Multi-channel light environment monitoring and information transmission equipment

Abstract

The invention discloses a multi-channel light environment monitoring and information transmission device, and relates to the field of light monitoring. The single-chip microcomputer module is used for filtering high-frequency noise on a power supply and storing energy, and providing a target impedance backflow path for a high-frequency signal; the spectrum detection module is used for collecting luminous flux values of spectrums with different wavelengths and is communicated with the single chip microcomputer module; the light intensity detection module is used for collecting illumination intensity of spectrums with different wavelengths and is communicated with the single chip microcomputer module; a data uploading module; the data arrangement and analysis module is used for acquiring growth stages of crops; based on the light environment data and in combination with a causal structure learning model, constructing a spectral regulation causal network in different growth stages; and extracting key spectrum factor contributions in multiple space-time scenes by adopting an adaptive interpretation type graph neural network, and finally generating optimal spectrum regulation and control suggestions in stages and parcels. According to the invention, the fineness and interpretability of spectrum regulation and control are improved.

Description

Technical Field

[0001] This invention relates to the field of optical monitoring, and more specifically, to a multi-channel optical environment monitoring and information transmission device. Background Technology

[0002] Light is a crucial environmental factor in plant growth. Different plants have varying spectral requirements for visible light at different stages. Compared to traditional agriculture, greenhouse agriculture requires data-driven cultivation, and light environment data is particularly important. Growers use existing light environment data to accurately collect, analyze, and evaluate it, thereby improving crop yield and quality. However, current technologies for collecting and utilizing light environment data still have the following shortcomings:

[0003] (1) Existing solutions mostly collect data in a decentralized manner and then store it simply. They cannot automatically integrate the physical quantities of light environment with agricultural structural information such as crop growth stage, plot code, and time tag, resulting in data silos and difficulty in accurately mapping subsequent analysis, which seriously restricts the basic conditions for refined regulation and intelligent analysis.

[0004] (2) Existing methods for analyzing crop outcome variables of agricultural environmental factors mainly rely on correlation statistics, traditional regression analysis or static black box machine learning, which cannot effectively reveal the complex causal network structure between multiple variables and multiple stages.

[0005] (3) Traditional data analysis or prediction models mostly use black-box neural networks or shallow algorithms, which lack interpretability for the multi-temporal and multi-level control effects between multiple parameters of the light environment and yield and quality. Most models only provide outputs in the overall or static scenario and lack a practical intelligent spectral control recommendation table.

[0006] No effective solutions have yet been proposed to address the problems in the relevant technologies. Summary of the Invention

[0007] In response to the problems in related technologies, this invention proposes a multi-channel optical environment monitoring and information transmission device to overcome the aforementioned technical problems existing in the existing related technologies.

[0008] Therefore, the specific technical solution adopted by the present invention is as follows:

[0009] Multi-channel optical environment monitoring and information transmission equipment, including:

[0010] The power module is used to regulate the voltage of the power supply according to the voltage regulator;

[0011] The microcontroller module is used to filter out high-frequency noise on the power supply and store energy, while providing a target impedance return path for high-frequency signals.

[0012] The spectral detection module is used to collect the luminous flux values ​​of different wavelengths and communicate with the microcontroller module.

[0013] The light intensity detection module is used to collect the light intensity of different wavelength spectra and communicate with the microcontroller module.

[0014] The data upload module is used to standardize light environment data, including luminous flux and illuminance, and upload it to the cloud server to generate multi-dimensional charts.

[0015] The data processing and analysis module is used to obtain the growth stages of crops; based on light environment data and combined with a causal structure learning model, a spectral regulation causal network for different growth stages is constructed; an adaptive interpretive graph neural network is used to extract the contribution of key spectral factors in multi-temporal and spatial scenarios, and finally generate optimal spectral regulation suggestions for different stages and plots.

[0016] Furthermore, the microcontroller module includes: chip U3, inductor L1, capacitor C6, and capacitor C7;

[0017] In this configuration, the eighth terminal of chip U3 is connected in sequence to one end of inductor L1, one end of capacitor C6, and one end of capacitor C7. The other end of inductor L1 is connected to the ninth terminal of chip U3, and the other end of capacitor C6 is connected to the other end of capacitor C7.

[0018] Furthermore, the power module includes: socket DC1, USB1, button SW1, diode D1, diode D2, capacitor C3, capacitor C4, chip U1, capacitor C1, capacitor C2, LED1 and resistor R1;

[0019] The USB1 is connected to the second terminal of the button SW1. The third terminal of the button SW1 is connected to the cathode of diode D2 and the anode of diode D1 in sequence. The cathode of diode D1 is connected to one end of capacitor C3, one end of capacitor C4 and the third terminal of chip U1 in sequence. The anode of diode D2 is connected to the fifth terminal of USB1, the other end of capacitor C3, the other end of capacitor C4, the first terminal of chip U1, one end of capacitor C1, one end of capacitor C2 and one end of resistor R1 in sequence. The other end of resistor R1 is connected to the cathode of LED1. The anode of LED1 is connected to the other end of capacitor C2, the other end of capacitor C1 and the fourth terminal of chip U1 in sequence. A socket DC1 is provided on the side of USB1.

[0020] Furthermore, the spectral detection module includes: terminal H2, capacitor C8, capacitor C9, resistor R14, resistor R15, resistor R16 and resistor R17;

[0021] Specifically, the second end of terminal H2 is connected to one end of capacitor C8 and one end of capacitor C9 in sequence, the other end of capacitor C8 is connected to the other end of capacitor C9, the third end of terminal H2 is connected to resistor R14, the fourth end of terminal H2 is connected to resistor R15, the fifth end of terminal H2 is connected to resistor R16, and the sixth end of terminal H2 is connected to resistor R17.

[0022] Furthermore, the light intensity detection module includes:

[0023] The ambient light signal is acquired by an ambient light sensor combined with an infrared photodiode. After being digitized by an analog-to-digital converter module, infrared compensation is performed using a digital signal processing module.

[0024] Furthermore, obtaining the growth stages of crops includes:

[0025] Crop images are organized by plot and time, and matched with light flux and light intensity to obtain a structured dataset including plot number, timestamp, light environment parameters and crop images;

[0026] The deep convolutional neural network model is used to extract features from crop images, automatically identify and classify the current growth morphology features of crops, and fuse them with crop physiological measurement data to obtain a comprehensive feature vector.

[0027] Based on the pre-set growth stage rules, the comprehensive feature vector is assigned to a specific stage, and the corresponding crop growth stage category label is output.

[0028] A three-dimensional structured mapping table for crops is constructed based on plot number, timestamp, and growth stage label.

[0029] Furthermore, based on ambient light data and combined with a causal structure learning model, a spectral regulation causal network for different growth stages is constructed, including:

[0030] Based on the crop growth stage, corresponding light environment data variables, crop physiological parameters, and final crop quality variables are extracted to form a staged dataset.

[0031] Using a causal structure learning algorithm, the causal relationship between the light environment data variables of crops at each growth stage and the final crop quality variables is modeled, outputting a causal directed acyclic graph for each stage, and constructing a spectral regulation causal network for each growth stage; the causal strength of each path in the spectral regulation causal network for each growth stage is quantified.

[0032] Based on the spectral regulation causal network of different growth stages, the causal directed acyclic graphs of each growth stage are integrated to identify the key causal chains and network regulation nodes between different stages.

[0033] Furthermore, an adaptive interpretive graph neural network is used to extract the contributions of key spectral factors in multi-temporal and spatiotemporal scenarios, ultimately generating optimal spectral regulation suggestions for different stages and plots, including:

[0034] A spatiotemporal multi-node graph model is constructed, with each plot of crop, spectral distribution, time period, and growth stage serving as nodes and attributes of the spatiotemporal multi-node graph model.

[0035] Attention-based graph neural networks are used to train spatiotemporal multi-node graph models, extract the contribution of different growth stages, different plots, and different spectral distributions to crop yield and quality, output the effect explanation, and locate potential optimal regulatory factors.

[0036] By aggregating the causal strength and effect explanations of all outputs, optimal spectral combination recommendation tables and control suggestions are generated for different stages and plots.

[0037] Furthermore, using a causal structure learning algorithm, the causal relationship between the light environment data variables of crops at each growth stage and the final crop quality variables is modeled, outputting a causal directed acyclic graph for each stage, and constructing a spectral regulation causal network for each growth stage, including:

[0038] In each growth stage, an undirected complete graph of variables is constructed using each light environment data variable as a node, and undirected edges connect any nodes.

[0039] For each pair of connected light environment data variables, the independence of each pair of connected light environment data variables is tested by successively increasing the size of the condition set and applying the conditional independence test.

[0040] If two light environment data variables are detected to be independent under a certain condition set, the corresponding light environment data variable is deleted from the undirected complete graph of variables; traverse all combinations of light environment data variables and all condition sets to complete the screening of all edges in the undirected complete graph of variables and obtain the causal skeleton;

[0041] Based on each pair of retained undirected adjacency variables in the causal skeleton, two opposite causal autoregressive flow models are constructed, and the parameters of the causal autoregressive flow models are optimized by maximum likelihood estimation; the marginal log-likelihood in each direction is calculated based on the standardized flow method.

[0042] Compare the marginal log-likelihoods in the two directions, select the direction with the larger marginal likelihood as the true causal direction, and obtain the causal directed acyclic graph;

[0043] The causal directed acyclic graphs of each growth stage are merged, and hierarchical links of variable causal paths between each stage are established by standardizing variable identifiers and time series labels.

[0044] The key nodes that regulate the final crop quality variables are identified, and all causal paths are assigned causal strength scores to form a spectral regulatory causal network for different growth stages.

[0045] Furthermore, the causal strength of each path in the spectral regulation causal network of the quantification growth stage includes:

[0046] Using a graph traversal algorithm, enumerate all directed paths that start from the light environment data variable node, pass through the crop physiological parameter node, and reach the final crop quality variable node.

[0047] Using the causal strength measurement method, the causal strength of each directed edge in the directed path is quantified to obtain the numerical strength of each pair of direct causal relationships.

[0048] The causal strengths of each directed edge are combined according to the order of the directed paths to obtain the overall causal strength of the directed paths.

[0049] The beneficial effects of this invention are as follows:

[0050] (1) This invention acquires multidimensional spectral information from different plots, time periods, and crop growth stages, forming structured big data with crop physiological and quality data, ensuring full-chain coverage and high precision of the data. It enables automatic identification of growth stages, eliminates subjective human intervention and classification errors, and improves the objectivity and intelligence of crop growth status judgment.

[0051] (2) This invention utilizes a causal structure learning algorithm to model the causal relationship between light environment variables and final quality variables at each crop growth stage. This enables the maximum identification of the direct and indirect action chains of different spectral channels, spatiotemporal dimensions, and physiological variables on the final crop yield at different stages. Through techniques such as undirected complete graphs of variables, conditional independence tests, and causal autoregressive flow models, the ontological causal structure and strength of the regulatory network are revealed, achieving automatic hierarchical causal linking at different growth stages. After synthesizing the staged network, key factors and core paths are identified and quantified, enhancing the pertinence and operability of theoretical inferences and practical regulation. In other words, the precision of spectral regulation is improved based on causal structure learning and staged network analysis.

[0052] (3) An adaptive interpretable graph neural network model was introduced. Through deep training and attribution of spatiotemporal multi-node graphs, the specific contribution of different attributes to crop quality in multi-spatial and multi-environment scenarios was measured. Combining the output of the causal network, causal strength and attention weights were integrated for the localization and contribution interpretation of key regulatory factors. Through aggregation analysis, personalized and hierarchical optimal spectral regulation recommendation tables and action guidance suggestions for different crop plots and different growth stages were finally formed. This improved the interpretability of refined management of light environment and improvement of crop quality. Attached Figure Description

[0053] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0054] Figure 1 This is a block diagram of a multi-channel optical environment monitoring and information transmission device according to an embodiment of the present invention;

[0055] Figure 2 This is a circuit schematic diagram of the MCU minimum system according to an embodiment of the present invention;

[0056] Figure 3 This is a circuit schematic diagram of a power supply circuit according to an embodiment of the present invention;

[0057] Figure 4 This is a circuit schematic diagram of a button circuit according to an embodiment of the present invention;

[0058] Figure 5 This is a circuit schematic diagram of a spectral detection circuit according to an embodiment of the present invention;

[0059] Figure 6 This is a circuit diagram of a light intensity detection circuit according to an embodiment of the present invention;

[0060] Figure 7 This is a circuit schematic diagram of an OLED display circuit according to an embodiment of the present invention;

[0061] Figure 8 This is a circuit schematic diagram of a transmission circuit according to an embodiment of the present invention;

[0062] Figure 9 This is a software function relationship diagram of the MCU minimum system according to an embodiment of the present invention;

[0063] Figure 10 This is a hardware principle framework diagram according to an embodiment of the present invention.

[0064] In the picture:

[0065] 1. Power supply module; 2. Microcontroller module; 3. Spectrum detection module; 4. Light intensity detection module; 5. Data upload module; 6. Data processing and analysis module. Detailed Implementation

[0066] To further illustrate the various embodiments, the present invention provides accompanying drawings, which are part of the disclosure of the present invention. These drawings are mainly used to illustrate the embodiments and can be used in conjunction with the relevant descriptions in the specification to explain the operating principles of the embodiments. With reference to these drawings, those skilled in the art should be able to understand other possible implementation methods and the advantages of the present invention. The components in the drawings are not drawn to scale, and similar component symbols are generally used to represent similar components.

[0067] According to an embodiment of the present invention, a multi-channel optical environment monitoring and information transmission device is provided.

[0068] The present invention will now be further described in conjunction with the accompanying drawings and specific embodiments, such as... Figure 1 As shown, the multi-channel optical environment monitoring and information transmission device according to an embodiment of the present invention includes:

[0069] Power module 1 is used to regulate the voltage of the power supply according to the voltage regulator; microcontroller module 2 is used to filter out high-frequency noise on the power supply and store energy, while providing a target impedance return path for high-frequency signals; spectral detection module 3 is used to collect luminous flux values ​​of different wavelength spectra and communicate with the microcontroller module; light intensity detection module 4 is used to collect light intensity of different wavelength spectra and communicate with the microcontroller module; data upload module 5 is used to standardize the light environment data including luminous flux values ​​and light intensity, and upload it to the cloud server to generate multi-dimensional charts; data processing and analysis module 6 is used to obtain the growth stage of crops; based on the light environment data and combined with the causal structure learning model, a spectral regulation causal network for different growth stages is constructed; an adaptive interpretive graph neural network is used to extract the contribution of key spectral factors in multi-temporal and spatial scenarios, and finally generate optimal spectral regulation suggestions for different stages and plots.

[0070] In a further embodiment, the microcontroller module includes: chip U3, inductor L1, capacitor C6 and capacitor C7; wherein, the eighth terminal of chip U3 is connected to one end of inductor L1, one end of capacitor C6 and one end of capacitor C7 in sequence, the other end of inductor L1 is connected to the ninth terminal of chip U3, and the other end of capacitor C6 is connected to the other end of capacitor C7.

[0071] In a further embodiment, the power module includes: a socket DC1, a USB1, a button SW1, diodes D1 and D2, capacitors C3 and C4, a chip U1, capacitors C1 and C2, an LED1, and a resistor R1; wherein, the first terminal of the USB1 is connected to the second terminal of the button SW1, the third terminal of the button SW1 is connected in sequence to the negative terminal of diode D2 and the positive terminal of diode D1, the negative terminal of diode D1 is connected in sequence to one end of capacitor C3, one end of capacitor C4, and the third terminal of chip U1, the positive terminal of diode D2 is connected in sequence to the fifth terminal of USB1, the other end of capacitor C3, the other end of capacitor C4, the first terminal of chip U1, one end of capacitor C1, one end of capacitor C2, and one end of resistor R1, the other end of resistor R1 is connected to the negative terminal of LED1, and the positive terminal of LED1 is connected in sequence to the other end of capacitor C2, the other end of capacitor C1, and the fourth terminal of chip U1; a socket DC1 is provided on the side of the USB1.

[0072] In a further embodiment, the spectral detection module includes: terminal H2, capacitor C8, capacitor C9, resistor R14, resistor R15, resistor R16, and resistor R17; wherein, the second end of terminal H2 is connected to one end of capacitor C8 and one end of capacitor C9 in sequence, the other end of capacitor C8 is connected to the other end of capacitor C9, the third end of terminal H2 is connected to resistor R14, the fourth end of terminal H2 is connected to resistor R15, the fifth end of terminal H2 is connected to resistor R16, and the sixth end of terminal H2 is connected to resistor R17.

[0073] In a further embodiment, the light intensity detection module includes: acquiring ambient light signals through an ambient light sensor combined with an infrared photodiode, digitizing the signals through an analog-to-digital converter (ADC), and then performing infrared compensation using a digital signal processing (DSP) module.

[0074] In a further embodiment, obtaining the growth stage of the crop includes:

[0075] Crop images are organized by plot and time, and matched with light flux and light intensity to obtain a structured dataset including plot number, timestamp, light environment parameters, and crop images. A deep convolutional neural network model is used to extract features from the crop images, automatically identifying and classifying the current growth morphology characteristics of the crops. These features are then fused with crop physiological measurement data (such as leaf area index, stem and leaf fresh weight, and plant height) to obtain a comprehensive feature vector. Based on pre-defined growth stage rules in a crop science knowledge base, the comprehensive feature vector is assigned to a specific stage, and the corresponding crop growth stage category label is output. Finally, a three-dimensional structured mapping table of crops is constructed based on plot number, timestamp, and growth stage label.

[0076] Crop images are processed using a deep convolutional neural network model to extract visual features such as morphology, color, and edges. Simultaneously, crop physiological measurements (e.g., leaf area index, stem and leaf fresh weight, plant height) are incorporated for multi-source fusion. This enhances the model's adaptability to different crop varieties, light environments, and mutational stresses (e.g., pests, diseases, nutrient deficiencies). Furthermore, by combining numerical scaling with semantic information, it improves the ability to distinguish the performance of different individuals or plots at the same stage. Fusion methods include feature-level stitching, attention mechanisms, and multimodal network structures.

[0077] Stage attribution determination relies on rules from a crop science knowledge base (such as tillering stage, heading stage, and milk stage classification standards), combined with comprehensive feature vectors for automatic classification. It employs traditional rule engines, decision tree models, or AI-based end-to-end growth stage classification models. The knowledge base is regularly updated and supports customized expansion based on crop varieties, regions, and management models, ensuring accuracy and broad applicability. A three-dimensional structured mapping table is formed using plot numbers, timestamps, and growth stage tags, enabling multi-dimensional integrated management of crops, time, space, and developmental stages.

[0078] In a further embodiment, based on light environment data and combined with a causal structure learning model, a spectral regulation causal network for different growth stages is constructed, including:

[0079] Based on the crop growth stages, corresponding light environment data variables, crop physiological parameters, and final crop quality variables are extracted to form stage-specific datasets. Using a causal structure learning algorithm, the causal relationship between the crop light environment data variables and the final crop quality variables at each growth stage is modeled, outputting a causal directed acyclic graph for each stage, and constructing a spectral regulation causal network for each growth stage. The causal strength of each path in the spectral regulation causal network for each growth stage is quantified. Based on the spectral regulation causal network for each growth stage, the causal directed acyclic graphs of each growth stage are integrated to identify key causal chains and network regulation nodes between different stages.

[0080] Among them, by performing data cleaning, missing value imputation, standardization and normalization on light environment data variables, crop physiological parameters and final crop quality variables, the accuracy and robustness of causal modeling are ensured.

[0081] By analyzing the integrated, staged spectral regulation causal network, key intermediary nodes with significant positive or negative effects on the final quality variables can be precisely located, allowing for the screening of controllable and interventionist pathways and variables. This not only serves the decision optimization for precise light environment regulation and crop quality improvement but also provides a basis for causal reasoning, achieving an engineering closed loop from observation to intervention.

[0082] In a further embodiment, an adaptive interpretive graph neural network is used to extract the contributions of key spectral factors in multi-temporal and spatiotemporal scenarios, ultimately generating optimal spectral modulation suggestions for different stages and plots, including:

[0083] A spatiotemporal multi-node graph model is constructed, with each crop plot, spectral distribution, time point, and growth stage serving as nodes and attributes. A graph neural network based on an attention mechanism is used to train the spatiotemporal multi-node graph model structure, extracting the contributions of different growth stages, plots, and spectral distributions to crop yield and quality, outputting explanations of effects, and locating potential optimal regulatory factors. The causal strength and explanations of all outputs are aggregated to form optimal spectral combination recommendation tables and regulatory suggestions for different stages and plots, and the optimal spectral regulation suggestions are integrated and output.

[0084] Node types include: plot nodes (reflecting geographical distribution and soil background), spectral nodes (different spectral channels or combinations), time nodes (continuous or discrete time tags), growth stage nodes (such as seedling stage, tillering stage, jointing stage, etc.), and crop physiological or quality status nodes (such as yield, quality indicators). Each node is accompanied by multidimensional features (such as light environment data, crop physiological parameters, etc.).

[0085] During the model inference phase, the attention score or attribution weight of the node is directly output, and spectral factors, special plots, key moments and growth nodes with high sensitivity to the results are used as candidates for regulation priorities.

[0086] The results of causal network inference at different stages, attribution weights of graph neural network nodes / paths, and other multi-source information are aggregated at multiple scales. Aggregation methods include weighted averaging, input mechanisms, and confidence interval assessments to ensure that recommended spectral combinations have theoretical causal basis and reflect the true contribution of data-driven attribution. Optimal spectral channel parameter combinations, control schemes, and their relative contribution rates for different plots and growth stages are provided for practical reference and benefit comparison by planting managers. Based on attribution analysis and combined with controllable spectral equipment parameters (such as the wavelength, intensity, and irradiation cycle of LED lights), the data is automatically converted into control commands or setting suggestions for physical equipment, directly intervening in the agricultural production system in a closed-loop intelligent control process.

[0087] In a further embodiment, a causal structure learning algorithm is used to model the causal relationship between crop light environment data variables and final crop quality variables at each growth stage, outputting a causal directed acyclic graph for each stage, and constructing a spectral regulation causal network for each growth stage, including:

[0088] In each growth stage, an undirected complete graph of variables is constructed using each light environment data variable as a node, with undirected edges connecting any node. For each pair of connected light environment data variables, the independence of each pair of connected light environment data variables is tested by successively increasing the size of the condition set (i.e., using other variables as condition variables step by step) and applying the conditional independence test. If two light environment data variables are found to be independent under a certain condition set, the corresponding light environment data variable is deleted from the undirected complete graph. All combinations of light environment data variables and all condition sets are traversed to complete the screening of all edges in the undirected complete graph, resulting in a causal skeleton. Based on each pair of retained undirected adjacent variables in the causal skeleton, two opposite causal relationships are constructed. A causal autoregressive flow model was developed, and the parameters of the causal autoregressive flow model were optimized using maximum likelihood estimation. The marginal log-likelihood of each direction was calculated based on the standardized flow method. The marginal log-likelihoods of the two directions were compared, and the direction with the larger marginal likelihood was selected as the true causal direction, resulting in a causal directed acyclic graph. The causal directed acyclic graphs of each growth stage were merged, and hierarchical links of the causal paths of variables between each stage were established through standardized variable labels and time series labels. The key nodes that have a regulatory effect on the final crop quality variables were located, and all causal paths were assigned causal strength scores to form a spectral regulatory causal network for different growth stages, namely, a hierarchical causal regulatory network of spectral variables-crop physiological processes-target quality.

[0089] In a further embodiment, the causal strength of each path in the spectral modulation causal network of the quantization growth stage includes:

[0090] Using a graph traversal algorithm, all directed paths starting from the light environment data variable node, passing through the crop physiological parameter node, and reaching the final crop quality variable node are enumerated. Using causal strength measurement methods (such as regression coefficient, log-likelihood change, average treatment effect, etc.), the causal strength of each directed edge in the directed path is quantified to obtain the numerical strength of each pair of direct causal relationships. According to the order of the directed paths, the causal strengths of each directed edge are synthesized. For example, the synthesis strategy is to multiply the causal strengths (when each edge is conditionally independent), or a Bayesian network path weight accumulation algorithm is used to obtain the overall causal strength of the directed paths.

[0091] Among them, the conditional independence test is the core of causal skeleton learning. The appropriate test method can be selected according to the distribution characteristics of the variable. For example, for Gaussian distribution, Fisher's Z test based on partial correlation coefficient is used; for categorical or mixed variables, the chi-square test based on empirical distribution G test is used; for nonlinear or complex relationships, it is recommended to use kernel density conditional independence test.

[0092] To determine the causal direction of variable pairs, a practical causal autoregressive flow model transparently learns the generation flow direction between variables. During maximum likelihood estimation, two flow models in opposite directions are trained, and the marginal log-likelihoods under standardized flow are calculated separately to avoid bias caused by differences in model complexity.

[0093] Through quantitative analysis, not only can the theoretically optimal control point for each stage and each pair of final qualities be found, but redundant variables can also be screened for paths with low control efficiency, short causal chains, or weak strength, thus slimming down the causal network. The strength scores of all causal chains and nodes can be used for real-time alarms in the intelligent control system. That is, once the state of a key point deviates from the optimal strength threshold, operations such as spectral adjustment and management optimization can be triggered.

[0094] In a further embodiment, the display circuit includes: chip U2, resistor R2, resistor R3 and capacitor C5; wherein, the second end of chip U2 is connected to capacitor C5, the third end of chip U2 is connected to one end of resistor R3, the fourth end of chip U2 is connected to one end of resistor R2, and the other end of resistor R2 is connected to the other end of resistor R3.

[0095] In a further embodiment, the transmission circuit includes: chip U11, resistor R5, capacitor C7, connector J11, and connector J9; wherein, the second and third ends of chip U11 are connected to connector J11, the sixth end of chip U11 is sequentially connected to one end of resistor R5 and the second end of connector J9, the first end of connector J9 is sequentially connected to the other end of resistor R5 and the seventh end of chip U11, the eighth end of chip U11 is connected to one end of capacitor C7, and the other end of capacitor C7 is connected to the fifth end of chip U11.

[0096] To facilitate understanding of the above technical solutions of the present invention, the working principle of the present invention in actual process will be described in detail below.

[0097] This invention collects and transmits light environment data at different locations. Light intensity measurement range: 0–3000 μmol / m⁻²·s⁻¹ (PPFD), accuracy 5%; spectral response wavelength range: 400–700 nm.

[0098] Hardware principle framework diagram of multi-channel optical environment monitoring and information transmission equipment, such as... Figure 1 As shown. Specifically includes:

[0099] Microcontroller (MCU) minimum system (data processing system)

[0100] The MCU minimum system uses the STC8H1K17, which has a built-in high-precision clock unit and hardware reset circuit, thus eliminating the need for external clock and reset circuits. A 2.2uF and a 100nF bypass capacitor are added to the microcontroller's power input section. Their main functions include:

[0101] 1. Filter out high-frequency noise from the power supply. 2. Store energy; when the load requires instantaneous current, the capacitor provides the current first, reducing power supply fluctuations. 3. Provide the shortest low-impedance return path for high-frequency signals, reducing interference to the power supplies of other chips. Circuit diagram description, such as... Figure 2 As shown.

[0102] Power supply circuit

[0103] The power supply circuit uses an LDO M1117-3.3 (forward low dropout regulator) to convert +5V DC to +3.3V DC. This LDO features low ripple noise, ensuring system stability and reliability. Circuit diagram description, as follows: Figure 3 As shown. DC1 - DC power adapter +5V input. USB1 - USB +5V power supply interface. SW1 - Power switch. D1 - Schottky diode, providing protection against reverse polarity. D2 - Transient voltage suppressor diode, utilizing its non-linear characteristics to clamp overvoltage to a lower value, protecting downstream circuits. U - LDO, converts +5V to +3.3V. C1 / C2 / C3 / C4 - Primarily used to filter out noise and interference from the DC output voltage, smoothing it into a stable DC output voltage. R1 - Current-limiting resistor. LED1 - Power indicator.

[0104] Button circuit

[0105] The button circuit uses touch buttons. When a user presses a button, the circuit detects the contact and generates a low-level signal, triggering the corresponding operation. Users can use these buttons to modify the timing and turn on the lights. Circuit diagram description, as follows: Figure 4 As shown. SW2 / SW3 / SW4 / SW5 - Point-touch buttons. The working principle of point-touch buttons: The button and the contact point act mechanically. When the button is pressed, the spring contracts, the contact point contacts the conductive strip, and the circuit is connected; when the button is released, the spring returns to its original state, the contact point leaves the conductive strip, and the circuit is disconnected.

[0106] Spectral detection circuit

[0107] The spectral detection circuit uses the AS7341 module. The AS7341 is a sensor based on the AS7341 visible spectrum sensing IC. It can detect the visible light components in different wavelengths of the environment and measure luminous flux. It exhibits considerable sensitivity and accuracy, and its very small size makes it suitable for use as a miniature spectrum analyzer. The AS7341 communicates with the microcontroller via the IIC bus. The circuit diagram is shown below. Figure 5 As shown.

[0108] Light intensity detection circuit

[0109] The MAX44009 (ambient light sensor) incorporates a built-in infrared photodiode for compensation. Signals from both sensors are digitized by an ADC module and then processed by a DSP (Digital Signal Processor) module for precise infrared compensation. Furthermore, the IC chip features a digital interface output and a compact size. A circuit diagram is shown below. Figure 6 As shown. A VISIBLE+IRPHOTODIODE is a photodetector that converts visible light and infrared light signals into electrical signals. A 16-BITADC is a 16-bit ADC (analog-to-digital converter). A 6-BIT RANGE CDR is a parameter range for clock data recovery configured using 6 bits of binary data. TIM CONTROL is timing control.

[0110] OLED display circuit

[0111] An OLED display is used to show spectral information. Furthermore, the OLED is also used for human-computer interaction during the timer setting process. Circuit diagram description, such as... Figure 7 As shown.

[0112] Transmission circuit

[0113] The 485-RX and 485-TX can be directly connected to the microcontroller's serial port. The 485-DIR can be connected to a general-purpose output I / O port to control the communication direction: it is always low by default (in receive mode), pulled high to start transmitting, and pulled low again after transmission is complete. Circuit diagram description, such as... Figure 8 As shown.

[0114] MCU Minimum System Software Design

[0115] Overall software function relationship diagram, such as Figure 9 As shown, the internal connections and logical structure between the main function and each sub-function are presented in detail and comprehensively.

[0116] Data processing design

[0117] Install electronic component development software and calculate color temperature using the correction matrix. The color temperature is calculated using correction data from mass-produced sensors and the CIE1931-based XYZ correction matrix, as shown in Table 1. The X, Y, and Z values ​​in the CIE1931 XYZ primary color system can be calculated using the following formulas:

[0118]

[0119] In the formula, X CIE1931 Y CIE1931 Z CIE1931 These represent the X, Y, and Z components in the XYZ color space calculated according to the CIE 1931 standard. F1 BasicCount To NIR BasicCount These represent spectral measurements at different wavelengths. The relative coefficients x, y, and z in the CIE 1931 xyz color space are calculated using the following formula:

[0120]

[0121] Color temperature calculation formula:

[0122]

[0123] Table 1 XYZ Correction Matrix

[0124] X Y Z F1 410 0.39814 0.01396 1.9501 F2 440 1.2954 0.16748 6.4549 F3 470 0.36954 0.23538 2.7801 F4 510 0.10902 1.4275 0.18501 F5 550 0.71942 1.8867 0.15325 F6 583 1.7818 1.142 0.09539 F7 620 1.1011 0.46497 0.10563 F8 670 -0.03991 -0.02702 0.08866 Clear 750 -0.27597 -0.24468 -0.6114 NIR 900 -0.02347 -0.01993 -0.00938

[0125] Among them, F1-F8 are different spectral bands, Clear is a channel without a specific filter, and NIR is the near-infrared spectral channel.

[0126] The luminous flux and illuminance values ​​of different wavelengths in the spectrum are collected, and then the values ​​are uploaded to a cloud server or PC via 4G in data form. The data in the cloud is then used by the cloud-based light analysis software system to generate charts.

[0127] Data processing and analysis

[0128] By integrating crop images, plot information, time series data, and light environment data, the system first automatically identifies and labels the growth stages of crops in each plot. Then, at each growth stage, causal structure learning methods are used to analyze the causal relationships between spectral data and crop physiological and quality indicators, constructing a stage-specific spectral regulation causal network. Finally, combined with an adaptive interpretive graph neural network, key spectral factors and their contributions to yield and quality are extracted from multi-plot, multi-temporal light environment data. Based on this, scientific and personalized optimal spectral regulation recommendations are generated for crops in different plots and at different growth stages.

[0129] Prototyping, Testing and Evaluation

[0130] Prototyping: Fabricating the PCB board and assembling various components. Functional Testing: Testing all functions of the product to ensure its proper operation. Performance Evaluation: Evaluating the product's stability and accuracy to ensure it meets design requirements.

[0131] In summary, this invention acquires multidimensional spectral information from different plots, time periods, and crop growth stages, forming structured big data with crop physiological and quality data, ensuring full-chain data coverage and high accuracy. It achieves automatic identification of growth stages, eliminating subjective human intervention and classification errors, and improving the objectivity and intelligence of crop growth status assessment. By utilizing causal structure learning algorithms, this invention models the causal relationship between light environment variables and final quality variables at each crop growth stage, maximizing the identification of direct and indirect effects of different spectral channels, spatiotemporal dimensions, and physiological variables on crop final output at different stages. Through techniques such as undirected complete graphs of variables, conditional independence tests, and causal autoregressive flow models, it reveals the ontological causal structure and strength of the regulatory network, achieving automatic growth stage hierarchical classification of causal links. After synthesizing the staged network, key factors and core paths are identified and quantified, enhancing the pertinence and operability of theoretical inferences and practical regulation. In short, causal structure learning and staged network analysis improve the precision of spectral regulation. An adaptive interpretable graph neural network model was introduced, and through deep training and attribution of spatiotemporal multi-node graphs, the specific contribution of different attributes to crop quality was measured in multi-spatial and multi-environmental scenarios. Combining the output of the causal network, causal strength and attention weights were integrated for the localization and contribution interpretation of key regulatory factors. Through aggregation analysis, personalized and hierarchical optimal spectral regulation recommendations and action guidelines were ultimately generated for different crop plots and different growth stages. This improved the interpretability of refined light environment management and crop quality improvement.

[0132] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-channel light environment monitoring and information transmission device, characterized in that, The application relates to a light spectrum regulation system for crops, which comprises the following modules: a power module for voltage regulation of a power supply according to a voltage stabilizer; a single-chip microcomputer module for filtering high-frequency noise on the power supply, storing energy, and providing a target impedance return path for high-frequency signals; a light spectrum detection module for collecting light flux values of different wavelength spectrums and communicating with the single-chip microcomputer module; a light intensity detection module for collecting light intensity of different wavelength spectrums and communicating with the single-chip microcomputer module; a data uploading module for standardizing light environment data including the light flux values and the light intensity and uploading the data to a cloud server to generate multidimensional charts; and a data arrangement and analysis module for obtaining a growth stage of crops, constructing a light spectrum regulation causal network of different growth stages based on light environment data and a causal structure learning model, extracting key light spectrum factor contributions in multiple spatiotemporal scenes by using an adaptive explanatory graph neural network, and finally generating optimal light spectrum regulation suggestions for different stages and different plots. The single-chip microcomputer module comprises a chip U3, an inductor L1, a capacitor C6 and a capacitor C7. The eighth end of the chip U3 is connected with one end of the inductor L1, one end of the capacitor C6 and one end of the capacitor C7 in sequence, the other end of the inductor L1 is connected with the ninth end of the chip U3, and the other end of the capacitor C6 is connected with the other end of the capacitor C7. The power module comprises a socket DC1, a USB1, a button SW1, a diode D1, a diode D2, a capacitor C3, a capacitor C4, a chip U1, a capacitor C1, a capacitor C2, an LED1 and a resistor R1. The first end of the USB1 is connected with the second end of the button SW1, the third end of the button SW1 is connected with the negative electrode of the diode D2 and the positive electrode of the diode D1 in sequence, the negative electrode of the diode D1 is connected with one end of the capacitor C3, one end of the capacitor C4 and the third end of the chip U1 in sequence, the positive electrode of the diode D2 is connected with the fifth end of the USB1, the other end of the capacitor C3, the other end of the capacitor C4, the first end of the chip U1, one end of the capacitor C1, one end of the capacitor C2 and one end of the resistor R1 in sequence, the other end of the resistor R1 is connected with the negative electrode of the LED1, the positive electrode of the LED1 is connected with the other end of the capacitor C2, the other end of the capacitor C1 and the fourth end of the chip U1 in sequence, and the side of the USB1 is provided with the socket DC1. The light spectrum detection module comprises a terminal H2, a capacitor C8, a capacitor C9, a resistor R14, a resistor R15, a resistor R16 and a resistor R17. The second end of the terminal H2 is connected with one end of the capacitor C8 and one end of the capacitor C9 in sequence, the other end of the capacitor C8 is connected with the other end of the capacitor C9, the third end of the terminal H2 is connected with the resistor R14, the fourth end of the terminal H2 is connected with the resistor R15, the fifth end of the terminal H2 is connected with the resistor R16, and the sixth end of the terminal H2 is connected with the resistor R17. ​ 2. The multi-pass light environment monitoring and information transmission device according to claim 1, characterized in that, ​ ​ 3. The multi-pass light environment monitoring and information transmission device according to claim 1, characterized in that, ​ ​ 4. The multi-pass light environment monitoring and information transmission device according to claim 1, characterized in that, ​ ​ 5. The multi-pass light environment monitoring and information transmission device according to claim 1, characterized in that, The light intensity detection module comprises: Through the ambient light sensor and combined with the infrared light sensing diode, the ambient light signal is acquired, and after digitalization through the analog-to-digital converter module, the infrared compensation is made by using the digital signal processing module.

6. The multi-pass light environment monitoring and information transmission device of claim 1, wherein, The growth stage of the crop is acquired by: The crop pictures are sorted according to the plots and time, and are matched with the light flux values and the light intensity to obtain a structured data set comprising plot numbers, time stamps, light environment parameters and crop images; The deep convolutional neural network model is used to extract features of the crop images, to automatically identify and classify the current growth morphological features of the crop, and to fuse the features with the crop physiological measurement data to obtain a comprehensive feature vector; According to the pre-set growth stage rules, the comprehensive feature vector is attributed to a stage, and a corresponding crop growth stage category label is outputted; According to the plot numbers, the time stamps and the growth stage labels, a three-dimensional structured mapping table of the crop is constructed.

7. The multi-pass light environment monitoring and information transmission device of claim 1, wherein, The light spectrum regulation causal network of each growth stage is constructed based on the light environment data and combined with the causal structure learning model, comprising: According to the growth stage of the crop, the corresponding light environment data variables, crop physiological parameters and final crop quality variables are extracted to form a data set for each stage; The causal relationship between the light environment data variables and the final crop quality variables of the crop in each growth stage is modeled by using the causal structure learning algorithm, a causal directed acyclic graph for each stage is outputted, and a light spectrum regulation causal network for each growth stage is constructed; the causal strength of each path in the light spectrum regulation causal network for each growth stage is quantified; Based on the light spectrum regulation causal network for each growth stage, the causal directed acyclic graphs for each growth stage are integrated, and the key causal action chains and network regulation nodes between different stages are identified.

8. The multi-pass light environment monitoring and information transmission device according to claim 7, characterized in that, The adaptive explainable graph neural network is used to extract key light spectrum factor contributions in multiple spatio-temporal scenarios, and finally generate optimal light spectrum regulation recommendations for each stage and each plot, comprising: A spatio-temporal multi-node graph model is constructed, taking each plot of the crop, spectrum distribution, each time and each growth stage as nodes and attributes of the spatio-temporal multi-node graph model; The spatio-temporal multi-node graph model structure is trained based on the attention mechanism of the graph neural network, the contribution of different growth stages, different plots and different spectrum distributions to the crop yield and quality is extracted, the action explanation is outputted, and the potential optimal regulation factor is located; The causal strength and action explanation of all outputs are aggregated to form an optimal light spectrum combination recommendation table and regulation suggestion for each stage and each plot.

9. The multi-pass light environment monitoring and information transmission device according to claim 7, characterized in that, The causal relationship between the light environment data variables and the final crop quality variables of the crop in each growth stage is modeled by using the causal structure learning algorithm, a causal directed acyclic graph for each stage is outputted, and a light spectrum regulation causal network for each growth stage is constructed, comprising: In each growth stage, each light environment data variable is taken as a node to construct a variable undirected complete graph, and undirected edges are connected between any nodes; For each pair of connected light environment data variables, the independence of each pair of connected light environment data variables is detected by sequentially increasing the size of the conditional set and applying the conditional independence test method. If two light environment data variables are detected to be independent under a certain condition set, the corresponding light environment data variable is deleted from the variable undirected complete graph; all combinations of light environment data variables and all condition sets are traversed to complete the screening of all edges in the variable undirected complete graph, and a causal skeleton is obtained; According to each pair of retained undirected adjacent variables in the causal skeleton, two opposite causal autoregressive flow models are respectively constructed, and the parameters of the causal autoregressive flow models are optimized by maximum likelihood estimation; the marginal log-likelihood is calculated in each direction based on the standardized flow method; The marginal log-likelihoods in two directions are compared, and the direction with greater marginal log-likelihood is selected as the real causal direction, and a causal directed acyclic graph is obtained; The causal directed acyclic graphs of each growth stage are combined, the standardized variable identifiers and time sequence labels are used to establish layer links of variable causal paths between stages; The key nodes having regulating effects on the final crop quality variables are located, and causal strength scores are assigned to all causal paths to form a growth-stage-specific spectral regulation causal network.

10. The multi-pass light environment monitoring and information transmission device of claim 7, wherein, The quantification of the causal strength of each path in the growth-stage-specific spectral regulation causal network includes: Using a graph traversal algorithm, all directed paths from the light environment data variable node, through the crop physiological parameter node, to the final crop quality variable node are enumerated; Using a causal strength measurement method, the causal strength of each directed edge in the directed path is quantified to obtain the numerical strength of each pair of direct causal relationship; According to the order of the directed path, the causal strengths of each directed edge are synthesized to obtain the overall causal strength of the directed path.

Images