 4:6 ratio. Once the unit is confirmed to have entered the primordium formation stage, the system immediately sends a pre-trigger command to other units on the same shelf with scores greater than 0.5, preparing for synchronous spectral switching. By combining fuzzy membership and coordinated pre-trigger, the adaptation problem of continuously changing growth stages is solved, and a group-wide collaborative response in the spatial dimension is achieved.

[0022] Step 4: Decompose the light formula parameters into driving signals for multiple independent spectral channels, drive the LED light source to output the corresponding spectrum, and when a linkage pre-trigger command is received, control multiple cultivation units to switch or transition the light formula simultaneously in a synchronous manner. The multiple independent spectral channels can refer to independent light-emitting units such as red, blue, green, near-infrared, and far-infrared light that constitute white light or specific composite light, with each channel equipped with an independent constant current drive circuit. The drive signal can be a PWM duty cycle signal or an analog voltage signal that controls the conduction time and current magnitude of the LED chips in each channel. It is obtained by converting the light intensity ratio in the light formula parameters into specific electrical signal parameters through spectral superposition calculations, and its function is to accurately reconstruct the target spectral distribution. The synchronization method can refer to the operation of updating the drive signal by the controllers of multiple cultivation units under the same clock cycle after receiving a linkage pre-trigger command. This is achieved through a clock synchronization protocol or hardware synchronization bus between distributed controllers, and its function is to eliminate the spectral switching time difference between different units and prevent growth differences caused by uneven illumination. For example, when the light formula requires a red light ratio of 60% and a blue light ratio of 40%, the control system sets the PWM duty cycle of the red light channel to 60% and the blue light channel to 40%. When a linkage command is received, the drivers of all units in the same layer simultaneously execute a duty cycle gradual change within milliseconds, smoothly transitioning from the old formula to the new formula. By decomposing the light formula into independent channels and implementing synchronous control, high-precision spectral customization and spatially consistent light environment regulation can be achieved, effectively avoiding cell stress caused by local light mutations.

[0023] Step 5: Collect the spectral parameters of the actual output of the LED light source, calculate the difference between it and the target light formula, and perform feedback correction on the above fuzzy membership parameters based on the difference and the prediction results of the microbial community growth status. The actual output spectral parameters of the LED light source can refer to the light intensity values ​​and spectral distribution curves of each band monitored in real time by a photosensitive sensor or multispectral imager. These are obtained through high-frequency sampling during the operation of the LED light source and are used to verify the output accuracy of the execution layer. The difference can refer to the deviation index between the actual output spectrum and the aforementioned target light formula, such as the color difference index ΔE or root mean square error. This is calculated by comparing measured data with target data, and its function is to quantify execution error. The fuzzy membership parameter can refer to the shape parameters of the membership function (such as steepness and center point position). It is dynamically adjusted based on the difference and the predicted microbial community growth status. Its function is to correct the decision logic to adapt to actual execution deviations and growth feedback. For example, if monitoring finds that the actual red light intensity is 10% lower than the target value (exceeding the difference standard), and the microbial community growth rate does not meet expectations, the system will automatically adjust the threshold of the fuzzy membership function, making the next stage of light formula decision more inclined to compensate for red light intensity or extend the duration of this stage. Conversely, if the difference is extremely small and growth is good, the parameters are maintained or fine-tuned to maintain stability. By introducing dual feedback from actual spectral differences and growth status, this step constructs a closed-loop control link, enabling the system to have the ability to self-correct and adaptively optimize.

[0024] This application establishes a cross-step collaborative relationship through the coordinated operation of steps S1 to S5. Specifically, the spectral difference and growth trend results calculated in step S5 are not only used for the current feedback correction, but also act as adjustment inputs to the preceding steps: the difference index can be used to dynamically adjust the acquisition frequency of the multispectral image in step S1, that is, to increase the sampling rate to obtain more refined correction data when the light efficiency quality is poor; the confidence score output in step S2 can trigger fine-tuning correction of the sampling phase difference in step S1 when it is continuously low, ensuring the temporal accuracy of the data source; the gradual duration coefficient carried by the linkage pre-trigger command generated in step S3 is dynamically calculated based on the standard deviation of the confidence score of the same layer unit, coupling the spatial growth difference to the temporal transition parameter, realizing deep spatial-temporal linkage; when the difference in step S5 exceeds the threshold, it can also trigger the retrospective correction mode in step S4, calling the historical parameters of the previous stage for weighted fusion. These bidirectional or cross-step closed-loop control links enable the entire system to form an intelligent cycle of perception-decision-execution-feedback-re-perception, significantly improving the accuracy, response speed and intelligence level of light regulation in the industrial cultivation of edible fungi, and effectively solving the problems of extensive regulation, lack of temporal coordination and insufficient adaptive ability in existing technologies.

[0025] Example 2 In another embodiment, the method further includes a specific implementation of the time phase synchronization alignment and dynamic adjustment of the acquisition frequency in step S1.

[0026] Step 1: In step S1, time-phase synchronization alignment is achieved by generating the exposure synchronization signal of the multispectral imager and the PWM drive signal of the LED light source using the same timing controller; In this context, time-phase synchronization alignment refers to maintaining a strict or fixed phase relationship between the image acquisition time of the multispectral imager and the stable emission time of the LED light source on the time axis. Specifically, this synchronization mechanism is uniformly scheduled and generated by a unified timing controller (such as an FPGA chip or a high-precision microcontroller) deployed at the system edge. This timing controller maintains a global time base clock, based on which two signals are simultaneously derived: one is the PWM drive signal for each channel of the LED light source, used to control the light source's on / off state and duty cycle to adjust the light intensity; the other is the exposure synchronization trigger signal for the multispectral imager. Through the generation of homogeneous signals at the hardware level, it is ensured that the multispectral imager performs exposure acquisition only during the steady-state high level of the PWM waveform of the LED light source (i.e., when the light source output spectrum is stable and without transient fluctuations). This setting eliminates the spectral reflectance calculation error caused by the phase misalignment between the light source switching moment and the image acquisition moment, ensuring high consistency and accuracy of the multispectral image data input to subsequent models. For example, when the LED red light channel operates at a frequency of 1kHz and a duty cycle of 50%, the timing controller strictly limits the imager's exposure window to a steady-state time of 200μs in the middle of each cycle, avoiding interference from the spectral composition caused by the current surges during on / off switching. This co-source signal generation and phase-locking mechanism fundamentally solves the problem of accumulated phase drift caused by traditional independent clock sources, significantly improving the data foundation quality for growth state identification.

[0027] Step 2: Furthermore, when the difference calculated in step S5 is lower than the preset threshold, the acquisition frequency of the multispectral image in step S1 is automatically increased.

[0028] The difference degree refers to the deviation between the actual output spectral parameters of the LED light source calculated in step S5 and the target light formula parameters, typically expressed as the color difference index (ΔE) or the percentage of relative light intensity error. The preset threshold is a judgment boundary pre-set according to the accuracy requirements of edible fungus growth monitoring, such as ΔE < 2.0 or light intensity error < 3%. The automatic increase in acquisition frequency mechanism in this step is a positive feedback strategy based on light control quality. Specifically, the system monitors the difference degree value output in step S5 in real time. Once the difference degree is lower than the preset threshold for multiple consecutive sampling periods, it indicates that the current output state of the LED light source is extremely stable and accurately matches the target formula, and the light environment is in a high-quality and controllable state. Under this premise, the control system automatically sends a command to the multispectral imager in step S1 to increase its acquisition frequency from the basic frequency (e.g., once per hour) to a high-frequency mode (e.g., once every 15 minutes or once every 5 minutes). This frequency increase does not occur randomly, but depends on the linkage between the real-time evaluation results of step S5 and the execution parameters of step S1. When light efficiency is excellent, increasing image acquisition density can capture more subtle growth trends in edible fungi (such as early signs of primordia germination or the instantaneous rate of cap unfolding), providing richer time-series data for multimodal time-series fusion models and thus improving the granularity of growth status prediction. For example, during the critical period of primordia formation in king oyster mushrooms, if the system detects that the spectral difference is below 1.5 for 10 consecutive minutes, the imaging frequency is immediately adjusted from 60 minutes / time to 10 minutes / time, thereby intensively recording the primordia differentiation process. Through this collaborative mechanism of high-quality light-controlled triggering high-density sensing, the waste of storage and computing resources caused by acquiring invalid data when light control is unstable is avoided, and the data value is maximized under optimal light conditions, achieving a dynamic balance between system resource allocation and monitoring accuracy.

[0029] This application constructs a deep closed loop between sensing quality and control precision through the synergistic effect of the aforementioned technical features. By generating the exposure synchronization signal of the multispectral imager and the PWM drive signal of the LED light source using the same timing controller, measurement noise caused by time phase misalignment is eliminated from the hardware source, ensuring the authenticity of single-frame image data. Based on this, the low difference calculated in step S5 is used as a confidence anchor point to dynamically trigger an increase in the acquisition frequency in step S1, forming a positive reinforcement loop of precise light control → high confidence verification → high-frequency sensing. This linkage not only solves the spectral distortion problem caused by phase asynchrony in traditional systems but also overcomes the deficiency of fixed sampling frequencies in adapting to different light control stability scenarios. When light control is precise, the system automatically enters a fine observation mode, using high-density image data streams to further explore growth patterns; while when fluctuations in light control lead to increased difference, the system maintains the base frequency to avoid noise data contaminating the model. The combined use of these two features allows the entire control system to adaptively optimize the monitoring granularity while ensuring data purity, significantly improving the ability to capture minute growth changes and the response speed in the industrial cultivation of edible fungi.

[0030] Example 3 In one embodiment, the method further includes a reverse feedback processing mechanism for abnormal confidence scores in step S2.

[0031] Step 1: Input the collected data into the multimodal time series fusion model and output the prediction results of the microbial community growth status of each cultivation unit and the confidence score of the current growth stage. The confidence score is a quantitative indicator characterizing the reliability of the multimodal temporal fusion model's determination of the current growth stage. Its value is typically set between 0 and 1; a higher value indicates clearer feature extraction and more accurate stage determination of the current data. This score originates from the maximum probability distribution output by the model decoder or the normalized feature matching degree. When the confidence score repeatedly falls below the first confidence threshold, it indicates that the system has detected a continuous decline in the current prediction quality. This usually suggests a temporal phase deviation between the multispectral image data and the LED driving signal, causing the image acquisition to fail to accurately capture the spectral reflectance characteristics of the steady-state output of the light source, thus leading to feature alignment failure. In this case, the system will automatically trigger a fine-tuning correction procedure for the sampling phase difference between image acquisition and LED driving in step S1. Specifically, the control module changes the delay time of the multispectral imager's exposure trigger signal relative to the LED light source PWM driving signal with a preset fine-tuning step size (e.g., 0.1 milliseconds each time), and re-executes data acquisition and model prediction after each adjustment, monitoring the new confidence score in real time. This process continues until the confidence score recovers to above the first confidence threshold, thus completing the adaptive calibration of phase synchronization. Through this reverse feedback mechanism that drives acquisition accuracy based on prediction quality, the system can automatically compensate for synchronization errors caused by hardware clock drift, changes in line delay, or device aging without interrupting the cultivation process. This ensures strict alignment of multimodal data in the time dimension and significantly improves the robustness and accuracy of growth status prediction.

[0032] Step 2: Based on the confidence score, the fuzzy membership method is used to determine the current growth stage or stage transition interval, and the corresponding light formula parameters are output; when any cultivation unit is detected to have entered the preset key growth stage, a linkage pre-trigger command is issued to other units in the same stage within the cultivation layer. The high-confidence score recovered after the aforementioned phase fine-tuning correction provides a reliable data foundation for the fuzzy membership determination in this step. The high-confidence score enables the fuzzy membership function to more accurately delineate growth stage boundaries, avoiding misjudgments or frequent jumps due to data noise, thus ensuring that the output light formula parameters truly meet the current physiological needs of edible fungi. Based on this, when the system detects a key growth stage (such as the primordia formation stage) based on accurate determination results, the issued linkage pre-trigger command has higher execution effectiveness, coordinating a smooth transition of the light environment between cultivation units at the same level.

[0033] This application achieves deep collaboration between steps S2 and S1 by constructing a prediction-acquisition bidirectional closed-loop link. Specifically, the confidence score output by step S2 not only serves as the basis for stage judgment but is also reused as a probe to diagnose the quality of data acquisition; once the score continuously falls below the first confidence threshold, it immediately triggers the sampling phase difference fine-tuning correction in step S1. This design breaks the limitations of traditional unidirectional data flow, allowing the uncertainty of model prediction to directly drive the parameter optimization of the front-end sensing hardware. By dynamically adjusting the relative timing of image acquisition and LED driving, the system eliminates the spectral reflectance calculation error caused by phase misalignment, improving the quality of input data from the source, thereby causing the confidence score to rise above the threshold and forming a virtuous self-healing cycle. This not only ensures the stability of the multimodal temporal fusion model in long-term operation but also ensures the accuracy of subsequent light formulation decisions and linkage control, effectively avoiding the risk of erroneous regulation caused by data asynchrony.

[0034] Example 4 In an optional embodiment, the method further includes the specific implementation process of the linkage pre-trigger instruction carrying the gradual duration coefficient in step S3.

[0035] Step 1: Based on the confidence score, the fuzzy membership method is used to determine the current growth stage or stage transition interval, and the corresponding light formula parameters are output; when any cultivation unit is detected to have entered the preset key growth stage, a linkage pre-trigger command is issued to other units in the same stage within the cultivation layer. The linkage pre-trigger command not only includes a switching signal for the target light formula parameters but also carries a dynamically calculated gradient duration coefficient. This gradient duration coefficient is an adjustment factor used to control the time required for the light formula to transition from its current state to the target state. Its function is to determine the smoothness of changes in light intensity and spectral composition, thereby avoiding stress on edible fungi caused by sudden changes in the light environment. This coefficient is dynamically calculated based on the standard deviation of the confidence scores of all linked units in the same layer. Specifically, the system first statistically analyzes the current growth stage confidence scores of all cultivation units on the same layer that have received the linkage pre-trigger command and calculates the standard deviation (σ) of these score data. This standard deviation reflects the consistency or dispersion of the growth states among units in the same layer: the smaller the standard deviation, the more synchronized the growth of each unit; the larger the standard deviation, the more significant the differences in growth among units.

[0036] For example, on a certain cultivation rack, 10 cultivation units are identified as about to enter the primordia formation stage. The system reads the confidence scores of these 10 units as follows: 0.82, 0.85, 0.83, 0.84, 0.81, 0.90, 0.45, 0.50, 0.86, 0.83. Calculations show that the standard deviation of this data set is relatively large (due to the presence of low scores of 0.45 and 0.50), indicating that the growth of some units within this layer is significantly lagging. At this point, the system calculates the transition duration coefficient according to the formula T_adj = T_base + k × σ, where T_base is the base transition duration (e.g., 2 seconds), and k is a preset proportional coefficient (e.g., 2 seconds / unit standard deviation). If the calculated σ is 0.15, the transition time corresponding to the final transition duration coefficient will be extended to 2.3 seconds or even longer; the specific value can be set according to the actual algorithm. Conversely, if all unit scores are between 0.80 and 0.85 with a very small standard deviation, the transition duration will be close to the baseline value, enabling rapid switching.

[0037] This mechanism, which maps spatially distributed growth variability (standard deviation) to a transition duration parameter over time, allows the switching process of the light formula to adaptively accommodate the growth differences between different units. For populations with large growth differences, the extended transition duration provides a more ample physiological adaptation window for lagging units, effectively avoiding localized light stress or developmental stagnation caused by abrupt, rapid switching. Conversely, for populations with uniform growth, the shorter transition time ensures timely regulation. The linkage pre-trigger command output in this step, with its transition duration coefficient, provides crucial time control for the synchronous and smooth switching of multiple cultivation units in subsequent steps.

[0038] Step 2: Decompose the light formula parameters into driving signals for multiple independent spectral channels, drive the LED light source to output the corresponding spectrum, and when a linkage pre-trigger command is received, control multiple cultivation units to switch or transition the light formula simultaneously in a synchronous manner. Upon receiving the generated linkage pre-trigger command, the control system analyzes the gradient duration coefficient carried in the command and executes synchronous transition control of multiple cultivation units accordingly. This step is executed by the edge-side controller, which is configured to initiate the cross-gradient algorithm upon receiving a synchronization pulse signal. Specifically, the controller sets the time window for the PWM duty cycle to change from its current value to the target value based on the analyzed gradient duration coefficient. Within this time window, the drive signals for the five independent spectral channels—red, blue, green, near-infrared, and far-infrared—do not undergo a step-like transition but rather a smooth interpolation transition according to linear or nonlinear curves.

[0039] For example, when the gradient duration coefficient indicates a transition time of 5 seconds, the controller will uniformly increase the duty cycle of the red light channel from 20% to 60% within 5 seconds, while simultaneously decreasing the blue light channel from 50% to 30%. The remaining channels will also be adjusted synchronously according to their respective target ratios. During this process, all linked cultivation units on the same layer strictly maintain temporal phase synchronization, meaning that all units begin the transition at the same moment and complete the transition at the same moment (determined by the gradient duration coefficient).

[0040] By combining the gradient duration coefficient dynamically determined based on standard deviation with the synchronously executed cross-gradient algorithm, a deep coupling regulation of spatial heterogeneity and temporal smoothness is achieved. This synergistic effect ensures that even in large-scale factory cultivation where there are significant differences in individual development, the system can flexibly adjust the time parameters to accommodate spatial inhomogeneities. This guarantees both the consistency of population regulation and the specificities of individual growth, thereby significantly improving the stability of the edible mushroom cultivation environment and the uniformity of the final product.

[0041] This application constructs an intelligent linkage response mechanism through the synergistic effect of the aforementioned technical features. Specifically, the standard deviation of the confidence score is used to quantify the growth dispersion of units in the same layer, and a gradual duration coefficient is dynamically generated accordingly, solving the technical problem that traditional fixed-duration switching cannot adapt to differences in population growth. This coefficient is then passed to the aforementioned mechanism as a time constraint for the generation of the driving signal, guiding the LED light source to perform a smooth transition that is synchronous but variable in duration. This cross-step parameter transfer and logical association enables the system to automatically optimize the physical execution strategy (light switching rate) based on the real-time monitored state of the organism (confidence distribution), forming a complete closed loop from perceiving differences to adaptive execution. As a result, not only is the risk of local light stress that may be caused by uniform switching eliminated, but the rate of deformed mushrooms caused by asynchronous individual development is also significantly reduced, realizing the refinement and flexibility of light environment control in the industrialized cultivation of edible fungi.

[0042] Example 5 In another optional embodiment, the method further includes the specific configuration of the spectral channels and the detailed implementation of the cross-stage backtracking correction mechanism in step S4.

[0043] Step 1: In step S4, there are multiple independent spectral channels, including five channels: red light, blue light, green light, near-infrared light, and far-infrared light. The five independent spectral channels refer to five physical light-emitting units within the LED light source module that can independently control the luminous intensity and duty cycle, each corresponding to a key wavelength band required for edible fungi growth. The dominant wavelength of the red light channel is typically set around 660nm, primarily used to promote stipe elongation and fruiting body development; the dominant wavelength of the blue light channel is typically set around 450nm, used to induce primordia formation and cap differentiation; the dominant wavelength of the green light channel is typically set around 520nm, assisting in regulating the activity of photosynthetic enzymes and completing the life cycle; the dominant wavelength of the near-infrared channel is typically set around 780nm, used to provide thermal effects and deep tissue penetration; and the dominant wavelength of the far-infrared light channel is typically set around 730nm, participating in the reversible regulation of photomorphogenesis. These five channels are powered by independent constant current drive circuits, and each channel is equipped with an independent current feedback loop to ensure that the output accuracy of each wavelength band does not interfere with each other. For example, during the primordium formation stage of king oyster mushrooms, the control system can be set to a 30% duty cycle for the red light channel, 50% for the blue light channel, and 20% for the green light channel, while the near-infrared and far-infrared light channels remain closed or in low-power standby mode. When the fruiting body development stage begins, the system automatically adjusts the red light channel duty cycle to 60%, reduces the blue light to 30%, and activates the far-infrared light channel to 10% to optimize the cap morphology. Through this five-channel independently adjustable structure, the system can construct continuous and precise spectral distribution curves, accurately matching the specific light requirements of different edible fungi varieties at specific growth stages, avoiding the spectral deficiencies caused by traditional monochromatic or dual-color light sources.

[0044] Step 2: Furthermore, when the difference reported in step S5 exceeds the second preset threshold, step S4 automatically retrieves the historical light formula parameters of the previous stage of the cultivation unit, performs weighted fusion with the current target light formula parameters, and re-executes the drive to form a cross-stage retrospective correction. The second preset threshold is a pre-defined upper limit for light efficiency quality tolerance, used to determine whether the current light output deviates significantly from expectations. Its value can be set based on the color difference index (e.g., ΔE>5) or the relative light intensity error (e.g., >15%). The difference degree is a quantitative index calculated by comparing the actual spectral data collected by the photosensitive sensor in step S5 with the target light recipe parameters output in step S3 in real time. When the difference degree exceeds this threshold, it indicates that the current driving strategy may fail due to hardware aging, environmental changes, or model mismatch, triggering a cross-stage backtracking correction mechanism. The core of this mechanism is to break the unidirectional limitation of the time dimension and use historical successful experience to assist in current error correction. Specifically, the system automatically retrieves the set of light recipe parameters that were successfully run and verified to be effective in the previous growth stage for this specific cultivation unit from local storage or cloud database. Subsequently, a weighted fusion algorithm is used to linearly or nonlinearly combine the retrieved historical parameters with the current target light recipe parameters. The weighting coefficients can be dynamically adjusted based on the current degree of difference: the greater the difference, the higher the weight of historical parameters, so as to utilize their stability to suppress the current drastic fluctuations; when the difference slightly exceeds the threshold, the current target parameter is mainly relied upon, with only a small number of historical parameters introduced as smoothing terms. For example, if the current stage is in the development of fruiting bodies, and the target formula is high red and low blue, but the actual output is severely biased towards blue due to a drive failure and cannot be recovered by conventional PID adjustment, the system will immediately retrieve the medium-high blue formula used by the unit in the primordium formation stage, fuse it with the current target formula at a ratio of 3:7, generate a transitional corrected formula, and re-drive the LED array. This processing method is equivalent to allowing the system to retreat to a known safe state and then gradually approach the target state, thereby avoiding cell light stress or growth stagnation caused by forcibly executing erroneous instructions. By combining the five-channel independent control capability in step S4 with the real-time monitoring feedback in step S5, and introducing cross-stage historical data reuse under abnormal operating conditions, the fault-tolerant self-healing and smooth transition of the control system are realized, significantly improving the robustness of light formula execution in complex factory environments.

[0045] This application constructs a highly robust dynamic light formulation control system through the synergistic effect of the aforementioned technical features. By defining five independent channels—red, blue, green, near-infrared, and far-infrared—a physical basis is provided for refined spectral control, enabling the system to flexibly respond to the complex light requirements of different growth stages. Furthermore, by utilizing real-time monitoring of light efficiency quality in step S5, once an execution deviation exceeds the safe range, cross-stage backtracking correction logic is immediately activated, using historically verified parameters from the previous stage as anchor points in the current control loop. This design not only leverages the independent adjustment advantages of the five-channel hardware to achieve precise output after weighted fusion but also solves the problem of inconsistent performance of a single current-moment parameter under extreme abnormal conditions by introducing historical knowledge in the time dimension. The combination of these two aspects ensures that the system maintains the richness and specificity of spectral components when facing hardware drift or sudden interference, while also preventing control divergence through the intervention of historical experience, thus ensuring the stability and controllability of the edible fungi growth environment throughout the entire process.

[0046] Example 6 In another optional embodiment, the method further includes establishing a progressive learning mechanism across batches while performing closed-loop feedback correction in step S5.

[0047] Step 1: Update the difference and growth status prediction results to a cross-batch historical database. This cross-batch historical database is a dataset used to persistently store data from multiple cultivation cycles, comprised of a distributed storage module in the cloud platform controller. The data stored in this database originates from the difference indices between the actual output spectrum and the target spectrum (such as color difference index ΔE and light field uniformity U) calculated in real-time during step S5, as well as the microbial community growth status prediction results (including three-dimensional vectors such as microbial community density, primordium density, and average fruiting body diameter) output by the multimodal time-series fusion model. Specifically, whenever a cultivation batch ends or reaches a preset key time node, the system automatically packages the difference sequences and growth status sequences for all time steps within that batch, adds the current edible mushroom variety identifier, growth stage label, and environmental context information, and writes them into the cross-batch historical database. For example, for the fruiting body development stage of *Pleurotus eryngii*, the system will record the ΔE fluctuation curve and its corresponding average fruiting body diameter growth rate throughout the entire cycle of that batch, forming a complete light control error-growth response correlation record. This structured data storage method breaks the time limitations of a single batch of data, providing a foundation for subsequent multi-batch data mining.

[0048] Step 2: Utilize a cross-batch historical database to perform offline optimization of the light formula parameter library for the same variety and the same growth stage, enabling progressive learning between multiple cultivation batches; Offline optimization refers to a parameter optimization process performed during non-real-time control periods based on accumulated historical data. This process is executed by a second controller deployed on a cloud platform. Its working principle involves periodically (e.g., after every three cultivation batches) retrieving all historical records of the same variety and growth stage from a cross-batch historical database. Bayesian optimization or genetic algorithms are used to analyze the nonlinear mapping relationship between differences and growth status, thereby uncovering a better combination of light formula parameters than the current parameter library (e.g., the optimal ratio of red to blue light, the optimal threshold for total light intensity). Progressive learning is reflected in the fact that as the number of batches increases, the sample space on which the optimization algorithm is based expands, and the generated light formula parameters gradually converge to the global optimum. The optimized parameters are then automatically updated to the light formula decision engine parameter library in step S3 for direct use in subsequent new batches. For example, if historical data shows that under certain temperature and humidity conditions, slightly increasing the proportion of far-red light can significantly reduce ΔE and improve fruiting body uniformity, the offline optimization module will automatically adjust the recommended formula under these conditions, enabling the system to have a more precise control strategy when the next batch starts. Figure 2 As shown in the figure, this is a control block diagram of closed-loop feedback correction and cross-batch learning. The figure clearly shows the data flow from the difference and growth status data collected in step S5 into the cross-batch historical database, and then after being processed by the offline optimization module, the data flow of the light formula parameter library is updated in reverse. It intuitively reflects the evolutionary path of single-batch feedback-multi-batch iteration.

[0049] This application achieves deep synergy of technical features by constructing a cross-batch historical database and an offline optimization mechanism. Specifically, the real-time calculation of the difference degree and growth status prediction results in step S5 not only serves as the basis for single-batch closed-loop control but also continuously accumulates as fuel for long-term learning. The cross-batch historical database acts as a bridge connecting the past and the future, transforming discrete single-cultivation experiences into continuous knowledge accumulation. The offline optimization process acts as an intelligent brain, extracting patterns from massive historical data and feeding back into the light formula parameter library. This synergy enables the system to move beyond adaptive adjustment within a single cultivation cycle and possess self-evolution capabilities across time: as the number of operating batches increases, the system's precision in controlling the light formula for the same variety and growth stage gradually improves, effectively overcoming control deviations caused by strain variation, environmental disturbances, or equipment aging, and significantly improving the long-term stability, yield, and quality of edible fungi industrial cultivation.

[0050] Example 7 In existing technologies for industrialized cultivation of edible fungi, light control systems often suffer from problems such as dispersed sensing devices, fragmented control logic, single execution units, and a lack of effective feedback mechanisms. Traditional systems typically perform environmental data acquisition and image acquisition independently, resulting in a lack of data synergy over time and difficulty in accurately reflecting the true growth status of the fungal community. Simultaneously, the control end often employs a single local controller or purely cloud-based control. The former has limited computing power, making it difficult to run complex predictive models, while the latter suffers from high network latency, hindering the stability of real-time drive control, and there is a lack of reliable communication mechanisms between the two. Furthermore, existing light source drives at the execution layer are mostly open-loop controls, lacking real-time calibration and feedback correction of the output spectrum accuracy, leading to deviations in the light formula during actual execution and affecting cultivation results.

[0051] To address the aforementioned issues, this application proposes a hierarchical, collaborative, and closed-loop feedback system architecture. By dividing perception, control, execution, and feedback into four logical levels, and establishing close data flow and control linkages within and between each level, the aim is to achieve intelligent management and control of the entire process, from high-precision time-series synchronous acquisition to cloud-based intelligent decision-making, and then to edge-driven real-time operation and optical field closed-loop correction.

[0052] Based on the above issues, please refer to Figure 3 This application provides an LED dynamic light formula control system suitable for the industrialized cultivation of edible fungi, used to implement any one of the methods in embodiments 1 to 6 above. The system includes: Sensing layer: includes a photosensitive sensor array, a temperature and humidity sensor group, a CO2 sensor group and a multispectral imager group deployed on each cultivation layer rack. The acquisition trigger signal of the multispectral imager group and the PWM drive signal of the LED light source are generated by the same timing synchronization controller. The control layer includes a first controller deployed at the edge and a second controller deployed on the cloud platform. The first controller is used to generate the drive signal and perform some online corrections in step S4, and the second controller is used to perform situation prediction in step S2 and optical formula decision-making in step S3. The first controller and the second controller use a breakpoint resume communication protocol. Execution layer: includes a five-channel independently dimmable LED array module, each channel equipped with an independent constant current drive circuit and current feedback loop; Feedback layer: includes a light field calibration module and a growth assessment module, which are used to correct the LED driver output and generate a growth status assessment report, respectively.

[0053] The sensing layer refers to the front-end data acquisition subsystem deployed in the edible fungi cultivation environment. Its function is to provide the entire system with multi-dimensional raw data on the environment and growth status. The photosensitive sensor array in this sensing layer can be used to monitor the light intensity distribution in the cultivation space in real time; the temperature and humidity sensor group can be used to collect air temperature and relative humidity; the CO2 sensor group can be used to detect carbon dioxide concentration; and the multispectral imager group can be used to acquire reflectance spectral images of edible fungi in different wavelength bands to characterize their growth morphology. A key feature of the sensing layer in this application is its internal time-phase synchronization mechanism, whereby the acquisition trigger signal of the multispectral imager group and the PWM drive signal of the LED light source are generated by the same timing synchronization controller. This setting ensures that the image acquisition action is strictly aligned with the stable output phase of the LED light source, eliminating the problem of inconsistent image exposure caused by light source flicker or switching, and ensuring the accuracy of the multispectral data. The sensing layer is connected to the control layer via wired or wireless communication, transmitting the acquired heterogeneous data to the control layer for processing in real time, forming the source of the system's data flow.

[0054] The control layer refers to the core data processing and decision-making center of the system, which adopts a cloud-edge collaborative dual-controller architecture. The first controller is deployed at the edge, for example, it can be directly installed in a local control cabinet in the cultivation workshop. It primarily undertakes tasks with extremely high real-time requirements, such as receiving correction instructions from the feedback layer and generating the drive signal for step S4, as well as executing some online correction logic to ensure that the LED light source can quickly respond to environmental changes. The second controller is deployed on the cloud platform, utilizing the powerful computing resources of the cloud to run the computationally intensive step S2 multimodal time-series fusion model for predicting the microbial community growth status, and to execute the complex light formulation decision logic in step S3. The first and second controllers use a breakpoint-resume communication protocol for data interaction. This means that when network fluctuations or interruptions occur, both parties can record the transmission breakpoint, and resume transmission of unfinished instructions or data from the breakpoint after the network is restored, thereby ensuring the reliability of control command issuance and the integrity of historical data. As a bridge connecting the perception layer and the execution layer, the control layer receives raw data from the perception layer and performs intelligent analysis on the one hand, and sends the generated control strategies to the execution layer on the other hand, while receiving the evaluation results from the feedback layer to optimize subsequent decisions.

[0055] The execution layer refers to the light output actuator that directly acts on the growth environment of edible fungi. Its core component is a five-channel independently dimmable LED array module. The five independent spectral channels in this module correspond to specific wavelengths such as red, blue, green, near-infrared, and far-infrared light. Each channel is equipped with an independent constant current drive circuit and a current feedback loop. The constant current drive circuit precisely adjusts the current flowing through the LED chip according to the PWM duty cycle signal sent by the control layer, thereby controlling the light intensity output of each channel. The current feedback loop monitors the actual output current in real time and compares it with the set value, forming a local hardware-level closed loop to prevent spectral shifts caused by temperature drift or device aging. The execution layer is directly electrically connected to the first controller of the control layer, receiving the drive signal it sends and feeding back the actual working status to the feedback layer. Through this independent channel design and hardware feedback mechanism, the execution layer can accurately reproduce the complex light formula calculated by the control layer, achieving precise reconstruction of the spatiotemporal illumination field.

[0056] The feedback layer can refer to the system's quality monitoring and closed-loop correction subsystem, which includes a light field calibration module and a growth evaluation module. The light field calibration module receives photosensitive sensor data or multispectral imaging data from the sensing layer, calculates the difference between the actual output spectrum and the target light formula, and sends correction commands to the control or execution layer based on this difference to correct the LED driver output and ensure the accuracy of the light environment. The growth evaluation module generates a growth status evaluation report based on multispectral image data and environmental data. This report not only includes a description of the current growth status but can also be stored as historical data in a database to guide parameter optimization for subsequent batches. The feedback layer interacts with both the sensing and execution layers. It uses the measured data provided by the sensing layer to evaluate the output effect of the execution layer and feeds the evaluation results back to the control layer, thus forming a complete closed-loop control link at the system level: perception-decision-execution-feedback.

[0057] Specifically, the working process and principle of this application system are as follows: After the system starts, various sensors in the perception layer and the multispectral imager, under the coordination of the same timing synchronization controller, synchronously collect data on light, temperature, humidity, CO2 concentration, and multispectral images of edible fungi in the cultivation environment, ensuring the consistency of the data in time phase. The collected data is transmitted to the control layer, where the second controller (cloud) uses a multimodal time-series fusion model to perform in-depth analysis of the data, outputs the prediction results of the fungal community growth status and confidence score, and determines the current target light formula parameters accordingly; the first controller (edge ​​side) receives the light formula parameters, decomposes them into five independent drive signals, and maintains stable communication with the cloud through a breakpoint resume protocol. Subsequently, the first controller drives the five-channel independent adjustable LED array module in the execution layer, using the independent constant current drive circuit of each channel to output a precise spectral combination. During operation, the feedback layer monitors the actual light field quality and bacterial growth status in real time, the light field calibration module calculates the spectral difference and triggers online correction, the growth assessment module generates an assessment report and feeds it back to the control layer, and the control layer dynamically adjusts the subsequent driving strategies or decision parameters accordingly, thereby achieving dynamic control and adaptive optimization of the entire process.

[0058] As a preferred embodiment, the specific implementation of this application is as follows: In the industrialized cultivation scenario of king oyster mushrooms, the multispectral imager group of the sensing layer and the PWM drive signal of the LED light source are both generated by an FPGA-based timing synchronization controller to ensure that the image exposure time strictly corresponds to the stable lighting period of the red or blue light channel. The second controller of the control layer is deployed on Alibaba Cloud or a private cloud platform, runs a growth prediction model based on the Transformer architecture, calculates the confidence level of the current king oyster mushroom in the primordia formation stage as 0.85, and decides on a light formula with a red:blue:green light ratio of 3:5:2; the first controller uses an embedded ARM chip deployed next to the cultivation rack, receives the formula, and generates the corresponding five PWM signals. In the LED array module of the execution layer, the constant current drive circuits of the five channels, including the red light channel (660nm) and the blue light channel (450nm), adjust the current according to the PWM signal, while the current feedback loop monitors the current fluctuation in real time and automatically compensates for it. The light field calibration module of the feedback layer detects that the red light intensity in a certain area is 5% lower than normal through the photosensitive sensor, and immediately notifies the first controller to fine-tune the duty cycle of that channel; after analyzing the image, the growth assessment module generates an assessment report on the good differentiation of primordia, and marks the data of this batch and stores it in the history database for use in optimizing the light formula parameters during the next batch of cultivation.

[0059] Through the above technical solutions, this application realizes a multi-level linkage and collaborative system architecture. By adopting a synchronous controller in the perception layer to generate acquisition trigger and drive signals, phase misalignment between light source switching and image acquisition is eliminated, ensuring the consistency of multispectral data. By constructing a collaborative architecture between the first controller on the edge side and the second controller in the cloud in the control layer and adopting a breakpoint resume protocol, the powerful computing power of the cloud is utilized for accurate prediction and decision-making, while ensuring the real-time performance and communication reliability of the edge-side drive control. By configuring a five-channel LED array module with an independent constant current drive circuit and current feedback loop in the execution layer, precise independent control of each spectral channel and stable hardware-level output are achieved. By setting up a feedback layer including a light field calibration module and a growth evaluation module, a complete closed loop from light efficiency quality monitoring to growth status evaluation is formed, enabling the system to dynamically correct drive output and decision parameters based on actual operating results, significantly improving the intelligence level of edible fungi factory cultivation and the accuracy of light formula execution.

[0060] Example 8 In one optional implementation, the method further includes: a multispectral imager assembly comprising at least four narrowband filters with center wavelengths set near 450nm, 540nm, 660nm and 730nm respectively.

[0061] At least four narrowband filters can refer to optical elements deployed in the optical path of a multispectral imager for filtering incident light into specific wavelengths. Their function is to physically isolate light, allowing only light within a specific center wavelength and its adjacent narrowband range to pass through and reach the image sensor, thereby acquiring reflectance image data of edible fungi at different characteristic wavelengths. In this application, these filters work in conjunction with other optical components of the multispectral imager to decompose composite ambient light or LED light into independent monochromatic channel images, providing a high signal-to-noise ratio raw data source for subsequent growth status analysis. The bandwidth of the narrowband filter can be set according to actual conditions; for example, the full width at half maximum (FWHM) can be 10nm, 20nm, or 30nm. This application embodiment does not impose special limitations on this, as long as it can effectively distinguish the target wavelength band.

[0062] The center wavelength around 450nm can refer to the characteristic band corresponding to the blue light region of the visible light spectrum. This band is chosen because, in the physiological mechanisms of edible fungi, blue light receptors are most sensitive to the spectrum around 450nm. Light signals in this band are primarily used to induce the formation of primordia and the differentiation of the cap. In the system linkage, image data in this band is collected and input into a multimodal temporal fusion model to assess whether the current cultivation unit possesses the spectral response characteristics for entering the primordia formation stage, thereby assisting in the growth stage determination in step S3.

[0063] The center wavelength around 540nm can refer to the characteristic band corresponding to the green region of the visible light spectrum. This band is often used as a reference band or to detect specific pigment changes, reflecting the absorption characteristics of chlorophyll analogs or other pigments that may be present on the surface of edible fungi. It is also used to construct derived indicators such as the normalized vegetation index to eliminate the influence of light intensity fluctuations. In the overall technical solution, this band data, together with other band data, constitutes a multidimensional feature vector, which is used to correct measurement errors caused by ambient light interference and improve the robustness of growth status prediction.

[0064] The center wavelength around 660nm can be considered the main peak band corresponding to the red light region of the visible spectrum. This band is the main absorption band in the form of phytochrome Pr, and it plays a significant role in promoting the elongation and growth of edible fungi stipes. In terms of coordination, the reflectance data in this band is directly related to the accumulation rate of edible fungi biomass. The system quantifies the mycelial coverage density and the development degree of fruiting bodies by monitoring the grayscale changes or texture features of images in this band, and uses this result as an important factor in the confidence score calculation.

[0065] The center wavelength around 730nm can be considered a characteristic band corresponding to the far-red light region. This band participates in the conversion regulation of phytochrome Pfr form, affecting the photoperiodic response and morphogenesis of edible fungi. Functionally, introducing this band enhances the system's ability to identify critical points in different growth stages, especially for varieties requiring a specific red / far-red light ratio (R / FR) to trigger morphological changes; data from this band provides crucial decision-making information. The ratio or difference between this band and the 660nm band data can serve as a core parameter for determining the transition range of growth stages.

[0066] Specifically, the setting of the four center wavelengths mentioned above is determined based on the spectral response mechanism of common varieties in industrialized edible fungi cultivation (such as king oyster mushrooms, shiitake mushrooms, and enoki mushrooms). 450nm, 540nm, 660nm, and 730nm respectively cover the key physiological action ranges of blue, green, red, and far-red light. In practical implementation, narrowband filters can be fabricated using thin-film interferometry and integrated onto the filter wheel of the multispectral imager or directly implemented using a Bayer array sensor. When the multispectral imager array is working, the timing synchronization controller controls the sequential switching of filters or the time-division exposure of the sensors to ensure that image sequences of these four bands are acquired at the same time phase. This configuration enables the system to capture subtle spectral changes in edible fungi under the influence of key growth factors, avoiding the spectral aliasing problem caused by broadband filtering, thereby supporting the high-precision prediction of the fungal community growth status in step S2 and the fuzzy membership division of growth stages in step S3.

[0067] As a preferred embodiment, the solution of this application is implemented as follows: In the cultivation scenario of king oyster mushroom, a multispectral imager is assembled with four narrowband filters with center wavelengths of 450nm, 540nm, 660nm, and 730nm, respectively. After the system is started, the imager sequentially acquires images of the cultivation unit through these four filters at the instant the LED light source outputs a steady-state spectrum. The image of the 450nm channel is used to analyze the density of primordia differentiation. If the reflectance characteristics of this channel indicate that a large number of primordia are formed, then combined with the stipe elongation reflected by the 660nm channel, the system determines that it has entered the early stage of fruiting body development. At the same time, the far-red ratio is calculated using the data from the 730nm channel. If the ratio is lower than a preset threshold and the background noise of the 540nm channel is stable, then the confidence score of the growth stage transition is higher than 0.7, triggering the switching command of the light formula from the primordia formation stage to the fruiting body development stage. During this process, if an image of a certain band shows abnormal brightness or darkness, the system can perform interpolation compensation based on the redundant data of the other three bands to ensure the continuity of growth status identification.

[0068] Through the above technical solution, this application achieves the ability to accurately capture the differences in the reflectance characteristics of edible fungi to specific spectra at different growth stages by setting a narrowband filter covering key physiological bands of blue, green, red and far-red light. This solves the technical problem in the prior art of inaccurate identification of growth status due to improper selection of spectral bands, thereby improving the contribution of multispectral image data to the prediction of fungal growth status and the pertinence of light formula regulation.

[0069] Example 9 In one embodiment, this application also provides the system of the above embodiment, wherein the main wavelength of the red light of the five-channel LED array module is 660nm, the main wavelength of the blue light is 450nm, the main wavelength of the green light is 520nm, the main wavelength of the near-infrared light is 780nm, and the main wavelength of the far-infrared light is 730nm; the LED chips of each channel are arranged in a triangular grid on the substrate, and a high reflectivity isolation grille is provided between the chips of different channels.

[0070] The dominant wavelengths of red light (660nm), blue light (450nm), green light (520nm), near-infrared light (780nm), and far-infrared light (730nm) refer to the peak wavelength positions of the emission spectra of each LED chip channel. These specific wavelength parameters are set according to the photobiological characteristics of edible fungi. For example, 660nm red light is mainly used to promote stipe elongation and fruiting body development, 450nm blue light is used to induce primordia formation and cap differentiation, 520nm green light assists in completing the life cycle, 730nm far-infrared light participates in photomorphogenesis regulation, and 780nm near-infrared light may be involved in thermal effect management or deep tissue penetration. The above wavelength values ​​can be fine-tuned according to the actual cultivated edible fungi varieties (such as king oyster mushrooms, shiitake mushrooms, enoki mushrooms, etc.) and their specific growth stages, for example, by an offset within ±10nm. This application does not impose any special limitations on this. In the system linkage, these wavelength parameters serve as the core control targets of the execution layer, directly corresponding to the light recipe parameters issued by the control layer. This ensures that the output spectral energy distribution can accurately match the physiological needs of the current growth stage, thereby achieving an accurate mapping from decision-making to execution.

[0071] The LED chips in each channel are arranged in a staggered triangular grid on the substrate. This means that the five different wavelengths of LED chips—red, blue, green, near-infrared, and far-infrared—are not arranged in a simple row and column matrix on the substrate plane, but rather staggered based on the vertices of triangles. This arrangement allows multiple wavelengths of light to be contained within any local area, thereby achieving a more uniform spectral mixing effect in space and avoiding local light spots or spectral unevenness caused by the concentrated distribution of a single wavelength of light. The specific geometric parameters of the triangular grid, such as the grid side length and included angle, can be set according to the actual situation based on the size of the substrate and the packaging specifications of the LED chips. For example, it can be an equilateral triangular grid or an isosceles triangular grid; this application embodiment does not make any special limitations on this. In the overall technical solution, this arrangement structure works closely with the light formula control method. When the control system issues a light mixing command, the triangular grid structure can ensure the natural superposition of multiple spectra within a very short distance, reducing dependence on external diffusers and improving the uniformity of the light field.

[0072] A high-reflectivity isolation grille refers to a physical barrier structure placed between LED chips in different channels. Its surface is coated with or made of a high-reflectivity material (such as silver-plated polymer, white ceramic, or high-reflectivity metal foil). The main function of this grille is to physically block the lateral diffusion of light from adjacent channels, preventing unintended mixing of different wavelengths of light near the light outlet (i.e., optical crosstalk), and ensuring that the light emitted from each channel propagates primarily axially until it reaches a preset mixing distance. The height, width, and reflectivity parameters of the isolation grille can be designed according to the divergence angle of the LED chip and the required mixing distance. For example, the height can be 1 to 3 times the chip thickness, and the reflectivity can be greater than 90%. This application does not impose any special limitations on these parameters. Furthermore, the grille structure also functions as a heat dissipation channel, increasing the contact area between the substrate and the air, which helps to quickly dissipate the heat generated by the LED chip during operation. In system linkage, the presence of the high-reflectivity isolation grille ensures the purity of the output spectrum of the execution layer, making the actual spectral parameters collected in step S5 more accurately reflect the set value of the drive signal, thereby improving the accuracy of closed-loop feedback correction.

[0073] Specifically, the working process of the five-channel LED array module in this application is as follows: when the control layer receives the light formula parameters for a certain growth stage, it resolves them into drive current or PWM duty cycle signals for five independent channels; the LED chips corresponding to each channel are lit simultaneously or at different times according to the physical layout of the triangular grid staggered arrangement; after the emitted specific wavelength light leaves the chip, it is first constrained by a high reflectivity isolation grid, which limits lateral crosstalk, and then propagates in the vertical direction and forms a uniformly mixed light field at a certain distance; this mixed light field irradiates the edible fungus cultivation unit, providing a light environment that meets the current growth requirements; at the same time, the grid structure helps to dissipate the chip heat, maintain the stability of the module's operating temperature, and ensure that the wavelength drift is controlled within the allowable range.

[0074] As a preferred embodiment, the solution of this application is implemented as follows: During the primordium formation stage of *Pleurotus eryngii*, the system sets the ratio of red light (660nm) to blue light (450nm) to 3:5. At this time, the control layer sends a drive signal to the five-channel LED array module. The red and blue light channels operate according to a predetermined duty cycle, while the green, near-infrared, and far-infrared light channels are in a closed or low-power standby state. Because the chip uses a triangular grid staggered arrangement, the red and blue light begin to mix initially near the light-emitting surface. Combined with the suppression of stray light by the high-reflectivity isolation grid, a mixed light field with a uniformity higher than 0.85 is finally formed in the cultivation layer 20cm below the module. If the system detects an abnormal temperature rise in a certain area, the structural advantage of the isolation grid can accelerate heat convection in that area, prevent local high temperature from affecting mycelial growth, and maintain the purity of the spectral components, avoiding the failure of the light formula due to thermal wavelength drift.

[0075] Through the above technical solutions, this application achieves the following: by setting precise main wavelength parameters, it ensures that each spectral channel produces the expected biological effect on the growth of edible fungi; by using a triangular grid staggered layout, it achieves spatial light field homogenization, avoiding inconsistent growth caused by local over-brightness or under-brightness; by setting a high-reflectivity isolation grid, it effectively suppresses light crosstalk between multiple channels, improves the purity of light mixing and control precision, and at the same time improves the heat dissipation performance of the module, enhancing the stability and reliability of the system in long-term operation.

[0076] The above description is merely a preferred embodiment of this application and is not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. LED dynamic light recipe regulation method suitable for industrial cultivation of edible mushrooms, characterized in that, Includes the following steps: Step S1: Collect multispectral image data of light intensity, temperature and humidity, CO2 concentration and edible fungi in each cultivation unit in the cultivation environment, and keep the acquisition of multispectral images and the control of LED light source synchronized in time phase. Step S2: Input the above-mentioned multi-type data collected in step S1 into the multimodal time series fusion model, and output the prediction results of the growth status of edible fungi community in each cultivation unit and the confidence score of the current growth stage. Step S3: Based on the confidence score, the current growth stage or stage transition interval is determined by the fuzzy membership method, and the corresponding light formula parameters are output; when any cultivation unit is detected to have entered the preset key growth stage, a linkage pre-trigger command is issued to other units in the same stage within the cultivation layer. Step S4: Decompose the light formula parameters into driving signals for multiple independent spectral channels, drive the LED light source to output the corresponding spectrum, and when a linkage pre-trigger command is received, control multiple cultivation units to switch or transition the light formula simultaneously in a synchronous manner. Step S5: Collect the spectral parameters of the actual output of the LED light source, calculate the difference between it and the target light formula, and perform feedback correction on the fuzzy membership parameter in step S3 based on the difference and the prediction results of the growth status of edible fungi. Among them, the above steps S1 to S5 are set with cross-step linkage and coordination relationship, including: at least one output result of the subsequent step is used as the adjustment input of the preceding step to form a two-way or cross-step closed-loop control link.

2. The method of claim 1, wherein, The time-phase synchronization alignment in step S1 is achieved by generating the exposure synchronization signal of the multispectral imager and the PWM drive signal of the LED light source by the same timing controller; and when the difference calculated in step S5 is lower than a preset threshold, the acquisition frequency of the multispectral image in step S1 is automatically increased.

3. The method of claim 1, wherein, In step S2, when the confidence score is lower than the first confidence threshold multiple times consecutively, a fine-tuning correction is triggered on the sampling phase difference between image acquisition and LED driving in step S1 until the confidence score recovers to above the threshold.

4. The method of claim 1, wherein, The linkage pre-triggering command in step S3 carries a gradual duration coefficient. This coefficient is dynamically calculated based on the standard deviation of the confidence scores of all linkage units in the same layer. The larger the standard deviation, the longer the gradual duration, in order to adapt to the growth differences between different units.

5. The method of claim 1, wherein, The multiple independent spectral channels mentioned in step S4 include five channels: red light, blue light, green light, near-infrared light, and far-infrared light. Furthermore, when the difference reported in step S5 exceeds the second preset threshold, step S4 automatically retrieves the historical light formula parameters of the previous stage of the cultivation unit, performs weighted fusion with the current target light formula parameters, and re-executes the drive to form a cross-stage retrospective correction.

6. The method of claim 1, wherein, In step S5, the difference and growth status prediction results are also updated to a cross-batch historical database. This database is used to perform offline optimization of the light formula parameter library for the same variety and the same growth stage, so as to realize progressive learning between multiple cultivation batches.

7. A LED dynamic light recipe regulating system suitable for industrial cultivation of edible mushrooms, characterized in that, The system for implementing the method of any one of claims 1 to 6 comprises: Sensing layer: includes a photosensitive sensor array, a temperature and humidity sensor group, a CO2 sensor group and a multispectral imager group deployed on each cultivation layer rack. The acquisition trigger signal of the multispectral imager group and the PWM drive signal of the LED light source are generated by the same timing synchronization controller. The control layer includes a first controller deployed at the edge and a second controller deployed on the cloud platform. The first controller is used to generate the drive signal and perform some online corrections in step S4, and the second controller is used to perform situation prediction in step S2 and optical formula decision-making in step S3. The first controller and the second controller use a breakpoint resume communication protocol. Execution layer: includes a five-channel independently dimmable LED array module, each channel equipped with an independent constant current drive circuit and current feedback loop; Feedback layer: includes a light field calibration module and a growth assessment module, which are used to correct the LED driver output and generate a growth status assessment report, respectively.

8. The system according to claim 7, characterized in that, The multispectral imager group includes at least four narrowband filters with center wavelengths set around 450nm, 540nm, 660nm and 730nm respectively.

9. The system according to claim 7, characterized in that, The five-channel LED array module has a red light main wavelength of 660nm, a blue light main wavelength of 450nm, a green light main wavelength of 520nm, a near-infrared main wavelength of 780nm, and a far-infrared main wavelength of 730nm. The LED chips of each channel are arranged in a triangular grid on the substrate, and a high-reflectivity isolation grille is provided between the chips of different channels.

Images