A method and system for predicting polysaccharides in Dendrobium officinale based on hyperspectral imaging

Abstract

An embodiment of the present invention provides a method and system for predicting polysaccharides of Dendrobium officinale based on hyperspectral imaging, belonging to the technical field of polysaccharide detection. The prediction method includes: obtaining a Dendrobium growth model of the growth stage of Dendrobium and the corresponding chlorophyll content; obtaining the current growth stage of Dendrobium; determining the corresponding chlorophyll content according to the current growth stage of Dendrobium and the Dendrobium growth model; and inputting the chlorophyll content into a polysaccharide content prediction model to obtain the polysaccharide content of Dendrobium. The present invention uses the chlorophyll content as an independent variable, which is beneficial to indicating the optimization direction of the gradient update of the model parameters in the modeling process. Moreover, by using the hyperspectral image data of the leaves and constructing a two-stage model of "hyperspectral inversion chlorophyll model - Dendrobium growth model - polysaccharide content prediction model", problems such as difficult parameter adjustment and poor model accuracy in directly constructing a regression model of "hyperspectral inversion chlorophyll model - polysaccharide content prediction model" are avoided, and it has strong applicability and is easy to implement.

Description

Technical Field

[0001] The present invention relates to the technical field of polysaccharide detection, and particularly to a method and system for predicting polysaccharides in Dendrobium officinale based on hyperspectral imaging. Background Art

[0002] Estimation of plant physiological phenotypes is an important task in agricultural production and planting management. With the improvement of the automation, informatization, and intelligentization levels of agricultural production, plant physiological phenotype estimation technology has gradually become a research hotspot. The current main plant phenotype prediction technologies include spaceborne and airborne remote sensing, proximal sensing, meteorological models, machine learning, etc. These technologies still have problems such as low modeling accuracy, poor universality, and low detection efficiency in practice. Estimation of plant physiological parameters based on spectral data and machine learning is a type of method that has received extensive attention in recent years. This type of method analyzes the spectral band data reflected or absorbed by plant organs such as leaves, roots, and rhizomes to obtain crop growth information, and then models and estimates the corresponding plant physiological phenotypes (such as chlorophyll content, vegetation index, etc.). However, the data obtained by existing plant physiological instruments such as spectrometers are mostly point data on a single leaf, lacking data on the leaves of the plant population; the instrument operation is complex; the plant species corresponding to the modeling algorithm are single, mostly leafy vegetables, and cannot meet the requirements for obtaining a large amount of plant spectral data and modeling accuracy in actual production.

[0003] As a high-value economic crop, the main component of Dendrobium officinale is polysaccharides, and the content of polysaccharides determines the quality of Dendrobium officinale. Therefore, there is an urgent need for a low-cost, simple, fast, and accurate method for estimating the polysaccharide content of Dendrobium officinale, which is not only applicable to large scientific research institutions but also for individual planting production, and can perform algorithm modeling on the spectral imaging data of specific wavelengths and the polysaccharide content of Dendrobium officinale to reduce the cost of physiological phenotype detection and quickly and scientifically improve the estimation accuracy and efficiency. Summary of the Invention

[0004] The purpose of the embodiments of the present invention is to provide a method and system for predicting polysaccharides in Dendrobium officinale based on hyperspectral imaging, which solves the problem of predicting the polysaccharide content of Dendrobium officinale in a low-cost, simple, fast, and accurate manner.

[0005] To achieve the above purpose, the embodiments of the present invention provide a method for predicting polysaccharides in Dendrobium officinale based on hyperspectral imaging, and the prediction method includes:

[0006] Obtaining a Dendrobium growth model with the growth stage of Dendrobium and the corresponding chlorophyll content;

[0007] Obtaining the current growth stage of Dendrobium;

[0008] Determining the corresponding chlorophyll content according to the current growth stage of Dendrobium and the Dendrobium growth model;

[0009] Input the chlorophyll content into the polysaccharide content prediction model to obtain the polysaccharide content of Dendrobium officinale.

[0010] Optionally, obtain a Dendrobium growth model for the growth stage of Dendrobium and the corresponding chlorophyll content, including:

[0011] Obtain the hyperspectral image data of Dendrobium leaves;

[0012] Construct a hyperspectral inversion chlorophyll model;

[0013] Input the hyperspectral image data into the hyperspectral inversion chlorophyll model to obtain the chlorophyll content of Dendrobium leaves;

[0014] Construct a Dendrobium growth model based on the chlorophyll content and the growth stage of Dendrobium.

[0015] Optionally, construct a hyperspectral inversion chlorophyll model, including:

[0016] Construct the hyperspectral inversion chlorophyll model according to the partial least squares regression method;

[0017] Divide the hyperspectral image data into a training set and a validation set;

[0018] Train the hyperspectral inversion chlorophyll model using the training set;

[0019] Estimate the performance index of the trained hyperspectral inversion chlorophyll model using the validation set;

[0020] Judge whether the performance index of the hyperspectral inversion chlorophyll model meets the requirements;

[0021] If it is determined that the performance index of the hyperspectral inversion chlorophyll model meets the requirements, save the hyperspectral inversion chlorophyll model.

[0022] Optionally, divide the hyperspectral image data into a training set and a validation set, including:

[0023] Sort the hyperspectral image data in ascending order according to the chlorophyll content;

[0024] Divide every four hyperspectral image data into a group;

[0025] Use the third hyperspectral image data in each group as the validation set, and the rest as the training set.

[0026] Optionally, input the chlorophyll content into the polysaccharide content prediction model to obtain the polysaccharide content of Dendrobium officinale, including:

[0027] Simplify the memory cell based on the long short-term memory network (LSTM), and set a reset gate and an update gate. The reset gate is used to control the influence weight of the historical state on the current candidate state, and the update gate is used to control the influence weights of the historical state and the current candidate state on the current hidden state;

[0028] Construct the polysaccharide content prediction model according to the chlorophyll content, the actual polysaccharide content, the reset gate and the update gate, and output the predicted value of the polysaccharide content;

[0029] Determine the loss function of the polysaccharide content prediction model, determine the training loss of the polysaccharide content prediction model according to the actual polysaccharide content and the predicted value of the polysaccharide content, and update the parameter weights of the polysaccharide content prediction model through backpropagation.

[0030] Optionally, constructing the polysaccharide content prediction model according to the chlorophyll content, the actual polysaccharide content, the reset gate and the update gate, and outputting the predicted value of the polysaccharide content includes:

[0031] Preprocess the chlorophyll content;

[0032] Input the chlorophyll content into the polysaccharide content prediction model, and output the predicted value of the polysaccharide content.

[0033] On the other hand, the present invention provides a Dendrobium officinale polysaccharide prediction system based on hyperspectral imaging. The system includes:

[0034] An image acquisition module, which is used to acquire hyperspectral image data of Dendrobium leaves;

[0035] A processor, which is communicatively connected to the image acquisition module, and is used to obtain the current Dendrobium polysaccharide content according to the hyperspectral image data.

[0036] Optionally, the processor is used to execute:

[0037] Obtain a Dendrobium growth model with the Dendrobium growth stage and the corresponding chlorophyll content;

[0038] Obtain the current growth stage of Dendrobium;

[0039] Determine the corresponding chlorophyll content according to the current growth stage of Dendrobium and the Dendrobium growth model;

[0040] Input the chlorophyll content into the polysaccharide content prediction model to obtain the Dendrobium polysaccharide content.

[0041] Optionally, obtaining a Dendrobium growth model with the Dendrobium growth stage and the corresponding chlorophyll content includes:

[0042] Obtain the hyperspectral image data of the Dendrobium leaves;

[0043] Construct a hyperspectral inversion chlorophyll model;

[0044] Input the hyperspectral image data into the hyperspectral inversion chlorophyll model to obtain the chlorophyll content of Dendrobium leaves;

[0045] Construct a Dendrobium growth model based on the chlorophyll content and the growth stage of Dendrobium.

[0046] Through the above technical solution, the present invention establishes a Dendrobium growth model by tracking the entire growth cycle of Dendrobium, and then constructs a polysaccharide content prediction model based on the chlorophyll content and the polysaccharide content. Using the chlorophyll content of Dendrobium leaves as an independent variable is conducive to indicating the optimization direction of the model parameter gradient update in the modeling process and reduces the optimization difficulty of the model. Moreover, the hyperspectral image data of the leaves is used instead of the point data obtained by a conventional near-infrared spectrometer. Through the construction of a two-stage model of "hyperspectral - chlorophyll time series - Dendrobium polysaccharide", problems such as difficult parameter adjustment and poor model accuracy in directly constructing a regression model of "hyperspectral inversion chlorophyll model - polysaccharide content prediction model" are avoided. The present invention solves the problem that it is difficult to simply, low-cost, accurately and efficiently obtain the polysaccharide content of Dendrobium in traditional prediction, has very important application value and social and economic benefits, can be widely applied to multiple fields such as agriculture and gardening, has strong applicability, the method is simple and effective, and is easy to implement.

[0047] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent specific implementation part. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings are used to provide a further understanding of the embodiments of the present invention, and constitute a part of the specification, and are used together with the following specific implementation to explain the embodiments of the present invention, but do not constitute a limitation to the embodiments of the present invention. In the drawings:

[0049] Figure 1 is a flowchart of a prediction method according to an embodiment of the invention;

[0050] Figure 2 is a flowchart of constructing a Dendrobium growth model according to an embodiment of the invention;

[0051] Figure 3 is a flowchart of constructing a hyperspectral inversion chlorophyll model according to an embodiment of the invention;

[0052] Figure 4 is a flowchart of dividing data according to an embodiment of the invention;

[0053] Figure 5 is a flowchart of constructing a polysaccharide yield prediction model according to an embodiment of the invention;

[0054] Figure 6 It is a flowchart for obtaining the predicted value of polysaccharide content according to an embodiment of the invention. Specific Embodiment

[0055] The following will describe in detail the specific embodiments of the embodiments of the present invention in conjunction with the accompanying drawings. It should be understood that the specific embodiments described herein are only for the purpose of illustrating and explaining the embodiments of the present invention, and are not used to limit the embodiments of the present invention.

[0056] It should be noted that the acquisition, transmission, storage, use, processing, etc. of data in the technical solution of this application all comply with the relevant regulations of national laws and regulations. In the embodiments of this application, some industry-existing solutions such as certain software, components, models, etc. may be mentioned, and they should be regarded as exemplary. The purpose is only to illustrate the feasibility in the implementation of the technical solution of this application, but it does not mean that the applicant has already or necessarily used this solution.

[0057] Figure 1 It is a flowchart of a prediction method according to an embodiment of the present invention. In this figure, the prediction method includes:

[0058] In step S1, a Dendrobium growth model of the Dendrobium growth stage and the corresponding chlorophyll content is obtained.

[0059] In step S2, the current growth stage of the Dendrobium is obtained.

[0060] In step S3, the corresponding chlorophyll content is determined according to the current growth stage of the Dendrobium and the Dendrobium growth model.

[0061] In step S4, the chlorophyll content is input into the polysaccharide content prediction model to obtain the Dendrobium polysaccharide content.

[0062] Compared with the prior art, the present invention tracks the entire growth cycle of the Dendrobium, establishes a Dendrobium officinale growth model, and then constructs a polysaccharide content prediction model according to the chlorophyll content and polysaccharide yield. Using the chlorophyll content of the Dendrobium leaves as the independent variable is beneficial to indicating the optimization direction of the model parameter gradient update in the modeling process and reduces the optimization difficulty of the model. The present invention solves the problem that it is difficult to simply, low-cost, accurately and efficiently obtain the Dendrobium polysaccharide content in traditional prediction, has very important application value and social and economic benefits, can be widely applied to multiple fields such as agriculture and gardening, has strong applicability, the method is simple and effective, and is easy to implement.

[0063] With the popularization of hyperspectral technology, researchers have found that hyperspectral reflectance data may contain various physiological and biochemical information of plants, and the hyperspectral reflectance data can be used to establish prediction methods and models for plant phenotypic traits. However, the existing hyperspectral technology is mostly used for "point" measurements of plant leaves and cannot meet the measurement requirements at the overall plant or canopy level. On the other hand, the plant physiological phenotype regression model established based on the spectral reflectance data measured at points is restricted by the limited point measurement data, and the modeling accuracy cannot meet the monitoring requirements of actual production and is difficult to apply. The hyperspectral imaging technology can obtain hyperspectral image data at the overall plant or canopy level of plants, making up for the deficiencies of conventional leaf "point" measurements and being suitable for rapid, non-destructive, and comprehensive growth observations of plants. However, the increase in the dimension of hyperspectral image data has increased the difficulty of data mining. When directly modeling the "hyperspectral image - polysaccharide content" regression model using traditional machine learning algorithms, it is difficult to adjust the model parameters and converge. Therefore, in this embodiment, the present invention constructs a two-stage model of "hyperspectral inversion chlorophyll model - Dendrobium growth model - polysaccharide content prediction model", avoiding problems such as difficult parameter adjustment and poor model accuracy in directly constructing the "hyperspectral inversion chlorophyll model - polysaccharide content prediction model" regression model. For the specific steps of the Dendrobium growth model for obtaining the growth stage of Dendrobium and the corresponding chlorophyll content, there are various methods known to those skilled in the art. In one example of the present invention, as Figure 2 shown, it may include:

[0064] In step S11, hyperspectral image data of Dendrobium leaves is obtained. Among them, the wavelength range of the obtained hyperspectral image data is 400 - 1000 nm, and the imaging area of the above-ground part of Dendrobium, namely the stem and leaves, is extracted.

[0065] In step S12, a hyperspectral inversion chlorophyll model is constructed. During the construction of the hyperspectral inversion chlorophyll model, historical data is required to train the hyperspectral inversion chlorophyll model. Among them, when collecting hyperspectral image data, a hyperspectral imager is used to measure the reflectance spectral imaging data, and the spectral acquisition range is 400 - 1000 nm. Set the sampling time points and intervals, and use the hyperspectral imaging device to regularly collect data of the above-ground part of Dendrobium at the same location for chlorophyll content detection. When sampling once, data is collected multiple times and averaged to improve the accuracy. The specific preparation method includes:

[0066] Place the prepared leaves on the light source imaging table to collect data. Before measurement, use a standard diffuse reflection whiteboard made of polytetrafluoroethylene for standard spectral calibration. Considering the non-uniform characteristics of the leaves, three hyperspectral images of fresh leafy branches from each plant are repeatedly collected. After collecting the spectral data of the fresh leaves of all the above samples, pick the leaves, number them, place them in centrifuge tubes, add an appropriate amount of solvents such as ethanol or acetone, make the chlorophyll fully contact and dissolve with the solvent, shake and mix, and then perform centrifugation to obtain the chlorophyll extract. Use a UV-visible spectrophotometer to measure the absorbance of the chlorophyll extract at different wavelengths, and calculate the chlorophyll content based on the absorbance values at specific wavelengths (663 nm and 645 nm), and finally obtain the hyperspectral image data of the leaves to be trained and the chlorophyll content.

[0067] In step S13, input the hyperspectral image data into the hyperspectral inversion chlorophyll model to obtain the chlorophyll content of Dendrobium leaves. The original plant physiological phenotype regression model based on the spectral reflectance data measured at points is restricted by the limited point measurement data and cannot reflect the overall level of the plant. Therefore, the modeling accuracy cannot meet the monitoring requirements of actual production and it is difficult to apply. Compared with the prior art, the key point of the present invention is to use the hyperspectral imaging data of the leaves instead of the point data obtained by a conventional near-infrared spectrometer. Establish a hyperspectral inversion model for the chlorophyll content of Dendrobium leaves based on hyperspectral imaging equipment, which is accurate, fast in collection, can effectively avoid the cumbersome processes that require a large amount of manual operations in traditional agricultural methods, and improves the accuracy and efficiency of predicting the yield of Dendrobium polysaccharide.

[0068] In step S14, construct a Dendrobium growth model based on the chlorophyll content and the growth stage of Dendrobium.

[0069] The PLSR (Partial Least Squares Regression) model is a statistical method for building prediction models. It builds a linear fitting prediction model by decomposing independent variables and dependent variables into new latent variables and performing linear regression on these latent variables. Specifically, the PLSR model consists of two parts: 1) the decomposition process, which decomposes independent variables and dependent variables into new latent variables; 2) the regression process, which builds a prediction model through linear regression based on the latent variables. In the decomposition process, the PLSR model uses methods such as Orthogonal Partial Least Squares (OPLS) or Non-Orthogonal Partial Least Squares (NPLS) to transform the original independent variable and dependent variable matrices into new latent variable matrices. In the regression process, the PLSR model avoids the problem of multicollinearity by reducing the dimensions of a large number of potentially correlated variables, performs linear regression on these latent variables using the least squares method, and determines the relationships between variables based on the regression coefficients. The PLSR method has strong interpretability, low requirements for computing power, and fast calculation speed. Based on the above statistical data, we use the PLSR algorithm to model the chlorophyll content of leaves and analyze and evaluate the performance of its model. The linear PLSR method uses the average spectral reflectance curve of the hyperspectral image data of each leaf sample as the input variable and the chlorophyll content as the output variable for data fitting. Therefore, in this embodiment, there are various specific steps for constructing the hyperspectral inversion chlorophyll model known to those skilled in the art. In one example of the present invention, as Figure 3 shown, it may include:

[0070] In step S121, a hyperspectral inversion chlorophyll model is constructed according to the partial least squares regression method.

[0071] In step S122, the hyperspectral image data is divided into a training set and a validation set.

[0072] In step S123, the training set is used to train the hyperspectral inversion chlorophyll model.

[0073] In step S124, the validation set is used to estimate the performance metrics of the trained hyperspectral inversion chlorophyll model.

[0074] In step S125, it is determined whether the performance indicators of the hyperspectral inversion chlorophyll model meet the requirements. If it is determined that the performance indicators of the hyperspectral inversion chlorophyll model meet the requirements, step S126 is executed; otherwise, step S123 is executed. Among them, for the types of performance indicators, there can be various types known to those skilled in the art. In an example of the present invention, the hyperspectral inversion chlorophyll model of the present invention uses the coefficient of determination (R2), the root mean square error of prediction (RMSEP), and the ratio of performance to deviation (RPD) to evaluate and compare the prediction performance of the hyperspectral inversion chlorophyll model. Rc2 and Rp2 represent the coefficients of determination of the calibration and prediction sets, respectively. The closer R2 is to 1 and the smaller RMSEP is, the higher the accuracy of the model. In addition, an RPD value less than 1.5 represents very poor model prediction performance; an RPD value between 1.5 and 2.0 indicates poor model prediction performance and can only be used for rough estimation; an RPD value between 2.0 and 2.5 allows for approximate quantitative prediction; an RPD value between 2.5 and 3.0 indicates good model prediction performance and can be used for quantitative analysis; a PRD value higher than 3.0 is classified as an excellent quantitative prediction model and can achieve high-precision quantitative analysis. Specifically, it includes:

[0075] Calculate the coefficient of determination according to formula (1),

[0076]

[0077] Calculate the root mean square error according to formula (2),

[0078]

[0079] Calculate the ratio of performance to deviation according to formula (3),

[0080]

[0081] Wherein, is the chlorophyll content predicted for the i-th one, is the average value of the actually measured chlorophyll content, y i is the actually measured chlorophyll content for the i-th one, and n is the sample data volume, SD p is the standard deviation of the prediction set, R 2 is the coefficient of determination, RMSEP is the root mean square error of prediction, and RPD is the ratio of performance to deviation.

[0082] In step S126, save the hyperspectral inversion chlorophyll model.

[0083] Considering that in order to avoid changing the original spectral information, we use the original spectral data to carry out data analysis. The data is divided into a training set and a validation set in a ratio of 3:1. To reduce the data difference between the training set and the validation set, all the data is sorted from low to high according to the chlorophyll content, and then the third data item is extracted from every four data items as the validation set data. Therefore, there are various steps known to those skilled in the art for dividing the hyperspectral image data into a training set and a validation set. In one example of the present invention, as Figure 4 shown, it includes:

[0084] In S1221, the hyperspectral image data is sorted from low to high according to the chlorophyll content.

[0085] In S1222, every four hyperspectral image data items are divided into a group.

[0086] In S1223, the third hyperspectral image data item in each group is used as the validation set, and the rest are used as the training set.

[0087] In this embodiment, in order to solve the problems of gradient vanishing or explosion existing in the traditional RNN, and the problem of unsatisfactory memory effect when processing long sequence data, the present invention tracks the entire growth cycle of Dendrobium officinale, establishes a growth model of Dendrobium officinale by detecting the chlorophyll content at different growth stages of Dendrobium officinale, and based on the GRU algorithm, establishes a polysaccharide yield prediction model. Specifically, as Figure 5 shown, it includes:

[0088] In step S41, the memory unit is simplified based on the long short-term memory network LSTM, and a reset gate and an update gate are set. The reset gate is used to control the influence weight of the historical state on the current candidate state, and the update gate is used to control the influence weights of the historical state and the current candidate state on the current hidden state. The GRU simplifies the memory unit structure (cell) based on the recurrent neural network RNN, and alleviates the vanishing gradient problem of traditional RNN and LSTM in long sequence learning by introducing a gating mechanism. The GRU controls the information flow through two main gating mechanisms: the reset gate and the update gate. The GRU does not have a separate memory unit, but directly maintains long-term memory through the hidden state. Among them, the reset gate controls how to use the past hidden state to calculate the candidate state at the current moment. If the value of the reset gate is 0, it means ignoring the past hidden state and only considering the input at the current moment. The update gate determines how much of the hidden state at the current moment comes from the hidden state at the previous moment and how much comes from the current candidate state. The update gate enables the GRU to maintain control of historical information at each moment. When calculating the candidate hidden state, the GRU will determine how to combine the input at the current moment and the hidden state at the previous moment according to the output of the reset gate. The final hidden state is obtained by weighted averaging the hidden state at the previous moment and the current candidate hidden state through the update gate. Specifically, it includes:

[0089] Determine the reset gate of the polysaccharide content prediction model according to formula (4),

[0090] r t = Sigmoid(W r [h t-1 , x t +b r ), (4)

[0091] Among them, r t is the output of the reset gate at the current moment, W r is the weight matrix, b r is the bias term, h t-1 is the hidden state at the previous moment, x t is the input at the current moment;

[0092] Determine the update gate of the polysaccharide content prediction model according to formula (5),

[0093] z t = Sigmoid(W z [h t-1 , x t +b z ), (5)

[0094] Among them, z t is the output of the update gate at the current moment, Wz is the weight matrix, b z is the bias term, h t-1 is the hidden state at the previous moment, x t is the input at the current moment;

[0095] Determine the candidate hidden state of the polysaccharide content prediction model according to formula (6),

[0096]

[0097] wherein, is the candidate hidden state at the current moment, r t is the output of the reset gate at the current moment, b h is the bias term, h t-1 is the hidden state at the previous moment, x t is the input at the current moment, and tanh is the hyperbolic tangent activation function;

[0098] Determine the final hidden state of the polysaccharide content prediction model according to formula (7),

[0099]

[0100] wherein, h t is the hidden state at the current moment, z t is the output of the update gate at the current moment, is the candidate hidden state at the current moment, h t-1 is the hidden state at the previous moment, x t is the input at the current moment.

[0101] In step S42, construct a polysaccharide content prediction model based on the chlorophyll content, actual polysaccharide content, reset gate and update gate, and output the predicted value of the polysaccharide content. Specifically, select smooth L1 loss as the loss function, including:

[0102] Determine the loss function of the prediction model according to formula (8)

[0103]

[0104] wherein, x is the predicted value of the polysaccharide content, is the actual polysaccharide content, and smoothL1loss is the loss function.

[0105] In step S43, determine the loss function of the polysaccharide content prediction model, determine the training loss of the polysaccharide content prediction model according to the actual polysaccharide content and the predicted value of the polysaccharide content, and update the parameter weights of the polysaccharide content prediction model through backpropagation.

[0106] In this embodiment, the step of constructing a polysaccharide content prediction model based on chlorophyll content, actual polysaccharide content, reset gate, and update gate and outputting a predicted polysaccharide content value can be various methods known to those skilled in the art. In one example of the present invention, as Figure 6 shown, it may include:

[0107] In step S421, preprocess the chlorophyll content. That is, perform normalization processing on the polysaccharide content detected at different growth stages of Dendrobium officinale, including:

[0108] Perform normalization processing on the chlorophyll content according to formula (9),

[0109]

[0110] where x t is the chlorophyll content at the t-th moment, x' t is the normalized chlorophyll content at the t-th moment, x max and x min are the maximum and minimum values of the chlorophyll content, respectively.

[0111] In step S422, input the chlorophyll content into the polysaccharide content prediction model and output the predicted polysaccharide content value.

[0112] On the other hand, the present invention provides a Dendrobium officinale polysaccharide prediction system based on hyperspectral imaging. The system includes an image acquisition module and a processor. Among them, the image acquisition module is used to collect hyperspectral image data of Dendrobium officinale leaves, and the processor is communicatively connected to the image acquisition module and is used to obtain the current polysaccharide content of Dendrobium officinale according to the hyperspectral image data.

[0113] In this embodiment, the processor is used to execute the following method, including:

[0114] In step S1, obtain a Dendrobium officinale growth model of the growth stage of Dendrobium officinale and the corresponding chlorophyll content.

[0115] In step S2, obtain the current growth stage of Dendrobium officinale.

[0116] In step S3, determine the corresponding chlorophyll content according to the current growth stage of Dendrobium officinale and the Dendrobium officinale growth model.

[0117] In step S4, input the chlorophyll content into the polysaccharide content prediction model to obtain the polysaccharide content of Dendrobium officinale.

[0118] Compared with the prior art, by tracking the entire growth cycle of Dendrobium officinale, establishing a growth model of Dendrobium officinale, and then constructing a polysaccharide content prediction model based on chlorophyll content and polysaccharide yield, using the chlorophyll content of Dendrobium officinale leaves as an independent variable is conducive to indicating the optimization direction of the model parameter gradient update in the modeling process and reducing the optimization difficulty of the model. The present invention solves the problem that it is difficult to simply, low-cost, accurately and efficiently obtain the polysaccharide content of Dendrobium officinale in traditional prediction, has very important application value and social and economic benefits, can be widely applied to multiple fields such as agriculture and gardening, has strong applicability, the method is simple and effective, and is easy to implement.

[0119] With the popularization of hyperspectral technology, researchers have found that hyperspectral reflectance data may contain various physiological and biochemical information of plants, and using hyperspectral reflectance data can establish prediction methods and models for plant phenotypic traits. However, existing hyperspectral technologies are mostly used for "point" measurements of plant leaves and cannot meet the measurement requirements at the whole plant or canopy level of plants. On the other hand, the plant physiological phenotype regression model established based on spectral reflectance data measured at points is restricted by limited point measurement data, and the modeling accuracy cannot meet the monitoring requirements of actual production and is difficult to apply. Hyperspectral imaging technology can obtain hyperspectral image data at the whole plant or canopy level of plants, making up for the deficiencies of conventional leaf "point" measurements and being suitable for rapid, non-destructive and comprehensive growth observation of plants. However, the increase in the dimension of hyperspectral image data increases the difficulty of data mining. Using traditional machine learning algorithms to directly model the "hyperspectral image-polysaccharide content" regression model, it is difficult to adjust model parameters and converge. Therefore, in this embodiment, the present invention constructs a two-stage model of "hyperspectral inversion chlorophyll model-Dendrobium growth model-polysaccharide content prediction model", avoiding problems such as difficult parameter adjustment and poor model accuracy in directly constructing the "hyperspectral inversion chlorophyll model-polysaccharide content prediction model" regression model. For the specific steps of the Dendrobium growth model for obtaining the growth stage of Dendrobium and the corresponding chlorophyll content, there are various ones known to those skilled in the art. In an example of the present invention, as Figure 2 shown, it may include:

[0120] In step S11, hyperspectral image data of Dendrobium officinale leaves is obtained. Among them, the wavelength range of the obtained hyperspectral image data is 400-1000 nm, and the imaging area of the above-ground part of Dendrobium officinale, namely the stem and leaves, is extracted.

[0121] In step S12, a hyperspectral inversion chlorophyll model is constructed. During the construction of the hyperspectral inversion chlorophyll model, historical data is needed to train the model. When collecting hyperspectral image data, a hyperspectral imager is used to measure the reflected spectral imaging data, and the spectral acquisition range is 400 - 1000 nm. Set the sampling time points and intervals, and use the hyperspectral imaging device to regularly collect the data of the above-ground part of Dendrobium at the same position for chlorophyll content detection. When sampling once, data is collected multiple times and averaged to improve the accuracy. The specific preparation method includes:

[0122] Place the prepared leaves on the light source imaging table to collect data. Before measurement, use a standard diffuse reflection whiteboard made of polytetrafluoroethylene for standard spectral calibration. Considering the uneven characteristics of the leaves, 3 fresh leafy branches of each plant are repeatedly collected for hyperspectral images. After collecting the spectral data of the fresh leaves of all the above samples, pick the leaves, number them, and place them in centrifuge tubes. Add an appropriate amount of solvents such as ethanol or acetone to make the chlorophyll fully contact and dissolve with the solvent. Oscillate and mix, and then perform centrifugation to obtain the chlorophyll extract. Use a UV-visible spectrophotometer to measure the absorbance of the chlorophyll extract at different wavelengths. According to the absorbance values at specific wavelengths (663 nm and 645 nm), calculate the chlorophyll content, and finally obtain the hyperspectral image data of the leaves to be trained and the chlorophyll content.

[0123] In step S13, the hyperspectral image data is input into the hyperspectral inversion chlorophyll model to obtain the chlorophyll content of Dendrobium leaves. The original plant physiological phenotype regression model based on the spectral reflectance data measured at points is restricted by the limited point measurement data and cannot reflect the overall level of the plant. Therefore, the modeling accuracy cannot meet the monitoring requirements of actual production and is difficult to apply. Compared with the prior art, the key point of the present invention is to use the hyperspectral imaging data of the leaves instead of the point data obtained by a conventional near-infrared spectrometer. Based on the hyperspectral imaging device, a hyperspectral inversion model for the chlorophyll content of Dendrobium leaves is established, which is accurate and fast in collection, can effectively avoid the cumbersome processes that require a large amount of manual operations in traditional agricultural methods, and improves the accuracy and efficiency of the prediction of Dendrobium polysaccharide yield.

[0124] In step S14, a Dendrobium growth model is constructed according to the chlorophyll content and the growth stage of Dendrobium.

[0125] Through the above technical solutions, the present invention tracks the entire growth cycle of Dendrobium, establishes a growth model of Dendrobium, and then constructs a polysaccharide content prediction model based on the chlorophyll content and polysaccharide content. Using the chlorophyll content of Dendrobium leaves as the independent variable is conducive to indicating the optimization direction of the model parameter gradient update in the modeling process and reduces the difficulty of model optimization. Moreover, the hyperspectral image data of the leaves is used instead of the point data obtained by a conventional near-infrared spectrometer. Through the construction of a two-stage model of "hyperspectral - chlorophyll time series - Dendrobium polysaccharide", problems such as difficult parameter adjustment and poor model accuracy in directly constructing a regression model of "hyperspectral inversion chlorophyll model - polysaccharide content prediction model" are avoided. The present invention solves the problem that it is difficult to simply, low-cost, accurately and efficiently obtain the polysaccharide content of Dendrobium in traditional prediction, has very important application value and social and economic benefits, can be widely applied to multiple fields such as agriculture and gardening, has strong applicability, the method is simple and effective, and is easy to implement.

[0126] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or further includes elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the existence of additional identical elements in the process, method, commodity or device comprising the element.

[0127] The above are only embodiments of the present application and are not used to limit the present application. For those skilled in the art, various changes and modifications can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the scope of the claims of the present application.

Claims

1. A method for estimating polysaccharides of Dendrobium officinale based on hyperspectral imaging, characterized in that: The estimation method includes: The growth model of Dendrobium officinale to obtain the growth stage of Dendrobium officinale and the corresponding chlorophyll content includes: Obtaining hyperspectral image data of Dendrobium leaves; Construct a hyperspectral inversion chlorophyll model; Inputting the hyperspectral image data into the hyperspectral inversion chlorophyll model to obtain the chlorophyll content of the dendrobium leaves; Constructing a growth model of Dendrobium officinale according to the chlorophyll content and the growth stage of Dendrobium officinale; Get the current growth stage of Dendrobium officinale; Determining the corresponding chlorophyll content according to the current growth stage of the dendrobium and the growth model of the dendrobium; Inputting the chlorophyll content into a polysaccharide content estimation model to obtain the dendrobium polysaccharide content, comprising: On the basis of the long short-term memory network LSTM, the memory unit is simplified, and a reset gate and an update gate are set. The reset gate is used to control the influence weight of the historical state on the current candidate state, and the update gate is used to control the influence weight of the historical state and the current candidate state on the current hidden state; The polysaccharide content estimation model is constructed according to the chlorophyll content, the actual polysaccharide content, the reset gate and the update gate, and the polysaccharide content estimation value is output, including: Pre-treating the chlorophyll content; Inputting the chlorophyll content into the polysaccharide content estimation model, and outputting the polysaccharide content estimation value; The loss function of the polysaccharide content estimation model is determined, the training loss of the polysaccharide content estimation model is determined according to the actual polysaccharide content and the polysaccharide content estimation value, and the parameter weights of the polysaccharide content estimation model are updated by back propagation.

2. The estimation method according to claim 1, characterized in that: Construct a hyperspectral inversion chlorophyll model, including: Constructing the hyperspectral inversion chlorophyll model according to the partial least squares regression method; Dividing the hyperspectral image data into a training set and a validation set; Using a training set to train the hyperspectral inversion chlorophyll model; Using the validation set to estimate the performance index of the hyperspectral chlorophyll inversion model after training; Determine whether the performance index of the hyperspectral inversion chlorophyll model meets the requirements; When it is judged that the performance index of the hyperspectral inversion chlorophyll model meets the requirement, the hyperspectral inversion chlorophyll model is saved.

3. The estimation method according to claim 2, characterized in that: The hyperspectral image data is divided into a training set and a validation set, including: Sorting the hyperspectral image data from low to high according to chlorophyll content; Divide every four of the hyperspectral image data into one group; The third hyperspectral image data in each group is used as a validation set, and the rest are used as training sets.

4. A system for estimating Dendrobium officinale polysaccharides using the method for estimating Dendrobium officinale polysaccharides based on hyperspectral imaging as described in any one of claims 1 to 3, characterized in that: The system comprises: An image acquisition module, used to collect hyperspectral image data of Dendrobium leaves; The processor is connected to the image acquisition module for obtaining the current content of dendrobium polysaccharides according to the hyperspectral image data.

5. The system according to claim 4, characterized in that The processor is configured to execute: Obtaining the growth stage of Dendrobium and the growth model of Dendrobium corresponding to the chlorophyll content; Get the current growth stage of Dendrobium officinale; Determining the corresponding chlorophyll content according to the current growth stage of the dendrobium and the growth model of the dendrobium; The chlorophyll content is input into a polysaccharide content estimation model to obtain the dendrobium polysaccharide content.

6. The system according to claim 5, characterized in that The growth model of Dendrobium officinale to obtain the growth stage of Dendrobium officinale and the corresponding chlorophyll content includes: Acquiring hyperspectral image data of the dendrobium leaves; Construct a hyperspectral inversion chlorophyll model; Inputting the hyperspectral image data into the hyperspectral inversion chlorophyll model to obtain the chlorophyll content of the dendrobium leaves; A dendrobium growth model is constructed according to the chlorophyll content and the dendrobium growth stage.

Images