Edible mushroom bin environment intelligent regulation and control method and system

Abstract

The invention relates to the technical field of environment intelligent regulation and control, in particular to an edible mushroom warehouse environment intelligent regulation and control method and system.The method comprises the steps that edge computing nodes are deployed in each independent cultivation area of a metal mushroom frame in an edible mushroom warehouse, and environment parameters and mushroom form visual data are transmitted to the edge computing nodes through a wireless communication module; performing real-time signal calibration on the environmental parameters through an edge computing node to obtain calibrated environmental parameters; carrying out localized feature extraction on the mushroom form visual data to obtain a mushroom form consistency score; generating an environment regulation and control instruction through a reinforcement learning decision model preset in an edge computing node according to the calibrated environment parameters and the mushroom form consistency score; and the actuator group adjusts the environment parameters according to the environment regulation and control instruction. Accurate sensing of edible mushroom growth environment parameters and reliable evaluation of mushroom forms are achieved, and therefore the intelligent level of edible mushroom planting is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of environmental intelligent regulation, and particularly relates to a method and system for intelligent regulation of an edible mushroom warehouse environment. BACKGROUND

[0002] In the modern large-scale edible mushroom planting industry, the edible mushroom warehouse is the core place for mushroom growth, and accurate regulation of internal environmental parameters is crucial to ensure mushroom quality and yield. The growth process of edible mushrooms is extremely sensitive to environmental factors such as temperature, humidity, light, and carbon dioxide concentration. Any slight environmental fluctuation can affect the growth rate, morphological characteristics, and final quality of mushrooms.

[0003] However, in the actual edible mushroom warehouse environment, the widespread use of metal mushroom shelves poses a serious technical challenge. While providing structural support, the strong electromagnetic properties of metal mushroom shelves significantly interfere with wireless communication signals, resulting in unstable and inaccurate data transmission from environmental parameter sensors, and even signal loss. In addition, the shielding effect of metal mushroom shelves also severely affects the observation effect of image acquisition devices on mushroom morphology, making visual-based mushroom growth state assessment extremely difficult. These problems directly lead to the decline in performance of existing environmental regulation systems in the metal mushroom shelf environment, making it difficult to achieve accurate perception of environmental parameters and reliable assessment of mushroom morphology, thereby affecting the accuracy of dynamic regulation decisions, making it difficult to maintain environmental parameters within the target threshold range, and severely restricting the intelligent level and production efficiency of edible mushroom cultivation.

[0004] To solve the above problems, the present application provides a method for intelligent regulation of the environment of an edible mushroom warehouse, which realizes accurate perception of environmental parameters and reliable assessment of mushroom morphology, thereby significantly improving the intelligent level of edible mushroom cultivation. SUMMARY

[0005] (1) Technical problem to be solved

[0006] The purpose of the present application is to provide a method and system for intelligent regulation of the environment of an edible mushroom warehouse, to solve the problem of strong electromagnetic interference and visual shielding in the metal mushroom shelf environment, which makes it difficult to accurately perceive the environmental parameters of edible mushroom growth and reliably assess the morphology of mushrooms, thereby affecting the accuracy of dynamic regulation of environmental parameters.

[0007] (2) Technical solution

[0008] To achieve the above purpose, on the one hand, the present application provides a method for intelligent regulation of the environment of an edible mushroom warehouse, which comprises:

[0009] S1. Deploy an edge computing node in each independent cultivation area of ​​the metal mushroom rack in the edible mushroom warehouse, wherein the edge computing node integrates a wireless communication module that is resistant to metal interference and is configured with an image acquisition device; collect environmental parameters in real time, and continuously obtain mushroom morphological visual data through the image acquisition device; transmit the environmental parameters and mushroom morphological visual data to the edge computing node via the wireless communication module.

[0010] S2. Performing real-time signal calibration on the environmental parameters through edge computing nodes to obtain calibrated environmental parameters; extracting localized features from the mushroom morphology visual data to obtain a mushroom morphology consistency score.

[0011] S3. Generate environmental control instructions based on the calibrated environmental parameters and the mushroom morphology consistency score through a reinforcement learning decision model pre-set in the edge computing node; send the environmental control instructions to the actuator group, and the actuator group adjusts the environmental parameters according to the environmental control instructions.

[0012] Furthermore, the method of performing real-time signal calibration on the environmental parameters through the edge computing node to obtain calibrated environmental parameters includes:

[0013] The edge computing node obtains an environmental parameter data packet originally transmitted by a multi-source sensor group, and separates a timestamp and a spatial location tag from the environmental parameter data packet.

[0014] An electromagnetic attenuation compensation model is established based on the metal mushroom rack distribution map and the spatial position tags, an electromagnetic carrier signal is extracted from the environmental parameter data packet, and the electromagnetic carrier signal is dynamically corrected according to the electromagnetic attenuation compensation model to obtain preliminary correction data; and the real-time measurement values ​​of the reference sensors deployed in the non-interference area of ​​the metal mushroom rack are read as a verification benchmark.

[0015] The multidimensional feature deviation vector between the preliminary correction data and the verification benchmark is calculated. When the cumulative amplitude of the multidimensional feature deviation vector in the time dimension and the space dimension exceeds the preset dynamic deviation threshold, the spatial position remapping algorithm is triggered according to the current spatial position tag to update the positioning coordinates, and the calibrated environmental parameters that eliminate the signal distortion and positioning drift error caused by the electromagnetic shielding of the metal mushroom rack are output.

[0016] Furthermore, the method for establishing an electromagnetic attenuation compensation model based on the metal mushroom rack distribution map and the spatial position tags includes:

[0017] According to the metal bracket distribution diagram, the metal structure density and geometric configuration of the area corresponding to the current spatial position tag are analyzed, a theoretical electromagnetic attenuation coefficient matrix of the position relative to the reference sensor is calculated through a pre-set electromagnetic field simulation model; a real-time electromagnetic carrier signal strength of the reference sensor under a non-metal interference condition is synchronously collected as a calibration reference value, and the theoretical electromagnetic attenuation coefficient matrix and the calibration reference value are fused and interpolated to obtain an electromagnetic attenuation compensation model.

[0018] Further, the method of extracting an electromagnetic carrier signal from the environmental parameter data packet and dynamically correcting the electromagnetic carrier signal according to the electromagnetic attenuation compensation model to obtain preliminary correction data comprises:

[0019] According to the compensation coefficient corresponding to the spatial position tag index, gain compensation of the electromagnetic carrier signal is performed through the electromagnetic attenuation compensation model; after gain compensation, the electromagnetic carrier signal is subjected to multipath interference filtering processing, and a pure carrier signal strength component reflecting a real environmental parameter is separated out; after the pure carrier signal strength component and other data in the environmental parameter data packet are re-integrated, preliminary correction data is obtained.

[0020] Further, the method of triggering a spatial position remapping algorithm according to a current spatial position tag to update a positioning coordinate comprises:

[0021] The edge computing node retrieves mushroom shape visual data of a corresponding area according to a current spatial position tag, analyzes a shape distortion pattern caused by metal bracket shielding through a pre-trained distortion feature extraction network, and outputs a distortion correction parameter representing a positioning drift error; the distortion correction parameter and the current spatial position tag are analyzed through coordinate remapping to obtain an updated positioning coordinate eliminating a metal shielding error.

[0022] Further, the method of obtaining a mushroom shape consistency score of the mushroom shape visual data through localized feature extraction comprises:

[0023] The mushroom shape visual data is separated into visible light and near-infrared channel data through a multispectral fusion algorithm, and interference of reflected light on the surface of the metal bracket on the image of the mushroom body is eliminated to obtain corrected mushroom shape data; the corrected mushroom shape data is subjected to a double-path deep convolutional neural network to obtain a mushroom shape feature vector, the mushroom shape feature vector is input into a pre-trained fully connected scoring layer, and a mushroom shape consistency score is generated in combination with a current growth stage parameter of the mushroom; a first path of the double-path deep convolutional neural network model is a feature pyramid network extracting a mushroom cap size distribution and surface texture feature, and a second path is a time sequence hollow convolution capturing a continuity feature of a growth trajectory of a mushroom stem.

[0024] Furthermore, the second path is a method for capturing the continuity characteristics of the mushroom root growth trajectory through temporal dilated convolution, which includes:

[0025] The corrected mushroom morphological data is passed through a cascade of time-series dilated convolutional layers in the time dimension with increasing dilation rates to extract the dynamic features of mushroom rhizome growth at different time scales. Multi-scale feature fusion is performed on the dynamic features of mushroom rhizome growth at different time scales to generate a mushroom rhizome growth feature map that integrates multi-scale time series information. The mushroom rhizome growth feature map is input into a residual adaptive gating unit to output a continuity feature vector of the mushroom rhizome growth trajectory.

[0026] The continuity feature vector of the mushroom rhizome growth trajectory is feature-adapted with the parameters of the current growth stage of the mushroom; the weight coefficients of different time scale features in the continuity feature vector of the mushroom rhizome growth trajectory are dynamically adjusted according to the parameters of the current growth stage of the mushroom to obtain a corrected continuity feature vector of the mushroom rhizome growth trajectory; the corrected continuity feature vector of the mushroom rhizome growth trajectory is cross-path feature-spliced ​​with the mushroom crown size distribution and surface texture feature vector output by the first path to obtain a mushroom morphological feature vector.

[0027] Furthermore, the method of generating environmental control instructions based on the calibrated environmental parameters and the mushroom morphology consistency score through a reinforcement learning decision model pre-installed in the edge computing node; issuing the environmental control instructions to the actuator group, and the actuator group adjusting the environmental parameters according to the environmental control instructions includes:

[0028] The calibrated environmental parameters and the mushroom morphology consistency score are subjected to heterogeneous data fusion to construct a feature vector containing the environmental state and the mushroom growth state as the input state of the reinforcement learning decision model; the reinforcement learning decision model analyzes the input state through a policy network and calculates the expected cumulative reward value of each candidate environmental control action, and the expected cumulative reward value is calculated according to a preset reward function; the optimal environmental control action is selected according to the expected cumulative reward value, and the corresponding environmental control instruction is generated.

[0029] After the environmental control instruction passes through the actuator group, the changing trend of the environmental parameters after the execution of the environmental control instruction is predicted; based on the changing trend of the environmental parameters and the current mushroom growth stage parameters, the potential growth risk index of the mushroom is evaluated; when the potential growth risk index exceeds the preset safety index threshold, the suboptimal environmental control action is recalculated according to the input state to generate an alternative environmental control instruction.

[0030] On the other hand, based on the same inventive concept, the present invention also provides an intelligent control system for an edible mushroom warehouse environment, the system comprising: a data acquisition module, a data calibration and analysis module, and an environmental control and management module, wherein the modules are sequentially connected in communication;

[0031] The data acquisition module is used to deploy an edge computing node in each independent cultivation area of ​​the metal mushroom rack in the edible mushroom warehouse. The edge computing node integrates a wireless communication module that is resistant to metal interference and is configured with an image acquisition device; collects environmental parameters in real time, and continuously obtains mushroom morphological visual data through the image acquisition device; and transmits the environmental parameters and mushroom morphological visual data to the edge computing node via the wireless communication module.

[0032] The data calibration and analysis module is used to perform real-time signal calibration on the environmental parameters through the edge computing node to obtain calibrated environmental parameters; and extract localized features from the visual data of the mushroom morphology to obtain a mushroom morphology consistency score.

[0033] The environmental control management module is used to generate environmental control instructions based on the calibrated environmental parameters and the mushroom morphology consistency score through a reinforcement learning decision model pre-set in the edge computing node; the environmental control instructions are sent to the actuator group, and the actuator group adjusts the environmental parameters according to the environmental control instructions.

[0034] (3) Beneficial effects

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. By deploying edge computing nodes with integrated metal-interference-resistant wireless communication modules and image acquisition devices in each independent cultivation area of ​​the metal mushroom rack, environmental parameters and mushroom morphology visual data are collected in real time and stably. Through real-time signal calibration and localized feature extraction technology, the electromagnetic interference and visual occlusion caused by the metal mushroom rack are effectively eliminated, achieving accurate perception of environmental parameters and reliable assessment of mushroom morphology.

[0037] 2. Utilizing the reinforcement learning decision-making model pre-installed in the edge computing node, combined with the calibrated environmental parameters and mushroom morphology consistency score, precise environmental control instructions are generated, and dynamic adjustment of environmental parameters is achieved through the actuator group.

[0038] 3. By predicting the changing trends of environmental parameters and evaluating the potential growth risk indicators of mushrooms, it is possible to recalculate and generate alternative environmental control instructions when necessary, realizing closed-loop optimization of environmental control, ensuring that environmental parameters can be stably maintained within the target threshold range, and improving the timeliness and accuracy of control. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a flowchart of an intelligent control method for an edible fungus warehouse environment according to Example 1 of the present invention.

[0040] Figure 2This is a schematic diagram of the module composition of an intelligent control system for an edible fungus warehouse environment according to Example 2 of the present invention. DETAILED DESCRIPTION

[0041] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0042] Before giving examples, it is necessary to explain the application scenarios of the present invention. The present invention is a method and system for intelligent control of the edible mushroom warehouse environment. When applied to an actual edible mushroom warehouse environment, while the metal mushroom rack provides structural support, its strong electromagnetic characteristics significantly interfere with the wireless communication signal, resulting in unstable and inaccurate data transmission of the environmental parameter sensor. The shielding effect of the metal mushroom rack also seriously affects the observation effect of the image acquisition device on the mushroom morphology, making the evaluation of the mushroom growth status difficult. As a result, the performance of the existing environmental control system in the metal mushroom rack environment is reduced, and the accurate perception of environmental parameters and reliable evaluation of mushroom morphology cannot be achieved, which in turn affects the accuracy of dynamic control decisions and seriously restricts the intelligence level and production efficiency of edible mushroom cultivation.

[0043] Example 1: Figure 1 As shown, this embodiment provides a method for intelligently controlling the environment of an edible mushroom warehouse, the method comprising:

[0044] S1. Deploy an edge computing node in each independent cultivation area of ​​metal mushroom racks within an edible mushroom warehouse. The edge computing node integrates a metal-interference-resistant wireless communication module and is equipped with an image acquisition device. Environmental parameters are collected in real time, while the image acquisition device continuously acquires visual data of mushroom morphology. These environmental parameters and visual data are transmitted to the edge computing node via the wireless communication module. For example, consider a 200-square-meter shiitake mushroom cultivation warehouse with 12 rows of metal mushroom racks, each 8 meters long and 2.5 meters high, arranged in four layers. Each layer is divided into six independent cultivation areas, each measuring 1.3 meters by 1.2 meters. Because the metal racks are made of galvanized steel, they shield and attenuate wireless signals, and reflections from the metal surface can cause multipath interference. An edge computing node is installed in the center of each independent cultivation area. The metal-interference-resistant wireless communication module uses LoRa technology, operates in the 433 MHz frequency band, has a transmit power of 20 dBm, and is equipped with a ceramic antenna to mitigate the effects of the metal environment. The image acquisition device is a 5-megapixel industrial camera equipped with a visible light and near-infrared dual-spectrum lens, mounted on an angle-adjustable bracket to ensure that the growth status of the mushrooms can be observed from the best angle. Real-time collection of environmental parameters is achieved through a sensor group deployed in each cultivation area. The sensor group includes a temperature sensor (accuracy ±0.1°C), a humidity sensor (accuracy ±2%RH), a carbon dioxide concentration sensor (accuracy ±30ppm) and a light intensity sensor (accuracy ±5%). The sensor collects data every 30 seconds and transmits it to the edge computing node. At the same time, the image acquisition device takes a photo of the mushroom growth every 5 minutes. Each shot contains an image of both visible light and near-infrared bands, with an image resolution of 2592×1944 pixels.

[0045] S2. Performing real-time signal calibration on the environmental parameters through edge computing nodes to obtain calibrated environmental parameters; extracting localized features from the mushroom morphology visual data to obtain a mushroom morphology consistency score;

[0046] S3. Generate environmental control instructions based on the calibrated environmental parameters and the mushroom morphology consistency score through a reinforcement learning decision model pre-set in the edge computing node; send the environmental control instructions to the actuator group, and the actuator group adjusts the environmental parameters according to the environmental control instructions.

[0047] The method of performing real-time signal calibration on the environmental parameters through the edge computing node to obtain calibrated environmental parameters includes:

[0048] The edge computing node obtains the environmental parameter data packet originally transmitted by the multi-source sensor group, and separates the timestamp and spatial location tag from the environmental parameter data packet;

[0049] An electromagnetic attenuation compensation model is established based on the metal mushroom rack distribution map and the spatial location tags, an electromagnetic carrier signal is extracted from the environmental parameter data packet, and the electromagnetic carrier signal is dynamically corrected according to the electromagnetic attenuation compensation model to obtain preliminary correction data; and real-time measurement values ​​of reference sensors deployed in non-interference areas of the metal mushroom racks are read as verification benchmarks.

[0050] The multidimensional feature deviation vector between the preliminary correction data and the verification benchmark is calculated. When the cumulative amplitude of the multidimensional feature deviation vector in the time dimension and the space dimension exceeds the preset dynamic deviation threshold, the spatial position remapping algorithm is triggered according to the current spatial position tag to update the positioning coordinates, and the calibrated environmental parameters that eliminate the signal distortion and positioning drift error caused by the electromagnetic shielding of the metal mushroom rack are output.

[0051] When edge computing nodes receive raw environmental parameter data packets from various sensors, they transmit them in a standardized format. Each environmental parameter packet contains three core components: the actual environmental measurement value (e.g., temperature 25.3°C, humidity 78% RH), a timestamp (accurate to the millisecond level), and a spatial location tag (using a three-dimensional coordinate system). The edge computing node first parses the environmental parameter packet and extracts these three pieces of information separately.

[0052] Reference sensors are deployed in an interference-free area outside the mushroom racks, typically more than 3 meters from the nearest metal structure. The reference sensors, using the same model, measure the same environmental parameters in an ideal environment free of metal interference to verify calibration results. The multidimensional feature deviation vector contains multiple components, including temperature deviation, humidity deviation, cumulative deviation in the temporal dimension, and offset in the spatial dimension. For example, if the corrected temperature is 25.8°C and the reference value is 25.3°C, and the humidity correction value is 79% and the reference value is 78%, the temperature deviation is 0.5°C and the humidity deviation is 1%. Preset dynamic deviation thresholds are determined based on the sensitivity of different environmental parameters: ±0.3°C for temperature, ±3%RH for humidity, and ±50ppm for CO2 concentration. If any component of the multidimensional feature deviation vector exceeds the corresponding threshold, and this deviation persists for more than 5 minutes in the temporal dimension or involves more than three adjacent nodes in the spatial dimension, the cumulative magnitude is considered to have exceeded the preset dynamic deviation threshold.

[0053] The method for establishing an electromagnetic attenuation compensation model based on the metal mushroom rack distribution map and the spatial position tags includes:

[0054] According to the metal mushroom rack distribution map, the metal structure density and geometric configuration of the area corresponding to the current spatial position tag are analyzed, and the theoretical electromagnetic attenuation coefficient matrix of this position relative to the reference sensor is calculated through a preset electromagnetic field simulation model; the real-time electromagnetic carrier signal strength of the reference sensor under metal interference-free conditions is synchronously collected as a calibration reference value, and the theoretical electromagnetic attenuation coefficient matrix and the calibration reference value are fused and interpolated to obtain an electromagnetic attenuation compensation model.

[0055] The pre-stored metal mushroom rack distribution map is a 3D digital model that records the complete structural information of the mushroom rack. The map includes the material (Q235 galvanized steel), diameter (48mm), wall thickness (2.5mm), height (2.5m), and precise spatial coordinates of each column. The crossbeam section records the cross-sectional dimensions (40mm×60mm rectangular steel tube), length (8m), and connection node locations. The shelf section records the material (1.5mm galvanized steel sheet), dimensions (1.3m×1.2m), and installation angle. When establishing an electromagnetic attenuation compensation model for a spatial location tag (12.5, 8.3, 1.8), the metal rack distribution map is first searched for all metal structures within a 5-meter radius around that location. The density of the metal structures is calculated using the volume density method. Geometric analysis is more complex, requiring consideration of the shape, orientation, and relative position of the metal structures. The column's geometric configuration parameters include: cross-sectional shape (circular), axial orientation (vertical), and angle (90°) relative to the sensor. The geometric parameters of the laminate include: cross-sectional shape (rectangular), normal orientation (horizontal), and angle relative to the sensor (0°). These parameters directly affect the reflection, scattering, and diffraction characteristics of electromagnetic waves. The pre-built electromagnetic field simulation model, based on the finite-difference time-domain (FDTD) method, accurately simulates the propagation of electromagnetic waves in complex metal environments. During the simulation, the source position is set to the reference sensor coordinates, the receiver position is set to the target sensor coordinates, and the intermediate metal structure is modeled as a scatterer. During the simulation, the attenuation coefficients for the direct path, the primary reflection path, the secondary reflection path, and the diffraction path are calculated. For example, at the position (12.5, 8.3, 1.8), the direct path is blocked by the pillar and has an attenuation coefficient of 0.45; the path reflected by the laminate has an attenuation coefficient of 0.23; and the path diffracted by the beam has an attenuation coefficient of 0.31. The theoretical electromagnetic attenuation coefficient matrix for that location is obtained by combining the power of each path. The calibration reference value collected in real time by the reference sensor is a dynamic value that is updated once per second. The calibration reference value reflects the electromagnetic carrier signal strength in an interference-free area under current environmental conditions. Fusion interpolation calculations utilize a three-dimensional spline interpolation method to combine the theoretical electromagnetic attenuation coefficient matrix with the real-time calibration reference value to establish an electromagnetic attenuation compensation model. The interpolation process considers the combined effects of time, space, and environmental factors. The time factor reflects changes in the electromagnetic environment at different times of the day, the space factor reflects the correlation of electromagnetic characteristics between adjacent locations, and the environmental factor reflects the impact of temperature and humidity on electromagnetic propagation.

[0056] The method of extracting the electromagnetic carrier signal from the environmental parameter data packet and dynamically correcting the electromagnetic carrier signal according to the electromagnetic attenuation compensation model to obtain preliminary correction data includes:

[0057] According to the compensation coefficient corresponding to the spatial position tag index, the electromagnetic carrier signal is gain compensated through the electromagnetic attenuation compensation model; the electromagnetic carrier signal after gain compensation is subjected to multipath interference filtering processing to separate the pure carrier signal strength component reflecting the real environmental parameters, and the pure carrier signal strength component is reintegrated with other data in the environmental parameter data packet to obtain preliminary corrected data.

[0058] When an edge computing node receives an environmental parameter data packet, it extracts the electromagnetic carrier signal carrying the actual environmental information. For example, a temperature sensor's digital output signal consists of multiple components: a digital encoding of the environmental measurement value (e.g., the hexadecimal code 0x19CC for a temperature of 25.3°C), sensor status information (e.g., battery voltage, signal strength indicator), communication protocol header information, and the RF carrier signal carrying this data. The indexing process for spatial location tags is automated. When processing data from a sensor at coordinates (12.5, 8.3, 1.8), it automatically searches the compensation coefficient database for the corresponding record. The database uses a spatial hash index structure, enabling accurate compensation parameter location within milliseconds. Assuming the extracted electromagnetic carrier signal strength is -58dBm, the theoretical electromagnetic carrier signal strength at that location, calculated using a compensation coefficient of 0.68, should be the baseline value (-42dBm) divided by the compensation coefficient: -42 / 0.68 = -61.76dBm. The difference between the actual measured value and the theoretical value is -58 - (-61.76) = 3.76 dB, which requires gain compensation. A 3.76 dB gain is applied to the electromagnetic carrier signal to bring the signal strength back to the theoretical level. In a metal trellis environment, electromagnetic waves undergo multiple reflections and scattering, forming multiple propagation paths to the receiving antenna. The main path signal carries true environmental parameter information, while the multipath signal is a delayed reflection signal, causing signal distortion and data errors. The filtering algorithm utilizes adaptive multipath cancellation technology. The filter employs a finite impulse response structure, suppressing multipath interference through time-domain filtering. For example, for the main path signal, the filter coefficient is 1.0; for the reflected signal with a 150 ns delay, the filter coefficient is -0.3; and for the reflected signal with a 280 ns delay, the filter coefficient is -0.15. This method effectively eliminates multipath interference and extracts the pure carrier signal component. During reassembly, the packet checksum must be updated to ensure reliable data transmission.

[0059] The method for updating positioning coordinates by triggering a spatial position remapping algorithm according to the current spatial position tag includes:

[0060] The edge computing node retrieves the mushroom morphological visual data of the corresponding area according to the current spatial position label, analyzes the morphological distortion pattern caused by the metal mushroom frame occlusion through a pre-trained distortion feature extraction network, and outputs distortion correction parameters that characterize the positioning drift error; the distortion correction parameters and the current spatial position label are analyzed through coordinate remapping to obtain updated positioning coordinates that eliminate the metal occlusion error; the distortion feature extraction network adopts a composite structure of a spatial attention mechanism and residual deformation convolution. The spatial attention mechanism focuses on the image distortion features of the edge area of ​​the metal frame, and the residual deformation convolution compensates for the morphological geometric distortion caused by signal drift.

[0061] When the cumulative amplitude of the multidimensional feature deviation vector in both the temporal and spatial dimensions exceeds the preset dynamic deviation threshold, it indicates that the current sensor's spatial positioning may have drifted. For example, for the location tag (12.5, 8.3, 1.8), if the temperature deviation at that location consistently exceeds the ±0.3°C threshold for 5 minutes, and similar deviations are observed at neighboring nodes, it is determined that positioning drift may have occurred, triggering the spatial position remapping algorithm. The edge computing node first retrieves the visual data of the mushroom morphology in the corresponding area based on the current spatial location tag, enabling rapid localization of all images captured within the last 30 minutes. The distortion feature extraction network is a deep learning model specifically designed for image distortion in metallic environments. Its input is the original mushroom morphology image, and its output is distortion correction parameters representing spatial positioning offsets. The spatial attention mechanism is specifically designed to identify image distortion at the edges of metal frames, which typically manifest as high-frequency noise, sudden brightness changes, or geometric distortion. In practice, the edges of metal frames are most susceptible to light reflection and refraction, resulting in localized image distortion. The attention mechanism automatically focuses on these critical areas by learning the importance weights of different regions. For example, in a mushroom image, the weight of the metal pillar edge may reach 0.85, while the weight of the mushroom body is 0.92, and the weight of the background area is only 0.23. Residual deformable convolution is the core component of the network, specifically designed to compensate for morphological geometric distortion caused by signal drift. Traditional standard convolution kernels use a fixed rectangular grid, while deformable convolution's sampling points can adaptively adjust their positions based on the input content. When the mushroom outline in the image is detected to be shifted southeastward, the deformable convolution automatically adjusts the distribution of sampling points to compensate for this geometric distortion. Suppose that the mushroom in a mushroom morphological image should be located at the center of the image (1296, 972), but due to optical distortion caused by metal occlusion, the visual center of the mushroom is shifted to (1318, 964). By analyzing this shift pattern, the network outputs distortion correction parameters: horizontal offset dx = 22 pixels, vertical offset dy = -8 pixels, and rotation angle θ = 1.2 degrees. Considering the correspondence between image resolution and actual physical size, the pixel offset is converted to actual spatial offset. Assuming each pixel corresponds to an actual distance of 0.5 mm, the horizontal offset is 22 × 0.5 = 11 mm, and the vertical offset is -8 × 0.5 = -4 mm. These values ​​constitute the distortion correction parameters for the positioning drift vector. The coordinate remapping function receives the current spatial position label and the distortion correction parameters as input and generates updated positioning coordinates.

[0062] The method for extracting localized features from the mushroom morphology visual data to obtain a mushroom morphology consistency score includes:

[0063] The mushroom morphology visual data is separated into visible light and near-infrared channel data through a multispectral fusion algorithm, and the interference of the reflected light on the metal mushroom rack surface on the mushroom body image is eliminated to obtain corrected mushroom morphology data; the corrected mushroom morphology data is passed through a dual-path deep convolutional neural network to obtain a mushroom morphology feature vector, which is input into a pre-trained fully connected scoring layer and combined with the parameters of the current growth stage of the mushroom to generate a mushroom morphology consistency score; the first path of the dual-path deep convolutional neural network model is to extract the mushroom crown size distribution and surface texture characteristics through a feature pyramid network, and the second path is to capture the continuity characteristics of the mushroom root growth trajectory through temporal void convolution.

[0064] Mushroom morphological visual data consists of images from two spectral channels: visible light (400-700nm) and near-infrared (850nm). A multispectral fusion algorithm geometrically registers the two-channel images to eliminate parallax caused by the positional differences between the two cameras. The reflectivity of the metal rack surface in the visible and near-infrared bands differs significantly, with a visible reflectivity of approximately 0.6 and a near-infrared reflectivity of approximately 0.8. The corrected mushroom morphological data preserves the true morphological information of the mushrooms while eliminating the visual interference of the metallic surroundings. A two-pathway deep convolutional neural network employs a parallel processing architecture. The first path extracts features from the cap, while the second path extracts features from the rhizome. The outputs of these two paths are then fused. The first path employs a feature pyramid network (FPN) architecture. The feature extraction process includes cap size distribution calculation and surface texture analysis. The size distribution is calculated using an ellipse fitting method to determine the major axis, minor axis, and eccentricity of the cap. Surface texture statistical features such as contrast, uniformity, and entropy are calculated using a gray-level co-occurrence matrix. Feature vectors from the two paths are concatenated and dimensionally reduced in a fully connected layer. For example, the first pathway outputs a 2048-dimensional cap feature vector, and the second pathway outputs a 2048-dimensional rhizome feature vector. After concatenation, they produce a 4096-dimensional comprehensive feature vector. The fully connected scoring layer consists of three layers: 4096→1024→256→1. Reluctant Unit (ReLU) is used as the activation function, and the final layer uses a Sigmoid function to output a continuous score between 0 and 1. This score is then multiplied by 100 to obtain the final mushroom morphological consistency score. The scoring criteria are: 90-100 indicates excellent quality, with a full cap and a robust rhizome; 80-89 indicates good quality, with generally normal morphology; 70-79 indicates acceptable quality, with minor defects; and scores below 70 indicate that environmental conditions require improvement.

[0065] The second path is a method for capturing the continuity characteristics of the mushroom root growth trajectory through temporal dilated convolution, which includes:

[0066] The corrected mushroom morphological data is passed through a cascade of time-series dilated convolutional layers in the time dimension with increasing dilation rates to extract dynamic features of mushroom rhizome growth at different time scales; multi-scale feature fusion is performed on the dynamic features of mushroom rhizome growth at different time scales to generate a mushroom rhizome growth feature map that integrates multi-scale temporal information; the mushroom rhizome growth feature map is input into a residual adaptive gating unit to output a continuity feature vector of the mushroom rhizome growth trajectory;

[0067] The continuity feature vector of the mushroom rhizome growth trajectory is feature-adapted with the parameters of the current growth stage of the mushroom; the weight coefficients of different time scale features in the continuity feature vector of the mushroom rhizome growth trajectory are dynamically adjusted according to the parameters of the current growth stage of the mushroom to obtain a corrected continuity feature vector of the mushroom rhizome growth trajectory; the corrected continuity feature vector of the mushroom rhizome growth trajectory is cross-path feature-spliced ​​with the mushroom crown size distribution and surface texture feature vector output by the first path to obtain a mushroom morphological feature vector.

[0068] The design of the cascaded temporal dilated convolution layers is based on the multi-time scale characteristics of mushroom stem growth. The images taken every 2 hours in the last 7 days are collected to form an image sequence containing 84 time points. After preprocessing, the image blocks of the mushroom stem region are extracted, with a uniform size of 128x128 pixels. The first layer of dilated convolution has a dilation rate of 1, which is actually a standard convolution, used to extract the basic morphological features of the mushroom stem. The convolution kernel size is 3x3, and the channel number is 64. This layer mainly identifies the edge profile, surface texture, and basic geometric shape of the stem. The processed feature map can accurately describe the morphological state of the stem at a single time point. The second layer of dilated convolution has a dilation rate of 2, with a receptive field of 5x5, which can capture the growth changes within a 2-4 hour time window. This layer mainly detects the small changes in stem length, gradual adjustments in thickness, and evolution of surface color. By comparing the features of adjacent time points, it can identify the stem elongation of about 0.1-0.3 mm per hour. The third layer of dilated convolution has a dilation rate of 4, with a receptive field of 9x9, corresponding to an 8-12 hour time window. This layer captures the medium-term growth trends of the stem, such as the periodic changes in growth speed, adjustments in curvature, and the appearance of branching. By analyzing the features of this time scale, it can predict the growth state of the mushroom within the next 6-12 hours. The fourth layer of dilated convolution has a dilation rate of 8, with a receptive field of 17x17, covering a 24-36 hour time range. This layer mainly analyzes the daily cycle patterns of stem growth, such as the differences in growth speed during the day and night, the effects of temperature and humidity changes on growth, etc. Healthy mushroom stems usually grow fastest between 2-6 am and slower between 14-18 pm. The fifth layer of dilated convolution has a dilation rate of 16, with a receptive field of 33x33, corresponding to a 2-4 day time window. This layer captures the stage characteristics of stem growth, such as the morphological transitions from the bud stage to the young mushroom stage, and from the young mushroom stage to the mature stage. Mushrooms in different growth stages exhibit different morphological features: the bud stage has a thicker and shorter stem, the young mushroom stage has rapid elongation, and the mature stage tends to be stable. The sixth layer of dilated convolution has a dilation rate of 32, with a receptive field of 65x65, covering the entire 7-day observation period. By extracting the complete trajectory features of stem growth, including overall growth speed, morphological change patterns, and identification of abnormal growth patterns. By analyzing the complete growth trajectory, it can evaluate the overall health status of the mushroom. The multi-scale feature fusion adopts a channel attention mechanism to weight and combine features of different time scales. The fusion process considers the importance and reliability of features at each time scale. The weight coefficients for short-term features (dilation rates 1-4) are [0.25, 0.30, 0.25, 0.20], for medium-term features (dilation rates 8-16) are [0.35, 0.45], and for long-term features (dilation rate 32) is 0.20. The generated mushroom stem growth feature map after fusion is a multi-dimensional tensor, containing complete temporal growth information.The numerical value of each position in the mushroom rhizome growth feature map reflects the growth activity and trend of the corresponding spatial position at different time scales. The residual adaptive gating unit receives the fused mushroom rhizome growth feature map as input and filters noise interference through a gating mechanism.

[0069] The feature adaptation process dynamically adjusts the feature weights according to the current growth stage parameters of the mushroom. Five main stages of Lentinula edodes growth are predefined: mycelium stage (0-7 days), primordium stage (8-14 days), pinhead stage (15-21 days), young mushroom stage (22-28 days), and mature stage (29-35 days). Each stage corresponds to a different feature weight configuration. In the pinhead stage, more attention is paid to short-term growth features, with a weight configuration of [0.4, 0.3, 0.2, 0.1]; in the young mushroom stage, mid-term features are more important, with a weight configuration of [0.2, 0.3, 0.4, 0.1]; in the mature stage, long-term stability features have the highest weight, with a configuration of [0.1, 0.2, 0.3, 0.4].

[0070] The corrected mushroom rhizome growth trajectory continuity feature vector comprehensively describes the growth characteristics of the stipe in the time dimension, and is spliced with the cap feature vector output by the first path to form a complete mushroom morphology feature vector.

[0071] The calibrated environmental parameters and the mushroom morphology consistency score are used to generate environmental control instructions through a reinforcement learning decision model pre-installed in the edge computing node; the method for adjusting environmental parameters by the actuator group according to the environmental control instructions comprises:

[0072] The calibrated environmental parameters and the mushroom morphology consistency score are used to generate environmental control instructions through a reinforcement learning decision model pre-installed in the edge computing node; the method for adjusting environmental parameters by the actuator group according to the environmental control instructions comprises:

[0073] After the environmental control instructions are executed by the actuator group, the change trend of the environmental parameters after the execution of the environmental control instructions is predicted; the potential growth risk index of the mushroom is evaluated according to the change trend of the environmental parameters and the current growth stage parameters of the mushroom; when the potential growth risk index exceeds the preset safety index threshold, the suboptimal environmental control action is recalculated according to the input state, and an alternative environmental control instruction is generated.

[0074] According to the change trend of the environmental parameters, the change rate of the key environmental parameters in a preset future time window and the maximum amplitude deviating from the target threshold interval are extracted, a predefined risk mapping rule library is called in combination with the current mushroom growth stage parameters, the predefined risk mapping rule library defines the correlation between the change characteristics of each environmental parameter and the potential growth risk under different mushroom growth stages, a corresponding single risk indicator is matched in the predefined risk mapping rule library according to the change rate, the maximum amplitude and the current mushroom growth stage parameters, and the single risk indicator is linearly weighted to obtain the potential growth risk indicator of the mushroom.

[0075] Heterogeneous data fusion is the first step of the decision-making process, and it is necessary to effectively integrate the calibrated environmental parameters and the mushroom morphology consistency score. Due to the differences in data types and dimensions, standardization processing is required. The standardization process adopts the Z-score method to convert each parameter into a standard normal distribution with a mean of 0 and a standard deviation of 1. Each element in the feature vector of the environmental state and the mushroom growth state represents the deviation of the current environment or mushroom state from the ideal state, with positive values indicating higher than the average level and negative values indicating lower than the average level.

[0076] The reinforcement learning decision model adopts an Actor-Critic architecture, including a policy network (Actor) and a value network (Critic). The policy network is responsible for selecting the optimal environmental regulation action according to the input state, and the value network is responsible for evaluating the value of the state and the quality of the action. The reinforcement learning decision model is trained through historical regulation cases to learn the complex relationship between environmental parameters and mushroom growth status. Twelve basic environmental regulation actions are defined: temperature increase / decrease (±0.5°C), humidity increase / decrease (±3% RH), carbon dioxide concentration increase / decrease (±100 ppm), light intensity increase / decrease (±20 lux), ventilation volume increase / decrease, and maintaining the status quo. The calculation of the expected cumulative reward value is based on a pre-set reward function. The reward function considers three aspects: environmental parameter stability, mushroom growth quality, and energy efficiency. For example, the expected cumulative reward value of each candidate environmental regulation action is calculated: the expected reward of increasing the temperature by 0.5°C is -2.3 points (the current temperature is already high), the expected reward of reducing the humidity by 3% RH is +6.8 points (helps to stabilize the environment), the expected reward of increasing the carbon dioxide concentration by 100 ppm is +4.2 points (beneficial to mushroom growth), and the expected reward of maintaining the status quo is +2.1 points. According to the expected cumulative reward value, the action with the highest reward value is selected as the optimal environmental regulation action. In this example, reducing the humidity by 3% RH obtains the highest reward (+6.8 points), so it is selected as the optimal action. The corresponding environmental regulation instruction contains specific execution parameters: for example, the target humidity is set to 75.2% RH, the humidifier is turned off for 15 minutes, the exhaust fan is started at 60% power, and the expected time to reach is 8 minutes. The prediction of environmental parameter change trend adopts a time series analysis method, based on the past 24 hours of environmental data, using an ARIMA model to predict the change trajectory of each parameter in the next 2 hours.

[0077] Analyze the predicted environmental change trend and identify factors that may adversely affect mushroom growth. Rapid humidity decrease can cause the mushroom surface to lose water, especially during the young mushroom stage, which is a significant risk. For example, if the potential growth risk indicator (3.2) is below the preset safety indicator threshold (5.0), it is determined that the current control scheme is safe and no alternative instructions need to be generated. The environmental control instructions are officially issued to the actuator group, and the actuator group begins to execute the corresponding control operation after receiving the instructions. The actual changes in environmental parameters are continuously monitored and compared with the predicted trend. The rate of change of key environmental parameters within the preset future time window is extracted from the predicted environmental parameter change trajectory. The maximum deviation from the target threshold interval requires reference to the preset optimal growth environment standard. For example, the current mushroom growth stage parameter shows that the mushroom is in the 3rd day of the young mushroom stage, which is a relatively sensitive stage to environmental changes. The young mushroom is growing rapidly and the cell division of the stem is active, making it more sensitive to humidity changes. According to this stage information, the corresponding risk mapping rules are called. The pre-defined risk mapping rule base is a knowledge base established based on a large amount of experimental data and expert knowledge, which contains the quantitative relationship between environmental parameter changes and growth risks at different growth stages. The rule base uses an "IF-THEN" structure, and each rule defines the risk score under specific conditions. When the potential risk indicator exceeds the preset safety indicator threshold of 5.0 points, an alternative environmental control instruction is immediately generated.

[0078] Example 2: Based on the same inventive concept, as Figure 2 shown, the present embodiment also provides an intelligent environmental control system for edible mushroom warehouse, which comprises a data acquisition module, a data calibration and analysis module, and an environmental control and management module, which are sequentially connected in communication between each other;

[0079] The data acquisition module is used to deploy edge computing nodes in each independent cultivation area of the edible mushroom warehouse. The edge computing nodes are integrated with anti-metal interference wireless communication modules and configured with image acquisition devices. Real-time environmental parameters are collected, and mushroom morphology visual data are continuously acquired through the image acquisition devices. The environmental parameters and mushroom morphology visual data are transmitted to the edge computing nodes through the wireless communication module;

[0080] The data calibration and analysis module is used to calibrate the environmental parameters in real time through the edge computing nodes to obtain calibrated environmental parameters, and to extract the mushroom morphology consistency score from the mushroom morphology visual data through localized feature extraction.

[0081] The environmental control and management module is used to generate environmental control instructions through the reinforcement learning decision model pre-installed in the edge computing nodes based on the calibrated environmental parameters and the mushroom morphology consistency score. The environmental control instructions are issued to the actuator group, and the actuator group adjusts the environmental parameters according to the environmental control instructions.

[0082] It should be noted that, as to the system in the above-mentioned embodiments, the specific manner in which each module performs operations has been described in detail in the embodiments related to the method, and will not be described in detail here.

[0083] Finally, it should be noted that although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or equivalently replace some of the technical features, as long as they are within the spirit and principles of the present application. Any modification, equivalent replacement, improvement, etc. made shall be included in the protection scope of the present application.

Claims

1. A method for intelligently controlling the environment of an edible mushroom warehouse, characterized in that: The method comprises: Deploy an edge computing node in each independent cultivation area of ​​a metal mushroom rack in an edible mushroom warehouse, wherein the edge computing node integrates a wireless communication module that is resistant to metal interference and is equipped with an image acquisition device; collect environmental parameters in real time, and continuously obtain visual data of mushroom morphology through the image acquisition device; transmit the environmental parameters and visual data of mushroom morphology to the edge computing node via the wireless communication module; The environmental parameters are calibrated in real time through edge computing nodes to obtain calibrated environmental parameters; the mushroom morphology visual data is extracted through localized features to obtain a mushroom morphology consistency score; The calibrated environmental parameters and the mushroom morphology consistency score are used to generate environmental control instructions through a reinforcement learning decision model pre-set in the edge computing node; the environmental control instructions are sent to the actuator group, and the actuator group adjusts the environmental parameters according to the environmental control instructions.

2. The method for intelligently controlling the environment of an edible mushroom warehouse according to claim 1, characterized in that: The method of performing real-time signal calibration on the environmental parameters through the edge computing node to obtain calibrated environmental parameters includes: The edge computing node obtains the environmental parameter data packet originally transmitted by the multi-source sensor group, and separates the timestamp and spatial location tag from the environmental parameter data packet; An electromagnetic attenuation compensation model is established based on the metal mushroom rack distribution map and the spatial location tags, an electromagnetic carrier signal is extracted from the environmental parameter data packet, and the electromagnetic carrier signal is dynamically corrected according to the electromagnetic attenuation compensation model to obtain preliminary correction data; and real-time measurement values ​​of reference sensors deployed in non-interference areas of the metal mushroom racks are read as verification benchmarks. The multidimensional feature deviation vector between the preliminary correction data and the verification benchmark is calculated. When the cumulative amplitude of the multidimensional feature deviation vector in the time dimension and the space dimension exceeds the preset dynamic deviation threshold, the spatial position remapping algorithm is triggered according to the current spatial position tag to update the positioning coordinates, and the calibrated environmental parameters that eliminate the signal distortion and positioning drift error caused by the electromagnetic shielding of the metal mushroom rack are output.

3. The method for intelligently controlling the environment of an edible mushroom warehouse according to claim 2, characterized in that: The method for establishing an electromagnetic attenuation compensation model based on the metal mushroom rack distribution map and the spatial position tags includes: According to the metal mushroom rack distribution map, the metal structure density and geometric configuration of the area corresponding to the current spatial position tag are analyzed, and the theoretical electromagnetic attenuation coefficient matrix of this position relative to the reference sensor is calculated through a preset electromagnetic field simulation model; the real-time electromagnetic carrier signal strength of the reference sensor under metal interference-free conditions is synchronously collected as a calibration reference value, and the theoretical electromagnetic attenuation coefficient matrix and the calibration reference value are fused and interpolated to obtain an electromagnetic attenuation compensation model.

4. The method for intelligently controlling the environment of an edible mushroom warehouse according to claim 2, characterized in that: The method of extracting the electromagnetic carrier signal from the environmental parameter data packet and dynamically correcting the electromagnetic carrier signal according to the electromagnetic attenuation compensation model to obtain preliminary correction data includes: According to the compensation coefficient corresponding to the spatial position tag index, the electromagnetic carrier signal is gain compensated through the electromagnetic attenuation compensation model; the electromagnetic carrier signal after gain compensation is subjected to multipath interference filtering processing to separate the pure carrier signal strength component reflecting the real environmental parameters, and the pure carrier signal strength component is reintegrated with other data in the environmental parameter data packet to obtain preliminary corrected data.

5. The method for intelligently controlling the edible mushroom warehouse environment according to claim 2, characterized in that: The method for updating positioning coordinates by triggering a spatial position remapping algorithm according to the current spatial position tag includes: The edge computing node retrieves the mushroom morphological visual data of the corresponding area according to the current spatial position label, analyzes the morphological distortion pattern caused by the metal mushroom rack occlusion through a pre-trained distortion feature extraction network, and outputs distortion correction parameters that characterize the positioning drift error; the distortion correction parameters and the current spatial position label are analyzed through coordinate remapping to obtain updated positioning coordinates that eliminate the metal occlusion error.

6. The method for intelligently controlling the environment of an edible mushroom warehouse according to claim 1, characterized in that: The method for extracting localized features from the mushroom morphology visual data to obtain a mushroom morphology consistency score includes: The mushroom morphology visual data is separated into visible light and near-infrared channel data through a multispectral fusion algorithm, and the interference of the reflected light on the metal mushroom rack surface on the mushroom body image is eliminated to obtain corrected mushroom morphology data; the corrected mushroom morphology data is passed through a dual-path deep convolutional neural network to obtain a mushroom morphology feature vector, which is input into a pre-trained fully connected scoring layer and combined with the parameters of the current growth stage of the mushroom to generate a mushroom morphology consistency score; the first path of the dual-path deep convolutional neural network model is to extract the mushroom crown size distribution and surface texture characteristics through a feature pyramid network, and the second path is to capture the continuity characteristics of the mushroom root growth trajectory through temporal void convolution.

7. The method for intelligently controlling the environment of an edible mushroom warehouse according to claim 6, characterized in that: The second path is a method for capturing the continuity characteristics of the mushroom root growth trajectory through temporal dilated convolution, which includes: The corrected mushroom morphological data is passed through a cascade of time-series dilated convolutional layers in the time dimension with increasing dilation rates to extract dynamic features of mushroom rhizome growth at different time scales; multi-scale feature fusion is performed on the dynamic features of mushroom rhizome growth at different time scales to generate a mushroom rhizome growth feature map that integrates multi-scale temporal information; the mushroom rhizome growth feature map is input into a residual adaptive gating unit to output a continuity feature vector of the mushroom rhizome growth trajectory; The continuity feature vector of the mushroom rhizome growth trajectory is feature-adapted with the parameters of the current growth stage of the mushroom; the weight coefficients of different time scale features in the continuity feature vector of the mushroom rhizome growth trajectory are dynamically adjusted according to the parameters of the current growth stage of the mushroom to obtain a corrected continuity feature vector of the mushroom rhizome growth trajectory; the corrected continuity feature vector of the mushroom rhizome growth trajectory is cross-path feature-spliced ​​with the mushroom crown size distribution and surface texture feature vector output by the first path to obtain a mushroom morphological feature vector.

8. The method for intelligently controlling the environment of an edible mushroom warehouse according to claim 1, characterized in that: The calibrated environmental parameters and the mushroom morphology consistency score are used to generate environmental control instructions through a reinforcement learning decision model pre-installed in the edge computing node; The method of sending the environmental control instruction to the actuator group, and the actuator group adjusting the environmental parameters according to the environmental control instruction includes: The calibrated environmental parameters and the mushroom morphology consistency score are subjected to heterogeneous data fusion to construct a feature vector containing the environmental state and the mushroom growth state as the input state of the reinforcement learning decision model; the reinforcement learning decision model analyzes the input state through a policy network and calculates the expected cumulative reward value of each candidate environmental control action, wherein the expected cumulative reward value is calculated according to a preset reward function; the optimal environmental control action is selected based on the expected cumulative reward value, and the corresponding environmental control instruction is generated; After the environmental control instruction passes through the actuator group, the changing trend of the environmental parameters after the execution of the environmental control instruction is predicted; based on the changing trend of the environmental parameters and the current mushroom growth stage parameters, the potential growth risk index of the mushroom is evaluated; when the potential growth risk index exceeds the preset safety index threshold, the suboptimal environmental control action is recalculated according to the input state to generate an alternative environmental control instruction.

9. An intelligent control system for edible mushroom warehouse environment, characterized in that: The system includes: a data acquisition module, a data calibration and analysis module, and an environmental control and management module, and each module is sequentially connected to communicate with each other; A data acquisition module is configured to deploy an edge computing node in each independent cultivation area of ​​a metal mushroom rack in an edible mushroom warehouse. The edge computing node integrates a wireless communication module that is resistant to metal interference and is configured with an image acquisition device. The module collects environmental parameters in real time and continuously obtains visual data of mushroom morphology through the image acquisition device. The module transmits the environmental parameters and visual data of mushroom morphology to the edge computing node via the wireless communication module. A data calibration and analysis module is used to perform real-time signal calibration on the environmental parameters through edge computing nodes to obtain calibrated environmental parameters; and extract localized features from the visual data of the mushroom morphology to obtain a mushroom morphology consistency score; The environmental control management module is used to generate environmental control instructions based on the calibrated environmental parameters and the mushroom morphology consistency score through a reinforcement learning decision model pre-set in the edge computing node; the environmental control instructions are sent to the actuator group, and the actuator group adjusts the environmental parameters according to the environmental control instructions.

Images