A factory mushroom growth monitoring and environment regulation system, method and medium

Abstract

The application discloses a kind of factory edible mushroom growth monitoring and environmental control system, comprising: multi-modal data acquisition module collects the multi-dimensional physical and biochemical data of edible mushroom growth environment;Edge computing module carries out data preprocessing to the data collected, inputs the data after preprocessing into visual smell fusion model respectively to carry out visual feature extraction and olfactory feature extraction, and the visual feature and olfactory feature are fused, output growth state vector, carries out growth stage identification and disease risk assessment, obtains disease risk index, according to the corresponding control strategy of disease risk index output, sends control strategy to executing mechanism;Executing mechanism carries out environmental control operation according to control strategy;Cloud platform is used to store historical image data and odor data, train and update deep reinforcement learning model, and send model parameters to edge computing module.The system realizes early warning of disease based on VOCs fingerprint.

Description

Technical Field

[0001] This invention belongs to the field of agricultural intelligent equipment technology, specifically relating to a system, method and medium for monitoring the growth and controlling the environment of edible fungi in a factory setting. Background Technology

[0002] With the acceleration of agricultural modernization, the cultivation model of edible fungi (such as shiitake, oyster, and enoki mushrooms) is shifting from traditional family workshops to intensive and intelligent factory-style production. Factory cultivation is characterized by high density, three-dimensional operation, and a full life cycle, requiring extremely high control over environmental factors (temperature, humidity, CO2 concentration, light, and airflow). Especially for wood-rotting fungi like shiitake, their growth cycle encompasses multiple physiological stages, including mycelial growth, color change (formation of a protective film), primordia differentiation (bud induction), and fruiting body development (fruiting). The environmental requirements at each stage are drastically different. For example, the color change stage requires specific temperature differences and light induction, while the fruiting stage requires strict humidity control to prevent disease.

[0003] Currently, environmental control systems for edible mushroom factories have mainly gone through three stages of development:

[0004] 1) Traditional automatic control: using PLC or microcontroller to perform simple logic control on environmental equipment (fans, humidifiers, air conditioners) based on preset fixed thresholds.

[0005] 2) IoT-based remote monitoring: Environmental data is collected by sensors and uploaded to the cloud, enabling remote viewing and manual intervention via a mobile app.

[0006] 3) Vision-based intelligent monitoring: In recent years, some advanced systems have introduced RGB cameras to monitor the growth of edible fungi (such as cap diameter) using image recognition technology.

[0007] Regarding patent literature, for example, Chinese invention patent application CN119002608A, entitled "Environmental Control System and Method for Plant Factory Based on Artificial Intelligence," discloses a dual-path control method based on "intelligent strategy + experience strategy." Chinese invention patent application CN120595885A, entitled "Intelligent Thermal Management Control System for Plant Factory," discloses a thermal management scheme based on an RGB-D camera and reinforcement learning.

[0008] While the aforementioned technologies have improved the automation level of environmental control to some extent, they still have significant limitations when facing the high-risk industrial production of edible fungi, specifically:

[0009] 1) Lack of "non-visual" early disease warning capabilities: Existing systems mainly rely on manual inspections or simple machine vision to monitor diseases (such as Trichoderma, Neurospora, and bacterial brown spot). However, visual recognition has a significant "lag effect." When the camera captures green mold spots or mycelial collapse on the surface of the mushroom log, the pathogens have often already completed spore maturation and begun to release spores. At this time, through a unified circulating ventilation system, the pathogenic spores may have already spread throughout the entire mushroom house. Existing technologies ignore the characteristic volatile organic compounds (VOCs) released by edible fungi and pathogens during metabolism, and cannot achieve early warning based on odor fingerprints.

[0010] 2) Insufficient depth of multimodal fusion, unable to cope with complex physiological changes: Most existing solutions only focus on environmental scalar data (temperature and humidity) or single visual phenotypic data. There is a lack of a mechanism to deeply integrate environmental parameters, visual features (color change, morphology), and olfactory features (metabolic odor). For example, during the "color change period" of shiitake mushrooms, it is difficult to judge whether the color change has penetrated deeply by sight alone. Without the support of multidimensional data, it is easy to cause color change failure or rotten mushrooms.

[0011] 3) Cloud-based decision-making suffers from latency, hindering real-time disease control: Existing intelligent control systems largely rely on cloud servers for model inference. When sudden disease risks occur (such as an electronic nose detecting abnormally high concentrations of musty VOCs), delays in data upload and cloud feedback may cause the optimal "blocking window" to be missed (e.g., failure to immediately shut down local fans, resulting in spores being dispersed). There is a lack of edge computing architectures with local online learning and real-time response capabilities.

[0012] 4) Lack of adaptive strategies for the growth stages of edible fungi: General-purpose plant factory systems are difficult to adapt to the complex stage-specific needs of edible fungi. For example, existing systems have difficulty automatically identifying the "end of the first harvest of mushrooms" node and automatically switching to the "mushroom cultivation and rejuvenation" mode, often still relying on frequent adjustments of settings based on human experience. Summary of the Invention

[0013] This invention provides a system, method, and medium for monitoring the growth and controlling the environment of industrialized edible fungi, in order to solve the problem that existing technologies cannot effectively block disease phenotypes before they appear.

[0014] In a first aspect, the first embodiment of the present invention provides a factory-scale edible fungus growth monitoring and environmental control system, comprising: a multimodal data acquisition module, an edge computing module, an actuator, and a cloud platform.

[0015] The multimodal data acquisition module is used to collect multidimensional physical and biochemical data of the edible fungi growth environment and transmit the collected data to the edge computing module.

[0016] The edge computing module is used to preprocess the collected data to obtain preprocessed data. The preprocessed data is then input into the visual-olfaction fusion model for visual feature extraction and olfactory feature extraction, and the visual and olfactory features are fused to output a growth state vector. The growth state vector is then input into a deep reinforcement learning model for growth stage identification and disease risk assessment to obtain a disease risk index. Based on the disease risk index, a corresponding control strategy is output and sent to the execution agency.

[0017] The actuator performs environmental control operations according to the control strategy;

[0018] The cloud platform is used to store historical image data and odor data, train and update deep reinforcement learning models, and send the updated deep reinforcement learning model parameters to the edge computing module.

[0019] Furthermore, the multimodal data acquisition module includes a microenvironment sensor group, a machine vision acquisition unit, and an olfactory sensing unit;

[0020] The microenvironment sensor group is used to collect ambient temperature and humidity, carbon dioxide concentration, light intensity, and wind speed.

[0021] The machine vision acquisition unit is used to acquire image data of the surface of the mushroom sticks;

[0022] The olfactory sensing unit is used to collect odor fingerprints of characteristic volatile organic compounds contained in the air in real time.

[0023] Furthermore, the olfactory sensing unit includes a gas sensor array, which is used to detect alcohols, sulfides, aldehydes and geosmin, respectively.

[0024] Furthermore, the edge computing module includes a visual-olfaction fusion model, which includes a visual branch, an olfactory branch, and a fusion unit;

[0025] The visual branch uses a first neural network model to extract visual features of mushroom stick color change, mushroom bud diameter and lesion.

[0026] The olfactory branch uses a second neural network model to extract olfactory features;

[0027] The fusion unit uses a feature splicing and attention mechanism to fuse visual and olfactory features and outputs a growth state vector.

[0028] Furthermore, the deep reinforcement learning model includes a reward function unit, which defines a comprehensive reward function that includes disease constraints. The formula for the comprehensive reward function is:

[0029] R = α·ΔGrowth - β·Energy – γ|Err| - δ·Drisk;

[0030] Wherein, α•ΔGrowth represents the effective biomass increment calculated through visual analysis, β•Energy represents the system's comprehensive energy consumption at the current moment, γ|Err| represents the Euclidean distance between the current environmental parameters and the target setpoint, and δ•Drisk represents the disease risk index output by the visual-olfactory fusion model.

[0031] Furthermore, the edge computing module also includes an adaptive optimization unit, which performs positive reinforcement and negative penalty based on the reward function.

[0032] Secondly, another embodiment of the present invention provides a method for monitoring and controlling the growth of edible fungi in industrialized production, applicable to edge computing modules, the method comprising:

[0033] To obtain multidimensional physical and biochemical data on the growth environment of edible fungi;

[0034] The acquired data is preprocessed to obtain preprocessed data;

[0035] The preprocessed data is input into the visual-olfactory fusion model for visual feature extraction and olfactory feature extraction respectively. The visual and olfactory features are then fused to output a growth state vector.

[0036] The growth state vector is input into a deep reinforcement learning model to identify growth stages and assess disease risk, thus obtaining a disease risk index.

[0037] Based on the disease risk index, the control and execution agency outputs control strategies for regulating the environment.

[0038] Furthermore, the specific method for inputting the preprocessed data into the visual-olfactory fusion model for visual feature extraction and olfactory feature extraction respectively, and then fusing the visual and olfactory features, includes:

[0039] The first neural network model was used to extract visual features of mushroom stick color change, mushroom bud diameter, and lesion.

[0040] Olfactory features are extracted using a second neural network model;

[0041] Visual and olfactory features are fused using feature splicing and attention mechanisms.

[0042] Furthermore, the deep reinforcement learning model defines a comprehensive reward function that includes disease constraints, and the formula for the comprehensive reward function is:

[0043] R = α·ΔGrowth - β·Energy – γ|Err| - δ·Drisk;

[0044] Wherein, α•ΔGrowth represents the effective biomass increment calculated through visual analysis, β•Energy represents the system's comprehensive energy consumption at the current moment, γ|Err| represents the Euclidean distance between the current environmental parameters and the target setpoint, and δ•Drisk represents the disease risk index output by the visual-olfactory fusion model.

[0045] Thirdly, another embodiment of the present invention provides a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, cause the processor to perform the method described in the second embodiment above.

[0046] The beneficial effects of this invention are:

[0047] The industrialized edible fungi growth monitoring and environmental control system, method, and medium provided in this invention have the following advantages:

[0048] 1. This invention overcomes the lag in visual detection, enabling early disease warning based on VOCs fingerprints. Existing technologies rely on morphological recognition and can only trigger alarms after lesions appear. This invention introduces gas sensor array technology, establishing a non-visual perception mechanism by monitoring changes in the concentration of specific volatile organic compounds released by edible fungi and competing bacteria during metabolism. This mechanism can detect abnormal biochemical indicators during the spore germination stage or mycelial incubation period of pathogens, significantly advancing the risk detection window and achieving a technological leap from phenotypic monitoring to metabolic monitoring.

[0049] 2. A low-latency edge architecture was constructed to achieve real-time blocking of aerosol transmission risks. Addressing the network latency and reliability issues inherent in cloud-based control, this embodiment of the invention employs an edge computing architecture, deploying the visual-olfaction fusion model on the local gateway. Once a peak of pathogenic odor characteristics is detected, the system can achieve a millisecond-level actuator response without waiting for cloud instructions. This instantaneous response mechanism effectively blocks the aerosol transmission of pathogenic spores through ventilation ducts, significantly reducing the risk of cross-infection caused by delayed regulation of environmental factors.

[0050] 3. A risk-constrained reinforcement learning strategy is proposed to solve the game between growth and disease resistance. Existing control logics mostly employ single PID threshold control or optimization algorithms that only aim at biomass accumulation. This invention introduces a disease risk penalty term into the deep reinforcement learning model, constructing a multi-objective optimization function that includes growth rate, energy consumption, and health status. During decision-making, the system can dynamically weigh the dual impact of environmental parameters on mycelial growth and the proliferation of miscellaneous bacteria, automatically generating a robust control strategy centered on antibacterial activity during high-risk disease periods, thus improving the biosafety of the cultivation system.

[0051] 4. Achieved full-cycle adaptive and precise control based on multimodal feature fusion. Addressing the problem that traditional single-parameter environmental control cannot adapt to the complex physiological cycle of edible fungi, this invention integrates visual and olfactory features to achieve automatic identification and status assessment of key physiological stages such as the mycelium growth stage, color change stage, primordium differentiation stage, and fruiting stage. The system can dynamically load the corresponding environmental control model based on the identification results, achieving precise spatiotemporal matching between environmental supply and fungal physiological needs. Attached Figure Description

[0052] Figure 1 This is a schematic diagram of a factory-scale edible fungus growth monitoring and environmental control system provided in the first embodiment of the present invention;

[0053] Figure 2 This is a structural block diagram of a factory-scale edible fungus growth monitoring and environmental control system provided in the first embodiment of the present invention;

[0054] Figure 3 This is a flowchart of a method for monitoring and controlling the growth of edible fungi in a factory setting, provided in the second embodiment of the present invention;

[0055] Figure 4 This is a flowchart of a method for monitoring the growth and controlling the environment of industrialized edible fungi, provided in the third embodiment of the present invention. Detailed Implementation

[0056] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, and to make the above-mentioned objectives, features and advantages of the embodiments of the present invention more apparent and understandable, the technical solutions in the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0057] In the description of this invention, unless otherwise specified and limited, it should be noted that the term "connection" should be interpreted broadly. For example, it can be a mechanical connection or an electrical connection, or it can be a connection between two internal components. It can be a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above term according to the specific circumstances.

[0058] See Figure 1 , 2 The first embodiment of the present invention provides a factory-scale edible fungus growth monitoring and environmental control system, comprising: a multimodal data acquisition module, an edge computing module, an actuator, and a cloud platform.

[0059] The multimodal data acquisition module is used to collect multidimensional physical and biochemical data of the edible fungi growth environment and transmit the collected data to the edge computing module.

[0060] The edge computing module is used to preprocess the collected data to obtain preprocessed data. The preprocessed data is then input into the visual-olfaction fusion model for visual feature extraction and olfactory feature extraction, and the visual and olfactory features are fused to output a growth state vector. The growth state vector is then input into a deep reinforcement learning model for growth stage identification and disease risk assessment to obtain a disease risk index. Based on the disease risk index, a corresponding control strategy is output and sent to the execution agency.

[0061] The actuator performs environmental control operations according to the control strategy;

[0062] The cloud platform is used to store historical image data and odor data, train and update deep reinforcement learning models, and send the updated deep reinforcement learning model parameters to the edge computing module.

[0063] In this embodiment, the multimodal data acquisition module includes a microenvironment sensor group, a machine vision acquisition unit, and an olfactory sensing unit. The microenvironment sensor group includes an air temperature and humidity sensor, a CO2 concentration sensor, a light intensity sensor (supporting scattered light detection), and a wind speed sensor, deployed at specific monitoring points on the mushroom house shelves to collect environmental parameters. The machine vision acquisition unit uses a high-resolution RGB camera or an RGB-D depth camera, installed above the cultivation rack or on an inspection robot, to collect image data of the mushroom log surface, including texture, color, and morphology. The olfactory sensing unit, i.e., an electronic nose odor acquisition device, consists of a miniature air pump, a gas pretreatment chamber (dehumidification and dust removal), and a gas sensor array. Sensor selection: The array contains several metal oxide semiconductor sensors, which have cross-sensitivity characteristics to alcohols (characteristic aroma of shiitake mushrooms), sulfides (rotten smell), aldehydes, and earthy compounds (characteristic odor of mold). Function: Real-time capture of the volatile organic compound (VOC) fingerprint spectrum in the mushroom house air. In this embodiment, the gas sensor array can also be replaced with a portable gas chromatography-mass spectrometry (GC-MS) or ion mobility spectrometry (IMS) device. Although GC-MS offers higher accuracy, it is expensive and difficult to deploy online in real time. The use of a MOS sensor array in this embodiment is the optimal solution in terms of cost and real-time performance, but it also covers other odor detection technologies. The machine vision acquisition unit can also employ a hyperspectral camera or a multispectral camera to acquire disease features across more spectral bands.

[0064] The edge computing module is the "brain" of this invention. The edge computing gateway deployed in the mushroom house can be an embedded device based on NVIDIA Jetson or high-performance ARM chips. Alternatively, it can use FPGA (Field Programmable Gate Array) or ASIC dedicated chips to achieve hardware acceleration of the neural network, with the same function as the GPU solution.

[0065] The edge computing module includes a data preprocessing unit, a visual-olfaction fusion model, and a decision-making agent. The data preprocessing unit performs noise reduction and cropping on the acquired images; it also performs baseline correction and feature extraction on the signals acquired by the electronic nose odor acquisition device, extracting features such as response amplitude, response integral, and recovery time. The visual-olfaction fusion model includes a visual branch, an olfactory branch, and a fusion unit. The visual branch uses a lightweight CNN (such as MobileNetV3) to extract visual features of mushroom log color change, mushroom bud diameter, and lesions. The olfactory branch uses a one-dimensional convolutional network (1D-CNN) or a long short-term memory network (LSTM) to extract VOCs time-series features. The fusion unit uses feature concatenation and attention mechanisms to fuse visual and olfactory features, outputting a growth state vector (S). t The decision-making agent is based on deep reinforcement learning (DQN or PPO algorithm), according to the state vector S.t Output control action A t .

[0066] The edge computing module also includes a deep reinforcement learning model, which is responsible not only for synchronizing multi-source data, but also for calculating the reward function of deep reinforcement learning to guide the online self-evolution of the control policy.

[0067] To address the challenge of traditional control algorithms in balancing growth rate, energy consumption, and disease risk, the deep reinforcement learning model in this invention includes a reward function unit. This reward function unit defines a comprehensive reward function R that incorporates disease constraints, as shown in the following formula:

[0068] R = α•ΔG rowth - β•E nergy – γ|E rr | - δ•D risk ;

[0069] In the formula, α•ΔG rowth (Growth Gain Term): Represents the effective biomass increment calculated through visual analysis. The definition differs depending on the stage of edible fungi growth: during the color-changing stage, it refers to the increase rate of the color-changing area of ​​the mycelium film; during the fruiting stage, it refers to the average growth rate of the cap diameter or the proportion of high-quality mushrooms (Grade A mushrooms). β•E nergy (Energy Constraint): Represents the system's total energy consumption at the current moment, including the real-time power of the air conditioner, humidifier, fan, and supplemental lighting. The goal is to find the minimum energy consumption path while satisfying growth requirements. γ|E rr | (Environmental Deviation Penalty): Represents the Euclidean distance between the current environmental parameters (temperature, humidity, CO2) and the target setpoint.

[0070] δ•D risk (Disease Risk Penalty Item): Represents the disease risk index (0~1) output by the visual-olfactory fusion model. When the electronic nose detects an odor (such as earthiness or mold) or the vision detects suspected lesions, δ•D risk A sharp increase in humidity causes the total reward R to become negative (i.e., a penalty), forcing the agent to abandon its current control strategy (such as high temperature and humidity) and switch to a strategy to suppress the disease (such as cooling and ventilation). α, β, γ, δ: These are the dynamic weighting coefficients for each objective, which can be adjusted according to the growth stage. For example, during the mycelium growth stage, the disease risk weight is set to the highest.

[0071] A disease risk penalty term was introduced into the deep reinforcement learning model, and a multi-objective optimization function was constructed that includes growth rate, energy consumption, and health status. The system can dynamically weigh the dual impact of environmental parameters on mycelial growth and the reproduction of miscellaneous bacteria during decision-making. It automatically generates robust control strategies centered on antimicrobial activity during high-risk disease periods, thereby improving the biosafety of the cultivation system.

[0072] The edge computing module also includes an adaptive optimization unit, which performs positive reinforcement and negative penalty based on the reward function.

[0073] The adaptive optimization unit performs online learning based on the above reward function:

[0074] Positive reinforcement: If performing a certain set of actions (such as "turning on fan level 2 + micro-spraying") results in good growth of edible fungi (ΔG) rowth No odor is produced (D) risk Furthermore, it has low energy consumption, and the model receives positive rewards, so the strategy is memorized and reinforced.

[0075] Negative penalty: If, after performing the action, the temperature and humidity meet the standards, but the electronic nose detects an increase in the concentration of musty odor (D... risk The model receives a huge negative reward, thus learning to avoid operating in such high-risk environments.

[0076] To ensure control stability and real-time performance, the edge computing module employs hierarchical control logic. The edge computing module includes upper-layer and lower-layer decision-making units. The upper-layer decision-making unit uses deep reinforcement learning algorithms for high-dimensional, long-term policy planning based on the current complex, multimodal growth state vector. Utilizing DQN or PPO algorithms, it generates optimal target control parameters based on the current state (visual + olfactory + environmental), such as "target humidity 85%, target wind speed 0.5 m / s". The lower-layer decision-making unit uses fuzzy logic to smooth upper-layer commands in response to fluctuations or conflicts in sensor data. For example, when the visual system displays "water shortage" but the electronic nose displays "high humidity and musty odor risk," the fuzzy logic prioritizes the risk of disease, limiting humidification commands to ensure consistency and safety across the entire control system.

[0077] The edge computing module also includes a co-evolution unit, which comprises an edge layer and a data fusion layer. The edge layer is responsible for millisecond-level real-time inference and local response (especially disease blocking). The data fusion layer is responsible for collecting reward value data from the edge, periodically uploading it to the cloud for iterative updates of the global model, and then sending smarter model parameters back to the edge computing module to achieve increasingly intelligent performance.

[0078] Multimodal data acquisition modules (cameras, electronic noses, environmental sensors) are connected to the edge computing module via wired (RS485 / Ethernet) or wireless (Wi-Fi / LoRa) connections. The edge computing module communicates with the PLC controller via an industrial bus (such as Modbus TCP). The PLC controller connects to various actuators (fans, lights, etc.) via relays or analog outputs. The edge computing module connects to the cloud platform via the internet (4G / 5G / fiber optic).

[0079] The implementing agencies include:

[0080] 1) Temperature and humidity control unit: including precision air conditioner, ultrasonic humidifier (for humidification) and dehumidifier.

[0081] 2) Airflow control unit: including variable frequency circulating fan, fresh air exchange fan and electric air valve.

[0082] 3) Light environment unit: LED diffused light source with adjustable spectrum and intensity (used for color conversion and induction of primordia).

[0083] 4) Local alarm device: audible and visual alarm, indicating the specific location of the infected shelf.

[0084] The cloud platform is used to store historical olfactory data, train and update deep reinforcement learning models, and distribute updated deep reinforcement learning model parameters to the edge computing module. Data is displayed through a visual terminal, and users can send remote commands to the cloud platform via the visual terminal.

[0085] The workflow of a factory-scale edible fungus growth monitoring and environmental control system provided in this embodiment of the invention includes:

[0086] An electronic nose odor collection device monitors volatile organic compounds in the air in real time, and a camera periodically captures images of the mushroom logs. If the electronic nose detects that the peak value of "earthy odor" or "rancid odor" exceeds the warning line, the edge node immediately activates the vision module to perform high-frequency scanning and marks the area as a "key concern area." The vision branch analyzes the surface of the mushroom logs in the key concern area to find small discoloration areas. If both visual and olfactory features point to "Trichoderma infection," a disease is diagnosed. Scenario A (Disease Diagnosis): The edge computing module controls the actuator to immediately shut down the circulating fan in the area (to prevent diffusion), activate local negative pressure ventilation, reduce humidity to 60%, and illuminate a red alarm light at that location. Scenario B (Normal Growth - Color Change Stage): The camera detects insufficient reddish-brown film coverage on the mushroom logs, and the odor is normal. The edge computing module controls the temperature and humidity control unit to automatically increase the day-night temperature difference, such as setting the daytime temperature to 25℃ and the nighttime temperature to 15℃, and increases the duration of diffused light from the light environment unit to promote color change.

[0087] The system adopts a three-layer architecture of sensing end-edge end-cloud. The core lies in deploying a vision-olfaction fusion model at the edge end, which utilizes the complementarity of gas sensor array and machine vision to achieve precise control of the growth process of edible fungi and early warning of diseases without visual field.

[0088] The industrialized edible fungi growth monitoring and environmental control system provided in this invention has the following advantages:

[0089] 1. This invention overcomes the lag in visual detection, enabling early disease warning based on VOCs fingerprints. Existing technologies rely on morphological recognition and can only trigger an alarm after lesions appear. This invention introduces a gas sensor array (electronic nose) technology, establishing a non-visual perception mechanism by monitoring changes in the concentration of specific volatile organic compounds (VOCs) released during the metabolism of edible fungi and competing fungi (such as Trichoderma). This mechanism can detect abnormal biochemical indicators during the spore germination stage or hyphal incubation period of pathogens, significantly advancing the risk detection window and achieving a technological leap from phenotypic monitoring to metabolic monitoring.

[0090] 2. A low-latency edge architecture was constructed to achieve real-time blocking of aerosol transmission risks. Addressing the network latency and reliability issues inherent in cloud-based control, this embodiment of the invention employs an edge computing architecture, deploying the visual-olfaction fusion model on the local gateway. Once a peak of pathogenic odor characteristics is detected, the system can achieve millisecond-level actuator responses (such as cutting off recirculating air or activating negative pressure isolation) without waiting for cloud commands. This instantaneous response mechanism effectively blocks aerosol transmission of pathogenic spores through ventilation ducts, significantly reducing the risk of cross-infection caused by delayed regulation of environmental factors.

[0091] 3. A risk-constrained reinforcement learning strategy is proposed to solve the game between growth and disease resistance. Existing control logics mostly employ single PID threshold control or optimization algorithms that only aim at biomass accumulation. This invention introduces a disease risk penalty term into the deep reinforcement learning model, constructing a multi-objective optimization function that includes growth rate, energy consumption, and health status. During decision-making, the system can dynamically weigh the dual impact of environmental parameters on mycelial growth and the proliferation of miscellaneous bacteria, automatically generating a robust control strategy centered on antibacterial activity during high-risk disease periods, thus improving the biosafety of the cultivation system.

[0092] 4. Achieved full-cycle adaptive and precise control based on multimodal feature fusion. Addressing the problem that traditional single-parameter environmental control cannot adapt to the complex physiological cycle of edible fungi, this invention integrates visual features (texture, color, morphology) and olfactory features to achieve automatic identification and status assessment of key physiological stages such as the mycelium growth stage, color change stage, primordium differentiation stage, and fruiting stage. The system can dynamically load corresponding environmental control models (such as a temperature difference stimulation model during the color change stage) based on the identification results, achieving precise spatiotemporal matching between environmental supply and fungal physiological needs.

[0093] like Figure 3 As shown, the second embodiment of the present invention provides a method for monitoring the growth and controlling the environment of industrialized edible fungi, comprising:

[0094] The multimodal data acquisition module collects multidimensional physical and biochemical data of the edible fungi growth environment;

[0095] Determine whether data collection is complete;

[0096] If not, continue data collection;

[0097] If so, the collected data is transmitted to the edge computing module. The edge computing module preprocesses the collected data to obtain preprocessed data. The preprocessed data is then input into the visual-olfaction fusion model for visual feature extraction and olfactory feature extraction, and the visual and olfactory features are fused to output a growth state vector. The growth state vector is then input into a deep reinforcement learning model for growth stage identification and disease risk assessment to obtain a disease risk index and determine whether there is a disease risk (odor or visual abnormality).

[0098] If so, implement emergency containment strategies, such as dehumidification, ventilation, and isolation.

[0099] If not, then implement growth adaptive regulation, such as strategies for bud induction, color change, and temperature difference adjustment;

[0100] Implementation effectiveness and environmental feedback monitoring;

[0101] Calculate the comprehensive reward function;

[0102] The cloud platform aggregates data and updates the deep reinforcement learning model, then distributes the updated deep reinforcement learning model parameters to the edge computing module.

[0103] like Figure 4 As shown, the third embodiment of the present invention provides a method for monitoring and controlling the growth of edible fungi in industrial settings, applicable to edge computing modules. The method includes:

[0104] To obtain multidimensional physical and biochemical data on the growth environment of edible fungi;

[0105] The acquired data is preprocessed to obtain preprocessed data;

[0106] The preprocessed data is input into the visual-olfactory fusion model for visual feature extraction and olfactory feature extraction respectively. The visual and olfactory features are then fused to output a growth state vector.

[0107] The growth state vector is input into a deep reinforcement learning model to identify growth stages and assess disease risk, thus obtaining a disease risk index.

[0108] Based on the disease risk index, the control and execution agency outputs control strategies for regulating the environment.

[0109] The specific methods for inputting the preprocessed data into the visual-olfactory fusion model for visual feature extraction and olfactory feature extraction, and then fusing the visual and olfactory features, include:

[0110] The first neural network model was used to extract visual features of mushroom stick color change, mushroom bud diameter, and lesion.

[0111] Olfactory features are extracted using a second neural network model;

[0112] Visual and olfactory features are fused using feature splicing and attention mechanisms.

[0113] The deep reinforcement learning model defines a comprehensive reward function that includes disease constraints. The formula for the comprehensive reward function is as follows:

[0114] R = α·ΔGrowth - β·Energy – γ|Err| - δ·Drisk;

[0115] Wherein, α•ΔGrowth represents the effective biomass increment calculated through visual analysis, β•Energy represents the system's comprehensive energy consumption at the current moment, γ|Err| represents the Euclidean distance between the current environmental parameters and the target setpoint, and δ•Drisk represents the disease risk index output by the visual-olfactory fusion model.

[0116] The present invention also provides an embodiment of a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, cause the processor to perform the method described in the third embodiment above.

[0117] The computer-readable storage medium can be an internal storage unit of the terminal described in the foregoing embodiments, such as the terminal's hard drive or memory. The computer-readable storage medium can also be an external storage device of the terminal, such as a plug-in hard drive, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the terminal. Furthermore, the computer-readable storage medium can include both internal storage units and external storage devices of the terminal. The computer-readable storage medium is used to store the computer program and other programs and data required by the terminal. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0118] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0119] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the terminals and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0120] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices or units, or may be electrical, mechanical or other forms of connection.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A system for monitoring and controlling the growth of edible fungi in a factory setting, characterized in that, include: Multimodal data acquisition module, edge computing module, actuator and cloud platform, The multimodal data acquisition module is used to collect multidimensional physical and biochemical data of the edible fungi growth environment and transmit the collected data to the edge computing module. The edge computing module is used to preprocess the collected data to obtain preprocessed data. The preprocessed data is then input into the visual-olfaction fusion model for visual feature extraction and olfactory feature extraction, and the visual and olfactory features are fused to output a growth state vector. The growth state vector is then input into a deep reinforcement learning model for growth stage identification and disease risk assessment to obtain a disease risk index. Based on the disease risk index, a corresponding control strategy is output and sent to the execution agency. The actuator performs environmental control operations according to the control strategy; The cloud platform is used to store historical image data and odor data, train and update deep reinforcement learning models, and send the updated deep reinforcement learning model parameters to the edge computing module.

2. The system as described in claim 1, characterized in that, The multimodal data acquisition module includes a microenvironment sensor group, a machine vision acquisition unit, and an olfactory perception unit; The microenvironment sensor group is used to collect ambient temperature and humidity, carbon dioxide concentration, light intensity, and wind speed. The machine vision acquisition unit is used to acquire image data of the surface of the mushroom sticks; The olfactory sensing unit is used to collect odor fingerprints of characteristic volatile organic compounds contained in the air in real time.

3. The system as described in claim 2, characterized in that, The olfactory sensing unit includes a gas sensor array, which is used to detect alcohols, sulfides, aldehydes and geosmin, respectively.

4. The system as described in claim 1, characterized in that, The edge computing module includes a visual-olf fusion model, which includes a visual branch, an olfactory branch, and a fusion unit. The visual branch uses a first neural network model to extract visual features of mushroom stick color change, mushroom bud diameter and lesion. The olfactory branch uses a second neural network model to extract olfactory features; The fusion unit uses a feature splicing and attention mechanism to fuse visual and olfactory features and outputs a growth state vector.

5. The system as described in claim 4, characterized in that, The deep reinforcement learning model includes a reward function unit, which defines a comprehensive reward function that includes disease constraints. The formula for the comprehensive reward function is as follows: R = α·ΔGrowth - β·Energy – γ|Err| - δ·Drisk; Wherein, α•ΔGrowth represents the effective biomass increment calculated through visual analysis, β•Energy represents the system's comprehensive energy consumption at the current moment, γ|Err| represents the Euclidean distance between the current environmental parameters and the target setpoint, and δ•Drisk represents the disease risk index output by the visual-olfactory fusion model.

6. The system as described in claim 5, characterized in that, The edge computing module also includes an adaptive optimization unit, which performs positive reinforcement and negative penalty based on the reward function.

7. A method for monitoring the growth and controlling the environment of industrialized edible fungi, characterized in that, Applicable to edge computing modules, the method includes: To obtain multidimensional physical and biochemical data on the growth environment of edible fungi; The acquired data is preprocessed to obtain preprocessed data; The preprocessed data is input into the visual-olfactory fusion model for visual feature extraction and olfactory feature extraction respectively. The visual and olfactory features are then fused to output a growth state vector. The growth state vector is input into a deep reinforcement learning model to identify growth stages and assess disease risk, thus obtaining a disease risk index. Based on the disease risk index, the control and execution agency outputs control strategies for regulating the environment.

8. The method as described in claim 7, characterized in that, The specific method for inputting the preprocessed data into the visual-olfactory fusion model for visual feature extraction and olfactory feature extraction respectively, and then fusing the visual and olfactory features, includes: The first neural network model was used to extract visual features of mushroom stick color change, mushroom bud diameter, and lesion. Olfactory features are extracted using a second neural network model; Visual and olfactory features are fused using feature splicing and attention mechanisms.

9. The method as described in claim 8, characterized in that, The deep reinforcement learning model defines a comprehensive reward function that includes disease constraints, and the formula for the comprehensive reward function is: R = α·ΔGrowth - β·Energy – γ|Err| - δ·Drisk; Wherein, α•ΔGrowth represents the effective biomass increment calculated through visual analysis, β•Energy represents the system's comprehensive energy consumption at the current moment, γ|Err| represents the Euclidean distance between the current environmental parameters and the target setpoint, and δ•Drisk represents the disease risk index output by the visual-olfactory fusion model.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method as described in any one of claims 7-9.

Images