A method and system for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics

Abstract

This invention relates to a method and system for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics. The method acquires sample data by setting different light gradient cultivation environments, preprocesses the data using standard normal variable transformation, extracts characteristic wavelengths using a competitive adaptive reweighting algorithm and a continuous projection method, and establishes a prediction model using a regression-based support vector machine algorithm to predict the photosynthetic potential of crops. The prediction system includes main control processing, a driving light source, reflected light detection, power supply, communication, control software, and display units. The main control processing unit controls the operation of each component and processes data; the driving light source unit provides detection light; the reflected light detection unit detects light intensity and converts and processes the signal; the communication unit uploads data; and the control software and display unit calculates and displays the results. This invention can quickly, accurately, and cost-effectively assess the photosynthetic potential of greenhouse crops, providing a scientific basis for agricultural production and assisting in the monitoring of crop growth status.

Description

Technical Field

[0001] This invention relates to the field of intelligent agriculture technology, specifically to a method and system for predicting the photosynthetic potential of facility crops based on spectral characteristics. Background Technology

[0002] Chlorophyll is the basic pigment of photosynthesis. When chlorophyll molecules absorb light energy, they transition from the ground state to an excited state and then back to the ground state. During this process, some energy is re-emitted in the form of fluorescence, forming chlorophyll fluorescence. By measuring this fluorescence, we can indirectly understand the relevant processes of photosynthesis. Among the commonly used parameters of chlorophyll fluorescence, the maximum photochemical quantum yield of PSII, Fv / Fm, represents the plant's photosynthetic potential, reflecting the maximum light energy conversion efficiency of the PSII reaction center, and is a key indicator in plant photosynthesis research. Real-time and non-destructive detection of plant photosynthetic potential is crucial for characterizing plant physiological states and their responses to the external environment. However, traditional Fv / Fm measurements require expensive chlorophyll fluorometers and necessitate long-term dark adaptation of plants, limiting their application in practical agricultural production.

[0003] Spectroscopic detection technology is an analytical technique based on the principle of matter-light interaction, characterized by its sensitivity, non-destructive nature, speed, and efficiency. Addressing the limitation of traditional Fv / Fm measurement methods in real-time measurement, numerous researchers have investigated the prediction of Fv / Fm using spectroscopic techniques. Results show a linear correlation between leaf reflectance spectra and chlorophyll fluorescence parameters, thus demonstrating the feasibility of using spectroscopic detection of fluorescence parameters.

[0004] However, most current research employs vegetation index-based modeling methods, which have limited spectral data dimensions and often rely on simple fitting techniques, thus limiting the accuracy of Fv / Fm evaluation. Furthermore, traditional spectral detection systems consist of multiple modules, requiring high stability of the measurement environment and a PC for data acquisition. This configuration not only makes the equipment inconvenient to carry but also makes the entire measurement process relatively complex, failing to meet the rapid detection needs in field or facility environments. Therefore, developing an efficient, accurate, and convenient method for predicting the photosynthetic potential of facility crops based on spectral characteristics to meet the needs of rapid and accurate detection of photosynthetic potential in actual agricultural production is an urgent problem to be solved. Summary of the Invention

[0005] To address the technical problems existing in the prior art, the first objective of this invention is to provide a method for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics. By integrating CARS-SPA characteristic wavelength screening, SNV data preprocessing, and GA-SVR modeling, this method enables non-destructive, rapid, and low-cost detection of the photosynthetic potential of greenhouse crops.

[0006] The second objective of this invention is to provide a system for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics, enabling rapid, accurate, and low-cost assessment of the photosynthetic potential of greenhouse crops, and providing a scientific basis for agricultural production.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A method for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics includes the following steps:

[0009] S1. Obtaining sample data: Set up multiple cultivation environments with the same temperature, humidity, and carbon dioxide concentration under different light gradients. After the crops under different light treatments show differences, randomly select crop leaves as experimental samples and measure the photosynthetic potential Fv / Fm and visible-near infrared reflectance spectra of the crop leaves as sample data.

[0010] S2. Preprocessing of sample data: Standard normal variable transformation is used for data preprocessing to eliminate the influence of sample surface particle size, surface scattering and optical path variation on the reflectance spectrum;

[0011] S3. Extracting Feature Wavelengths: First, a competitive adaptive reweighting algorithm is used to extract feature wavelength combinations that are beneficial to the accuracy of the prediction model. Then, the continuous projection method is used to filter out redundant variables with high repetition. After that, the training set and the test set are randomly divided in a 4:1 ratio to reduce model complexity and improve prediction accuracy.

[0012] S4. Using the reflected light intensity of the spectral characteristic wavelengths of the training set samples as input and the Fv / Fm of the crop as output, a crop photosynthetic potential prediction model is established using a regression-type support vector machine algorithm.

[0013] S5. Use photosynthetic potential prediction models to predict crop photosynthetic potential.

[0014] Furthermore, the steps of the standard normal variable transformation are as follows:

[0015] Calculate the mean of each variable using the following formula:

[0016]

[0017] in, Let x represent the mean of the j-th wavelength point, n represent the number of samples, and x represent the mean of the wavelength points. ij This represents the spectral value of the i-th sample at the j-th wavelength.

[0018] The standard deviation of each variable is calculated using the following formula:

[0019]

[0020] Among them, s j This represents the standard deviation of the j-th wavelength point;

[0021] The SNV transformation is calculated using the following formula:

[0022]

[0023] Among them, z ij This represents the spectral value of the i-th sample after SNV transformation at the j-th wavelength point.

[0024] Furthermore, the steps of the competitive adaptive reweighting algorithm are as follows:

[0025] The Monte Carlo sampling method is used. In each sampling iteration, 80% of the samples are randomly selected from the dataset to enter the modeling set, and the remaining 20% ​​is used as the prediction set to build the PLS model. The number of sampling iterations N for the Monte Carlo method is predetermined, and the absolute value weight of the regression coefficients in the PLS model is recorded for each sampling process. The calculation formula is as follows:

[0026]

[0027] Among them, b i Let be the regression coefficient of the i-th variable;

[0028] By forcibly removing wavelengths with relatively small absolute value weights in the regression coefficients using the exponential decay function (EDF), the proportion R of retained wavelength points is obtained from the EDF when establishing the PLS model based on Monte Carlo sampling for the i-th time. i The calculation formula is as follows:

[0029]

[0030] The formulas for calculating μ and k are as follows:

[0031]

[0032]

[0033] Where N is the number of samplings, and n is the number of original wavelength points;

[0034] The number of samples selected each time is R. i For *n wavelength variables, perform PLS modeling and calculate the root mean square error of cross-validation (RMSECV).

[0035] After N sampling cycles, the CARS algorithm obtains N candidate feature wavelength subsets and their corresponding RMSECV values, and selects the wavelength variable subset corresponding to the minimum RMSECV value as the feature wavelength.

[0036] Furthermore, the steps of the continuous projection method are as follows:

[0037] Choose an initial wavelength by randomly selecting one from the spectral matrix as a characteristic variable;

[0038] To perform projection calculations, for each remaining variable, calculate its projection onto the set of selected feature variables. The formula for this calculation is as follows:

[0039]

[0040] Where, x j The variable currently under consideration, x k(n-1) P is the feature variable selected in the previous iteration. xj It is x j In x k(n-1) Projection on;

[0041] Based on the projection results, the variable that maximizes the magnitude of the projection vector is selected as the next feature variable:

[0042]

[0043] Where s is the set of indices of the features that were not selected into the set of features, and k(n) is the index of the features selected in this iteration;

[0044] Remove the selected feature variables from the original dataset and update the set of unselected feature variables;

[0045] Repeat the above steps until the preset number of features is reached, and output the final set of selected feature variables.

[0046] Furthermore, the photosynthetic potential prediction model includes the following steps:

[0047] Read in the modeling data and perform data normalization processing;

[0048] Select the modeling parameters, including the regularization parameter c, the kernel function, and the kernel function parameter g;

[0049] A regression-type support vector machine model is established based on the training set samples and optimal parameters to obtain a photosynthetic potential prediction model.

[0050] Furthermore, the data normalization process employs a linear normalization method with an interval of [0,1], and its formula is:

[0051]

[0052] Where, x * Here, x represents the normalized data, and x represents the data to be normalized. max and x minThese are the maximum and minimum values ​​in the data sequence, respectively.

[0053] The kernel function used is the radial basis function. The optimal values ​​of parameters c and g are found using a genetic algorithm. The index range of c is set to [0.01, 10], and the number of genes is 6.

[0054] Set the index range of g to [0.01, 5], the number of genes to 5, and the encoding method to binary encoding. The encoding formula is as follows:

[0055]

[0056] The decoding formula is:

[0057]

[0058] Where b is the encoded binary string, m is the number of characters in the binary string taken from the chromosome, and a is the required decimal encoding number. max a is the largest decimal number in the encoding space. min It is the smallest decimal number in the encoding space.

[0059] A system for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics, comprising:

[0060] The main control processing unit has a microcontroller, which is used to control the switching of the driving light source unit, collect the reflected light signal of the reflected light detection unit, and transmit the processed data to the communication unit.

[0061] A driving light source unit includes a constant current driving circuit and a light source control circuit, wherein the constant current driving circuit is used to provide a stable constant current to the light source, and the light source control circuit is used to control the light source to emit detection light;

[0062] The reflected light detection unit is used to detect the intensity of diffuse reflected light from the leaves and to perform signal conversion and processing. It includes a photoelectric conversion circuit, a signal amplification circuit, and an analog-to-digital conversion circuit. The photoelectric conversion circuit is used to convert the optical signal into an electrical signal, the signal amplification circuit is used to amplify the electrical signal, and the analog-to-digital conversion circuit is used to convert the analog signal into a digital signal that can be processed by the microcontroller.

[0063] The power supply unit is used to provide operating voltage to each unit;

[0064] The communication unit includes a wireless communication module and peripheral circuitry. The wireless communication module supports Wi-Fi or Bluetooth protocols and is used to upload the collected data to a server database.

[0065] The control software and display unit are used to read data from the server database, call the crop photosynthetic potential prediction model to calculate the Fv / Fm value, store the detection results and send them to the mobile terminal for display.

[0066] Furthermore, the microcontroller is an STM32F103 microcontroller, which communicates with the constant current drive circuit and the analog-to-digital conversion circuit through standard I / O interfaces, and is connected to the wireless communication module through a USART module.

[0067] Furthermore, the driving light source unit and the reflected light detection unit are disposed inside the reflected light detection probe. The reflected light detection probe includes a housing with an open top. Multiple LEDs are arranged circumferentially on the lower inner sidewall of the housing. The light from the multiple LEDs converges at a certain angle toward the center of the top of the housing, thereby illuminating the blade to be tested. A photoelectric sensor is disposed at the bottom of the housing, which is used to receive the reflected light formed after the blade to be tested is illuminated by the LEDs.

[0068] Furthermore, the bottom of the housing is also provided with a light-shielding platform for blocking stray light emitted by the LED from interfering with the photoelectric sensor.

[0069] The present invention has the following advantages:

[0070] 1. The prediction method of this invention abandons the traditional method of relying on chlorophyll fluorometers to measure Fv / Fm. Traditional methods not only require long-term dark adaptation of plants, but also involve expensive equipment, limiting their application in actual agricultural production. This invention constructs a prediction model based on crop spectral characteristics, directly predicting photosynthetic potential by analyzing the intensity of reflected light at specific wavelengths of the plant's reflectance spectrum. This avoids the cumbersome dark adaptation process, shortens detection time, reduces equipment costs, improves detection efficiency, and makes detection work more economical. In data processing and model construction, on the one hand, standard normal variable transformation is used to preprocess sample data, effectively eliminating the interference of factors such as sample surface particle size, surface scattering, and optical path changes on the reflectance spectrum, significantly improving data quality and providing strong support for the accuracy of subsequent modeling and analysis. On the other hand, by combining the Competitive Adaptive Reweighting Algorithm (CARS) and the Continuous Projection Method (SPA) to extract feature wavelengths, the number of feature wavelengths is minimized while ensuring model accuracy, simplifying the model structure and further improving the model's prediction speed. Furthermore, when using the regression-type support vector machine (SVR) algorithm to establish a crop photosynthetic potential prediction model, the data is first normalized to unify the data magnitude and avoid the problem of imbalanced samples. Then, the regularization parameter c, kernel function and kernel function parameter g of the SVR model are optimized by using a genetic algorithm (GA) to achieve optimal model performance and significantly improve the accuracy and reliability of Fv / Fm photosynthetic potential prediction.

[0071] 2. The prediction system of this invention integrates multiple functional modules. The main control processing unit, based on an STM32F103 microcontroller, controls each unit and processes and transmits data. The communication unit uploads the collected data to a cloud server database. The control software and display unit read data from the database, call the prediction model to calculate the Fv / Fm value, store the results, and send them to a mobile device for display, achieving rapid, non-destructive, low-cost intelligent detection of crop Fv / Fm. The light source control circuit of the driving light source unit uses a 6-channel ring-shaped narrow-bandwidth LED array. These LEDs are arranged at a specific angle within the reflected light detection probe. When they are working, they can achieve uniform illumination of crop leaves, providing a stable and uniform light source for detection. The reflected light detection unit is also located within the reflected light detection probe. It uses photoelectric conversion, signal amplification, and analog-to-digital conversion circuits to accurately and effectively detect the intensity of diffuse reflected light from the leaves. To ensure the accuracy and high quality of the detection results, narrow-bandwidth LED light sources and photodiodes are selected, effectively improving the detection precision. On the other hand, a light-shielding platform is installed inside the reflected light detection probe to block stray light generated by the LED lamp itself and the surrounding environment, preventing it from interfering with the detection signal. Simultaneously, a first-order low-pass filter circuit is used to eliminate high-frequency noise, further purifying the signal. This makes the detection structure simpler, significantly reduces costs, and facilitates the portability and widespread adoption of the equipment. It effectively meets the needs for rapid detection of crop photosynthetic potential in field or facility environments, providing strong technical support for agricultural production. Attached Figure Description

[0072] Figure 1 This is a flowchart of the method for predicting the photosynthetic potential of facility crops based on spectral characteristics, as proposed in this invention.

[0073] Figure 2 This is a flowchart of the photosynthetic potential prediction model established by the GA-SVR algorithm in the spectral characteristics-based photosynthetic potential prediction method for facility crops of the present invention.

[0074] Figure 3 This is a comparison chart of the predicted and actual values ​​of the Fv / Fm prediction model of this invention.

[0075] Figure 4 This is a control flowchart of the facility crop photosynthetic potential prediction system based on spectral characteristics according to the present invention.

[0076] Figure 5 This is a schematic diagram of the structure of the facility crop photosynthetic potential prediction system based on spectral characteristics according to the present invention.

[0077] Figure 6 This is a schematic diagram of the structure of the facility crop photosynthetic potential prediction system based on spectral characteristics according to the present invention.

[0078] Figure 7This is a flowchart of the process of the spectral characteristics-based photosynthetic potential prediction system for facility crops of the present invention.

[0079] Among them, 1 is the reflected light detection probe, 101 is the housing, 101a is the light shielding platform, 2 is the driving light source unit, 201 is the LED lamp, 202 is the constant current driving circuit, 203 is the light source control circuit, 3 is the reflected light detection unit, 301 is the photoelectric sensor, 302 is the photoelectric conversion circuit, 303 is the signal amplification circuit, 304 is the analog-to-digital conversion circuit, 4 is the main control processing unit, 5 is the power supply unit, 6 is the communication unit, 601 is the wireless communication module, 7 is the control software and display unit, 701 is the photosynthetic potential prediction model, 702 is the server, and 703 is the mobile terminal. Detailed Implementation

[0080] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the embodiments set forth herein; rather, they are provided so that this application will be thorough and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted.

[0081] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0082] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0083] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all promotional information and operations / steps, nor do they necessarily need to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0084] It should be understood that although the terms first, second, third, etc., may be used herein to describe various components, these components should not be limited by these terms. These terms are used to distinguish one component from another. Therefore, the first component discussed below may be referred to as the second component without departing from the teachings of this application. As used herein, the terms "and / or" and similar terms include all combinations of any, multiple, and all of the associated listed items.

[0085] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of exemplary embodiments, and the modules or processes in the drawings are not necessarily essential for implementing this application, and therefore cannot be used to limit the scope of protection of this application.

[0086] Before detailing the embodiments of this application, some terms used in the embodiments of this application will be explained first, so that those skilled in the art can understand them.

[0087] Photosynthetic potential (Fv / Fm), or maximum photochemical quantum yield of PSII, is a key indicator in plant photosynthesis research, used to measure a plant's photosynthetic potential. In photosynthesis, chlorophyll, as a basic pigment, absorbs light energy and transitions from its ground state to an excited state, then returns to the ground state, emitting some energy as fluorescence, forming chlorophyll fluorescence. Fv / Fm reflects the maximum light energy conversion efficiency of the PSII reaction center. Its calculation formula is Fv / Fm = (Fm - Fo) / Fm, where Fo is the initial fluorescence, representing the fluorescence yield when all PSII reaction centers are open; and Fm is the maximum fluorescence, the fluorescence yield when all PSII reaction centers are closed. A higher Fv / Fm value indicates a higher light energy conversion efficiency of the PSII reaction center, a stronger ability of the plant to convert light energy into chemical energy, and thus a better photosynthetic potential. Real-time, non-destructive testing of plant Fv / Fm helps to accurately grasp the physiological state of plants and understand their response to the external environment, which is of great significance in agricultural production and plant research.

[0088] The Standard Normal Variate Transformation (SNV) algorithm is a commonly used method for spectral data preprocessing. Its purpose is to eliminate the influence of differences in the physical properties of samples (such as surface particle size, surface scattering, and optical path variation) on spectral reflectance, thereby improving data quality and enhancing the accuracy of subsequent modeling and analysis.

[0089] Competitive adaptive reweighted sampling (CARS) is a feature variable selection method that combines Monte Carlo sampling with PLS model regression coefficients. Its aim is to optimize the prediction accuracy of the calibration model. It iteratively builds the prediction model to obtain the feature variables most suitable for improving model accuracy. CARS involves progressively evaluating, analyzing, screening, and eliminating features at each wavenumber point in the spectrum, making it suitable for high-dimensional spectral screening.

[0090] Successive projections (SPA) is a forward iterative selection method that extracts wavelength combinations with minimal redundancy and collinearity through vector projection analysis. SPA is a forward feature variable selection method that starts with the first wavenumber and merges a new wavenumber in each iteration until a predetermined number of wavenumbers is reached. Its key feature is selecting wavenumber combinations with minimal redundancy and information content, thus addressing the problem of collinearity.

[0091] Support Vector Regression (SVR) is an extension of Support Vector Machine (SVM) to regression problems and belongs to the supervised learning algorithm. It aims to find an optimal regression function to predict continuous value outputs as accurately as possible. The core idea of ​​SVR is to map the data in the input space to a high-dimensional feature space through a nonlinear mapping, and then find an optimal hyperplane in this high-dimensional space to fit the data. To handle data noise and errors, SVR introduces an insensitive loss function (usually an ε-insensitive loss function), allowing a certain error range (ε) between the predicted and true values. As long as the error is within this range, the prediction is considered accurate, and no loss is incurred.

[0092] First refer to Figure 1 Taking grapes as an example, the following is a specific embodiment of a method for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics.

[0093] S1. Obtain sample data

[0094] Grapes were cultured under the same temperature, humidity, and carbon dioxide concentration conditions, with six light intensity gradients, and the photon flux densities were 90, 140, 220, 280, and 340 μmol·m⁻¹. -2 ·s -1To determine the differences in grape leaves under different light treatments, leaves were randomly selected as experimental samples. The photosynthetic potential (Fv / Fm) and visible-near-infrared reflectance spectra of the crop leaves were measured as sample data.

[0095] S2. Preprocess the sample data

[0096] The standard normal transformation algorithm is an effective method for spectral data preprocessing, capable of eliminating interference factors such as scattering effects and baseline drift, thereby improving data modeling performance. The steps for sample data preprocessing using the standard normal transformation are as follows:

[0097] Calculate the mean of each variable: For the collected spectral data matrix, calculate the average value of each wavelength point across all samples. The formula is as follows:

[0098]

[0099] in, Let x represent the mean of the j-th wavelength point, n represent the number of samples, and x represent the mean of the wavelength points. ij This represents the spectral value of the i-th sample at the j-th wavelength point.

[0100] Calculate the standard deviation of each variable: Calculate the standard deviation of each wavelength point across all samples using the following formula:

[0101]

[0102] Among them, s j This represents the standard deviation at the j-th wavelength point.

[0103] Perform SNV transformation: Perform standard normal variable transformation on each wavelength point of each sample. The calculation formula is as follows:

[0104]

[0105] Among them, z ij This represents the spectral value of the i-th sample after SNV transformation at the j-th wavelength point.

[0106] S3. Extract characteristic wavelengths

[0107] The high linear repeatability and redundancy of the leaf reflectance spectrum bands lead to model complexity and decreased accuracy. Feature wavelength screening can reduce the number of wavelength variables and improve the model's prediction speed. Therefore, a competitive adaptive reweighting algorithm is first used to extract feature wavelength combinations that are beneficial to the prediction model's accuracy, and then a continuous projection method is used to screen out redundant variables with high repeatability.

[0108] The competitive adaptive reweighting algorithm first employs Monte Carlo sampling, randomly selecting 80% of the samples from the dataset each time to enter the modeling set, and using the remaining 20% ​​as the prediction set to build the PLS model. The number of Monte Carlo sampling iterations, N, is pre-defined. The absolute value weight of the regression coefficients in the PLS model is recorded for each sampling process, calculated using the following formula:

[0109]

[0110] Among them, b i Let be the regression coefficient of the i-th variable.

[0111] By forcibly removing wavelengths with relatively small absolute value weights in the regression coefficients using the exponential decay function (EDF), the proportion R of retained wavelength points is obtained from the EDF when establishing the PLS model based on Monte Carlo sampling for the i-th time. i The calculation formula is as follows:

[0112]

[0113] The formulas for calculating μ and k are as follows:

[0114]

[0115] Where N is the number of samplings and n is the number of original wavelength points.

[0116] The number of samples selected each time is R. i Using n wavelength variables, perform PLS modeling and calculate the root mean square error (RMSECV) of cross-validation.

[0117] After N sampling cycles, the CARS algorithm obtains N candidate feature wavelength subsets and their corresponding RMSECV values, and selects the wavelength variable subset corresponding to the minimum RMSECV value as the feature wavelength.

[0118] The continuous projection method first selects an initial wavelength, randomly choosing an initial wavelength from the spectral matrix as a characteristic variable;

[0119] To perform projection calculations, for each remaining variable, calculate its projection onto the set of selected feature variables. The formula for this calculation is as follows:

[0120]

[0121] Where, x j The variable currently under consideration, x k(n-1) P is the feature variable selected in the previous iteration. xj It is x j In x k(n-1) Projection on;

[0122] Based on the projection results, the variable that maximizes the magnitude of the projection vector is selected as the next feature variable:

[0123]

[0124] Where s is the set of indices of the features that were not selected into the set of features, and k(n) is the index of the features selected in this iteration;

[0125] Remove the selected feature variables from the original dataset and update the set of unselected feature variables;

[0126] Repeat the above steps until the preset number of features is reached, and output the final set of selected feature variables.

[0127] By combining CARS and SPA, it is possible to minimize the number of feature wavelengths while considering model accuracy.

[0128] S4. Establish a photosynthetic potential prediction model using GA-SVR.

[0129] Reference Figure 2 It illustrates the overall process of establishing a photosynthetic potential prediction model using the GA-SVR algorithm. The establishment of a crop photosynthetic potential prediction model includes the following steps:

[0130] Data normalization is performed to ensure that data across all dimensions are on the same order of magnitude, preventing imbalanced samples caused by excessive data differences, which could lead to the final model deviating from the accurate hyperplane. Linear normalization is used for the normalization operation, with the interval [0,1], and the formula is as follows:

[0131]

[0132] x * Here, x represents the normalized data, and x represents the data to be normalized. max and x min These are the maximum and minimum values ​​in the data sequence, respectively.

[0133] The radial basis function was selected as the kernel function. The genetic algorithm was used to find the optimal values ​​of parameters c and g. The index range of c was set to [0.01, 10], and the number of genes was 6.

[0134] Set the index range of g to [0.01, 5], the number of genes to 5, and the encoding method to binary encoding. The encoding formula is as follows:

[0135]

[0136] The decoding formula is:

[0137]

[0138] Where b is the encoded binary string, m is the number of characters in the binary string taken from the chromosome, and a is the required decimal encoding number. max a is the largest decimal number in the encoding space. min It is the smallest decimal number in the encoding space.

[0139] Using the decoded parameters c and g as key parameters, a regression-based support vector machine is trained on the training set. The fitness of chromosomes is evaluated by calculating the determination coefficients of the test set data, and a roulette wheel selection mechanism is used for population elimination. In the generated offspring, the individual with the highest fitness from the parent generation is retained, while the individual with the lowest fitness is replaced. Subsequently, the offspring further evolves through crossover, mutation, and selection operations, eventually iterating to the optimal parameters c and g. A selection mutation coefficient of 0.2 and a crossover coefficient of 0.8 are used, and 50 generations of iterative evolution are performed to obtain the optimal parameters c and g.

[0140] Based on the training set samples and optimal parameters, an SVR model is established, resulting in the photosynthetic potential model f(x) = w T x+b, where w is the regression coefficient matrix and b is the bias vector.

[0141] Introducing slack variable ζ into the model i With soft margins, the regression-type support vector machine can be transformed into:

[0142]

[0143] Among them, y i Let Fv / Fm be the value of the i-th sample, ε be the insensitive loss function, and ζ be the value of the i-th sample. i and Let m be the total number of training samples for the model, and m be the slack variable. By introducing Lagrange multipliers and taking the partial derivative of the Lagrange function in the above equation, it can be transformed into the dual problem of a regression-type support vector machine, and the hyperplane can be solved according to the KKT conditions.

[0144] Validation of the Fv / Fm photosynthetic potential prediction model

[0145] The Fv / Fm photosynthetic potential prediction model was validated using test set data, such as... Figure 3 As shown, the coefficient of determination R between the predicted and actual values 2 =0.93, and the root mean square error (RMSE) =0.021, indicating that the model has high prediction accuracy.

[0146] To test the robustness of the prediction model, the experiment was repeated and the model performance was tested under fluctuating temperature (20-30℃) and humidity (50-70%RH) conditions. The results are shown in Table 1, which shows the coefficient of determination R of the model under different conditions. 2The value is stable at 0.90-0.93, and the root mean square error (RMSE) is ≤0.025, indicating that it has strong environmental adaptability.

[0147] Table 1 shows the model performance under different environmental conditions.

[0148]

[0149] Reference Figure 4 This paper illustrates the control flow of a facility crop photosynthetic potential prediction system based on spectral characteristics. It mainly includes a main control processing unit 4, a driving light source unit 2, a reflected light detection unit 3, a power supply unit 5, a communication unit 6, and control software and a display unit 7.

[0150] The main control processing unit 4 has a microcontroller used to control the switching of the driving light source unit 2, collect the reflected light signal from the reflected light detection unit 3, and transmit the processed data to the communication unit 6. The microcontroller is an STM32F103 microcontroller, which communicates with the constant current driving circuit 202 and the analog-to-digital converter circuit 304 via standard I / O interfaces, and is connected to the wireless communication module 601 via a USART module.

[0151] The driving light source unit 2 includes a constant current driving circuit 202 and a light source control circuit 203. The constant current driving circuit 202 provides a stable constant current to the light source, and the light source control circuit 203 controls the emission of detection light from the light source. The constant current driving circuit 202 uses a linear LED driver chip to provide a stable constant current to the light source to meet the requirements of the LED's DC driving mode. The light source control circuit 203 uses a 6-channel ring-shaped narrow bandwidth LED array to emit detection light with a characteristic wavelength as the center band to the crop leaves, and uses MOSFETs to realize the switching of each LED.

[0152] LED lights are typical constant current devices, using DC drive. Therefore, the linear LED driver chip CN5501 is used to provide a stable constant current. The formula for calculating the output current of CN5501 is as follows:

[0153]

[0154] Among them, I LED For the output current, R CS This is the resistor between the CS pin of CN5501 and ground.

[0155] The light source control circuit 203 can use an N-channel MOSFET to implement LED switching. When the gate-source voltage of the MOSFET is 0, there is no conductive channel between its drain and source, which can be regarded as an open circuit. When the gate-source voltage is greater than the MOSFET's turn-on voltage, a conductive channel is formed between its drain and source. The gate of each MOSFET is connected in series with a resistor and then connected to the core processor's I / O port. When the I / O port outputs a high level, its drain-source is turned on, and the current generated by the constant current drive circuit 202 flows into the light source.

[0156] The reflected light detection unit 3 is used to detect the intensity of diffuse reflected light from the leaf blades and to perform signal conversion and processing. It includes a photoelectric conversion circuit 302, a signal amplification circuit 303, and an analog-to-digital conversion circuit 304. The photoelectric conversion circuit 302 converts the optical signal into an electrical signal, the signal amplification circuit 303 amplifies the electrical signal, and the analog-to-digital conversion circuit 304 converts the analog signal into a digital signal that can be processed by a microcontroller. The photoelectric conversion circuit 302 uses a photodiode as the photoelectric sensor, which has characteristics such as high sensitivity, good high-frequency performance, small size, and low cost. Combined with a narrow-wavelength bandwidth light source, it meets the design requirements of this detection equipment.

[0157] Specifically, a first-order low-pass filter circuit can be used to eliminate high-frequency noise in the power supply signal. A photodiode is placed under a 5V reverse voltage; when illuminated, it generates a photocurrent that passes through a 1kΩ sampling resistor. As the light intensity increases, the photocurrent increases, and the voltage across the sampling resistor increases, thus achieving the conversion from optical signal to electrical signal.

[0158] Signal amplifier circuit 303 is a non-inverting proportional operational circuit, and its signal gain coefficient is calculated as follows:

[0159]

[0160] Among them, R f R is the feedback resistor, and R is the external resistor connected to the reverse terminal.

[0161] Specifically, signal amplification can be achieved using an operational amplifier. In the signal amplification circuit 303, the voltage signal across the sampling resistor is input from the non-inverting terminal of the operational amplifier, amplified, output from the output terminal, and then transmitted to the analog-to-digital converter circuit 304 after decoupling by a capacitor.

[0162] The analog-to-digital converter circuit 304 uses an analog-to-digital converter chip to convert voltage signals. After the voltage signals are converted, they are transmitted to the STM32 core processor.

[0163] like Figure 5 and Figure 6As shown, a preferred embodiment is illustrated where the driving light source unit 2 and the reflected light detection unit 3 are disposed within the reflected light detection probe 1. The reflected light detection probe 1 includes a housing 101 with an open top, the opening of which corresponds to the target area of ​​the blade to be detected. A photoelectric sensor 301, serving as a detection element, is placed at the center of the bottom of the housing 101. This sensor receives the reflected light generated after the blade to be detected is illuminated by the LED lamp 201. To ensure that the received signal originates solely from the target surface and to avoid interference from other sources such as scattered light from the LED lamp 201 itself or the surrounding environment, a light-shielding platform 101a is also provided at the bottom of the housing 101. The light-shielding platform 101a is positioned around the upper outer side of the photoelectric sensor 301, blocking stray light emitted by the LED lamp 201 from entering the photoelectric sensor 301.

[0164] Multiple LEDs 201 are circumferentially arranged on the lower inner wall of the housing 101. Specifically, six LEDs 201 with different emission wavelengths are arranged in a ring, with a 60° angular interval between each pair of adjacent LEDs 201. This design enables uniform illumination of the target area. The light from the multiple LEDs 201 converges at a certain angle toward the center of the top of the housing 101, and the light direction of the LEDs 201 forms a 45° angle with the blade to be inspected, thus illuminating the blade.

[0165] Power supply unit 5 provides operating voltage to each unit. Specifically, a lithium battery can be selected as the power source. Lithium batteries have high rated voltage, high energy density, small size, light weight, and long lifespan. They also have high power handling capacity and strong temperature adaptability, making them widely used in small electronic devices and precision instruments, and thus an ideal power supply option for this device. Meanwhile, to meet the operational requirements of the hardware circuitry, a linear regulator can be used to generate different stable voltages; capacitors are connected to the input and output terminals of the linear regulator to enhance voltage signal stability.

[0166] Communication unit 6 includes a wireless communication module 601 and peripheral circuitry, used to upload the collected reflected light data to the database of server 702. Specifically, the wireless communication module 601 (Wi-Fi module) can adopt the ESP8266 series module, and the STM32 core processor uses the USART transceiver module to send the collected light intensity data to the ESP8266 chip through the RX and TX ports.

[0167] The control software and display unit 7 is used to read data from the database of server 702, call the crop photosynthetic potential prediction model 701 to calculate the Fv / Fm value, store the detection results, and send them to the mobile terminal 703 for display. Specifically, the control software and display unit 7 reads reflected light data from the database of server 702, then calls the prediction model to calculate the Fv / Fm value corresponding to the reflected light data, stores the detection results in the database and sends them to the mobile terminal, and finally displays the detection results.

[0168] The workflow of the spectral characteristic-based photosynthetic potential prediction system for greenhouse crops includes:

[0169] Power supply unit 5 provides a stable operating voltage to all units of the system, and the system starts up;

[0170] The STM32F103 microcontroller in main control processing unit 4 initializes each I / O interface;

[0171] The main control processing unit 4 sequentially turns on the LEDs via the light source control circuit 203, with each LED turned on for 300ms. During this period, the reflected light detection unit collects, converts, and processes the reflected light signal. Specifically, the photoelectric conversion circuit 302 converts the light signal into an electrical signal, the signal amplification circuit 303 amplifies the electrical signal, and the analog-to-digital conversion circuit 304 converts the analog signal into a digital signal and transmits it to the STM32 core processor.

[0172] Repeat step 3 until all 6 LEDs have completed data acquisition;

[0173] The STM32 core processor uploads the stored reflected light data to the server 702 database through the ESP8266 module of communication unit 6;

[0174] The control software and display unit 7 reads data from the database of server 702, calls the crop photosynthetic potential prediction model 701 to calculate the Fv / Fm value, stores the detection results and sends them to the mobile terminal 703 for display.

[0175] It should be noted that the above hardware design is merely a specific example illustrating the specific structure and appearance of the facility crop photosynthetic potential detection system provided by this invention in practical applications, the installation positions of each module, and the specific hardware selection in each module. The system's software settings are matched to the hardware design. The specific settings of the hardware and software designs can be adjusted according to actual needs; this embodiment does not limit this.

[0176] The present invention proposes a method and detection system for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics, which can be used to accurately assess the physiological state of crops. By utilizing specific spectral characteristic wavelengths for detection, this method not only reduces costs but also improves detection speed, better meeting practical needs and demonstrating high practicality and convenience.

[0177] The embodiments and examples presented herein are provided to best illustrate embodiments of this application and its particular applications, thereby enabling those skilled in the art to implement and use this application. However, those skilled in the art will understand that the above description and examples are provided for ease of illustration and example only. The descriptions presented are not intended to cover all aspects of this application or to limit this application to the precise forms disclosed.

Claims

1. A method for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics, characterized in that, Includes the following steps: S1. Obtaining sample data: Set up multiple cultivation environments with the same temperature, humidity, and carbon dioxide concentration under different light gradients. After the crops under different light treatments show differences, randomly select crop leaves as experimental samples and measure the photosynthetic potential Fv / Fm and visible-near infrared reflectance spectra of the crop leaves as sample data. S2. Preprocessing of sample data: Standard normal variable transformation is used for data preprocessing to eliminate the influence of sample surface particle size, surface scattering and optical path variation on the reflectance spectrum; S3. Extracting Feature Wavelengths: First, a competitive adaptive reweighting algorithm is used to extract feature wavelength combinations that are beneficial to the accuracy of the prediction model. Then, the continuous projection method is used to filter out redundant variables with high repetition. Finally, the training set and the test set are randomly divided at a ratio of 4:

1. S4. Using the reflected light intensity of the spectral characteristic wavelengths of the training set samples as input and the crop's Fv / Fm as output, a crop photosynthetic potential prediction model is established using a regression-type support vector machine algorithm. S5. Use photosynthetic potential prediction models to predict crop photosynthetic potential; The steps of the competitive adaptive reweighting algorithm are as follows: The Monte Carlo sampling method is used. In each sampling iteration, 80% of the samples are randomly selected from the dataset to enter the modeling set, and the remaining 20% ​​is used as the prediction set to build the PLS model. The number of sampling iterations N for the Monte Carlo method is predetermined, and the absolute value weight of the regression coefficients in the PLS model is recorded for each sampling process. The calculation formula is as follows: Among them, b i Let be the regression coefficient of the i-th variable; The exponential decay function (EDF) is used to forcibly remove wavelengths with relatively small absolute value weights in the regression coefficients. In the i-th iteration of the PLS model based on Monte Carlo sampling, the proportion R of the retained wavelength points is obtained from the EDF. i The calculation formula is as follows: The formulas for calculating μ and k are as follows: Where N is the number of samplings, and n is the number of original wavelength points; The number of samples selected each time is R. i For *n wavelength variables, perform PLS modeling and calculate the root mean square error of cross-validation (RMSECV). After N sampling cycles, the CARS algorithm obtains N candidate feature wavelength subsets and their corresponding RMSECV values, and selects the wavelength variable subset corresponding to the minimum RMSECV value as the feature wavelength.

2. The method for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics according to claim 1, characterized in that, The steps for the standard normal variable transformation are as follows: Calculate the mean of each variable using the following formula: in, Let x represent the mean of the j-th wavelength point, n represent the number of samples, and x represent the mean of the wavelength point. ij This represents the spectral value of the i-th sample at the j-th wavelength. The standard deviation of each variable is calculated using the following formula: Among them, s j This represents the standard deviation of the j-th wavelength point; The SNV transformation is calculated using the following formula: Among them, z ij This represents the spectral value of the i-th sample after SNV transformation at the j-th wavelength point.

3. The method for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics according to claim 1, characterized in that, The steps of the continuous projection method are as follows: Choose an initial wavelength by randomly selecting one from the spectral matrix as a characteristic variable; To perform projection calculations, for each remaining variable, calculate its projection onto the set of selected feature variables. The formula for this calculation is as follows: Where, x j The variable currently under consideration, x k(n-1) P is the feature variable selected in the previous iteration. xj It is x j In x k(n-1) Projection on; Based on the projection results, the variable that maximizes the magnitude of the projection vector is selected as the next feature variable: Where s is the set of indices of the features that were not selected into the set of features, and k(n) is the index of the features selected in this iteration; Remove the selected feature variables from the original dataset and update the set of unselected feature variables; Repeat the above steps until the preset number of features is reached, and output the final set of selected feature variables.

4. The method for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics according to claim 1, characterized in that, The photosynthetic potential prediction model includes the following steps: Read in the modeling data and perform data normalization processing; Select the modeling parameters, including the regularization parameter c, the kernel function, and the kernel function parameter g; A regression-type support vector machine model is established based on the training set samples and optimal parameters to obtain a photosynthetic potential prediction model.

5. The method for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics according to claim 4, characterized in that, The data normalization process uses a linear normalization method with an interval of [0,1], and the formula is as follows: Where, x * Here, x represents the normalized data, and x represents the data to be normalized. max and x min These are the maximum and minimum values ​​in the data sequence, respectively. The kernel function used is the radial basis function. The optimal values ​​of parameters c and g are found using a genetic algorithm. The index range of c is set to [0.01, 10], and the number of genes is 6. Set the index range of g to [0.01, 5], the number of genes to 5, and the encoding method to binary encoding. The encoding formula is as follows: The decoding formula is: Where b is the encoded binary string, m is the number of characters in the binary string taken from the chromosome, and a is the required decimal encoding number. max a is the largest decimal number in the encoding space. min It is the smallest decimal number in the encoding space.

6. A system for predicting the photosynthetic potential of greenhouse crops based on spectral characteristics, characterized in that, To implement the prediction method according to any one of claims 1-5, comprising: The main control processing unit has a microcontroller, which is used to control the switching of the driving light source unit, collect the reflected light signal of the reflected light detection unit, and transmit the processed data to the communication unit. A driving light source unit includes a constant current driving circuit and a light source control circuit, wherein the constant current driving circuit is used to provide a stable constant current to the light source, and the light source control circuit is used to control the light source to emit detection light; The reflected light detection unit is used to detect the intensity of diffuse reflected light from the leaves and to perform signal conversion and processing. It includes a photoelectric conversion circuit, a signal amplification circuit, and an analog-to-digital conversion circuit. The photoelectric conversion circuit is used to convert the optical signal into an electrical signal, the signal amplification circuit is used to amplify the electrical signal, and the analog-to-digital conversion circuit is used to convert the analog signal into a digital signal that can be processed by the microcontroller. The power supply unit is used to provide operating voltage to each unit; The communication unit, which includes a wireless communication module and peripheral circuits, is used to upload the collected data to the server database. The control software and display unit are used to read data from the server database, call the crop photosynthetic potential prediction model to calculate the Fv / Fm value, store the detection results and send them to the mobile terminal for display.

7. The photosynthetic potential prediction system for facility crops based on spectral characteristics according to claim 6, characterized in that, The microcontroller is an STM32F103 microcontroller, which communicates with the constant current drive circuit and the analog-to-digital conversion circuit through standard I / O interfaces, and is connected to the wireless communication module through a USART module.

8. The photosynthetic potential prediction system for facility crops based on spectral characteristics according to claim 6, characterized in that, The driving light source unit and the reflected light detection unit are disposed inside the reflected light detection probe. The reflected light detection probe includes a housing with an open top. Multiple LEDs are arranged circumferentially on the lower inner sidewall of the housing. The light from the multiple LEDs converges at a certain angle toward the center of the top of the housing, thereby illuminating the blade to be tested. A photoelectric sensor is disposed at the bottom of the housing, which is used to receive the reflected light formed after the blade to be tested is illuminated by the LEDs.

9. A system for predicting the photosynthetic potential of facility crops based on spectral characteristics according to claim 8, characterized in that, The bottom of the housing is also provided with a light-shielding platform to block stray light emitted by the LED from interfering with the photoelectric sensor.

Images