Sunlight-induced chlorophyll fluorescence multispectral imaging method and imaging system

Through the multi-spectral imaging method of sunlight-induced chlorophyll fluorescence, combined with plant image distribution data and growth spectrum data, and using classification model and dual-flow network model, the problem that traditional spectrometers cannot intuitively reflect plant deformation and chlorophyll changes is solved, and the plant growth trend is achieved Efficient monitoring.

CN119091190APending Publication Date: 2024-12-06BEIJING YUTIANXING ELECTRONIC TECH CO LTD

Patent Information

Application Number
CN202411084010.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-08
Publication Date
2024-12-06

AI Technical Summary

Technical Problem

When traditional spectrometers acquire spectral data, due to the slit screening method, one-dimensional spatial resolution information loss cannot be intuitively reflected in changes such as deformation and development in plant production. It is difficult for the existing technology to effectively monitor the growth trend and changes in the chlorophyll content of plants.

Method used

The multi-spectral imaging method of sunlight-induced chlorophyll fluorescence is used to obtain plant image distribution data and image intensity data, calculate the growth spectrum data set, and use classification model and dual-stream network model for correlation analysis, and fuse spectral data and image data to form a visual spectrum Data image.

Benefits of technology

It realizes an intuitive feeling of changes in plant growth trends and chlorophyll content, improves the monitoring effect of plant growth trends, and can promptly detect problems during the growth process, such as early identification of wheat stripe rust.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119091190A_ABST
    Figure CN119091190A_ABST
Patent Text Reader

Abstract

The invention relates to a sunlight-induced chlorophyll fluorescence multispectral imaging method and imaging system. The method comprises the following steps: acquiring target plant data and background data under a target wavelength; calculating a plant growth spectrum data set through the plant image intensity data and the background image intensity data; performing classification model training by using an unmarked sample set in the plant image distribution data and a preset marked sample set to obtain a classification model; inputting unmarked samples in the unmarked sample data set into a classification model to obtain a classification result data set; inputting each unmarked sample in the to-be-analyzed category in each plant growth spectrum data set and the classification result data set into a mapping network for correlation analysis, and if the analysis result is correlation, endowing the plant image distribution data areas corresponding to the unmarked samples with the plant growth spectrum data; and endowing the plant growth spectral data on the plant image distribution data with different colors according to data intensity to obtain a visual spectral data image.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of agricultural remote sensing, and in particular to a sunlight-induced chlorophyll fluorescence multi-spectral imaging method and an imaging system. Background Art

[0002] Sunlight-induced chlorophyll fluorescence (SIF) is calculated based on the Fraunhofer fluorescence detection (FLD) method of atmospheric oxygen absorption dark lines. The principle is that the earth's atmosphere absorbs solar radiation in a specific band, forming O2-A and O2-B absorption dark lines near the 688 and 760nm bands, called Fraunhofer dark lines. Since the fluorescence spectrum radiation has a certain "filling" effect on the reflected radiation dark lines, the observed reflectivity increases. Assuming that the reflectivity and SIF values ​​inside and outside the absorption line are equal, the SIF value can be calculated using the difference between the solar downlink irradiance E at two wavelengths inside and outside the absorption line and the reflected radiance L of the vegetation canopy.

[0003] Traditional spectrometers mostly use the form of slit screening + grating splitting to expand the collected spectral information on the CCD, which can obtain spectral resolution of different wavelengths within a certain spectral range; however, in order to ensure sufficient spectral resolution, traditional spectrometers use slit selection, which causes one-dimensional spatial resolution information to be completely lost. Therefore, only spectral data can be obtained, and spectral data alone cannot intuitively reflect changes such as deformation development during plant production. The above problems need to be solved urgently. Summary of the invention

[0004] The invention discloses a sunlight-induced chlorophyll fluorescence multi-spectral imaging method and an imaging system, aiming to solve the technical problems existing in the prior art.

[0005] The present invention adopts the following technical solutions:

[0006] In one aspect, the present invention provides a method for sunlight-induced chlorophyll fluorescence multispectral imaging, comprising the following steps:

[0007] Acquire target plant data and background data at a target wavelength; wherein the target plant data includes plant image distribution data and plant image intensity data, and the background data includes background image intensity data;

[0008] Calculating a plant growth spectrum data set through the plant image intensity data and the background image intensity data at different target wavelengths;

[0009] A classification model is trained using an unlabeled sample set and a preset labeled sample set in the plant image distribution data to obtain a classification model; the labeled sample data set includes at least two labeled samples of different categories;

[0010] Inputting each unlabeled sample in the unlabeled sample data set into the classification model to obtain a classification result data set;

[0011] Inputting each of the plant growth spectrum data sets and each of the unlabeled samples of the categories to be analyzed in the classification result data sets into the mapping network for correlation analysis,

[0012] If the analysis result is related, then the plant growth spectrum data is assigned to the plant image distribution data area corresponding to the unlabeled sample;

[0013] The plant growth spectrum data on the plant image distribution data are assigned different colors according to data intensity to obtain a visualized spectrum data image.

[0014] In the sunlight-induced chlorophyll fluorescence multispectral imaging method of the present invention, the step of training the classification model using the unlabeled sample set and the preset labeled sample set in the plant image distribution data comprises:

[0015] A simple linear iterative clustering method is adopted to divide the plant image distribution data into a superpixel group including a plurality of superpixel blocks of different sizes and shapes, and the labeled samples are randomly assigned to the superpixel blocks, and at least one of the pixel blocks does not contain the labeled sample;

[0016] According to whether the super pixel block includes a labeled sample, dividing the plurality of super pixel blocks into a labeled group and an unlabeled group;

[0017] Propagate the category corresponding to the largest number of labeled samples in each superpixel block to the unlabeled samples in the same superpixel block to obtain a first propagation sample group;

[0018] Calculating the similarity between each unlabeled sample in the superpixel block in the unlabeled group and all the categories, and propagating the category corresponding to the maximum similarity value to each unlabeled sample to obtain a second propagation sample group;

[0019] The samples in the first propagation sample group and the second propagation sample group are classified according to categories to obtain a plurality of sample sets of different categories, and unlabeled samples labeled by propagation are selected in equal amounts from each of the sample sets to form an unlabeled sample set, and the classification model is trained using the unlabeled sample set and the labeled sample set.

[0020] In the sunlight-induced chlorophyll fluorescence multispectral imaging method of the present invention, the step of training the classification model using the unlabeled sample set and the labeled sample set comprises:

[0021] Constructing a two-stream network model, wherein the two-stream network model includes a supervised stream and an unsupervised stream;

[0022] By processing the labeled samples through the supervision flow, the supervised cross entropy loss function is calculated

[0023] Where H(·) is the cross entropy equation, Nx is the total number of labeled samples, and ω(·) corresponds to the unlabeled samples with different perturbations implemented by the supervision flow;

[0024] The unsupervised loss function is calculated by processing the unlabeled samples in the unlabeled sample set through the unsupervised flow.

[0025] Among them, H(·) is the cross entropy equation, Nu is the total unlabeled samples, and Ω(·) corresponds to the unlabeled samples with different perturbation versions implemented by the unsupervised flow;

[0026] The final loss function L = L X +L U The training endpoint is determined by minimization, and a two-stream network for classification is obtained as the classification model.

[0027] In the sunlight-induced chlorophyll fluorescence multispectral imaging method of the present invention, before the step of processing the unlabeled samples in the unlabeled sample set by the unsupervised flow, the method further includes:

[0028] Calculate the predicted probability of the unlabeled sample Select a threshold corresponding to the category of the unlabeled sample, compare the predicted probability with the threshold, and if the predicted probability is greater than the threshold, the unlabeled sample can be used to calculate the unsupervised loss parameter L U .

[0029] In the sunlight-induced chlorophyll fluorescence multispectral imaging method of the present invention, each of the unlabeled samples in the unlabeled sample set is processed by the unsupervised flow after being processed by consistency regularization.

[0030] In the sunlight-induced chlorophyll fluorescence multispectral imaging method of the present invention, the step of selecting a threshold value corresponding to the category of the unlabeled sample comprises:

[0031] Determine the number of categories corresponding to the unlabeled samples, and for each category η, count the classification probabilities σ in all unlabeled samples η Unlabeled sample data larger than the initial threshold §0;

[0032]

[0033] The unlabeled sample data is normalized to obtain normalized data

[0034] Constructing nonlinear weight mapping function

[0035] The threshold is calculated by the following formula:

[0036]

[0037] In the sunlight-induced chlorophyll fluorescence multispectral imaging method of the present invention, the nonlinear weight mapping function It is determined by quadratic parabola fitting.

[0038] in,

[0039] Here, δ is obtained by minimizing L U The value is determined, and δ∈(0,1].

[0040] In the sunlight-induced chlorophyll fluorescence multispectral imaging method of the present invention, the step of calculating and obtaining a plant growth spectrum data set by using the plant image intensity data and the background image intensity data at different target wavelengths comprises:

[0041] The plant image intensity data and the background image intensity data are binarized, normalized and averaged, and fluorescence intensity information is obtained based on the conversion relationship between the system-calibrated light power density and the actual intensity. The light energy utilization rate is obtained through the fluorescence intensity information, and the plant growth spectrum data is calculated based on the light energy utilization rate.

[0042] In the sunlight-induced chlorophyll fluorescence multispectral imaging method of the present invention, the plant growth spectrum data includes primary productivity, photosynthesis rate, chlorophyll quantum yield, chlorophyll fluorescence value, canopy total excited chlorophyll fluorescence value, and drought stress parameters.

[0043] In a second aspect, the present invention also provides an imaging system, comprising:

[0044] An acquisition module, used to acquire target plant data and background data at a target wavelength; wherein the target plant data includes plant image distribution data and plant image intensity data, and the background data includes background image intensity data;

[0045] A first calculation module, used for calculating a plant growth spectrum data set through the plant image intensity data and the background image intensity data at different target wavelengths;

[0046] A training module, used for training a classification model using an unlabeled sample set and a preset labeled sample set in the plant image distribution data to obtain a classification model; the labeled sample data set includes at least two labeled samples of different categories;

[0047] A classification module, used for inputting each unlabeled sample in the unlabeled sample data set into the classification model to obtain a classification result data set;

[0048] An analysis module is used to input each unlabeled sample of the category to be analyzed in each of the plant growth spectrum data sets and the classification result data sets into a mapping network for correlation analysis,

[0049] If the analysis result is related, then the plant growth spectrum data is assigned to the plant image distribution data area corresponding to the unlabeled sample;

[0050] The visualization module is used to assign different colors to the plant growth spectrum data on the plant image distribution data according to the data intensity to obtain a visualized spectrum data image. The technical solution adopted by the present invention can achieve the following beneficial effects:

[0051] The present invention mainly provides a sunlight-induced chlorophyll fluorescence multispectral imaging method, which is based on the fusion of spectral data of plant growth and plant image distribution data to form a visualized spectral data image. It can provide an intuitive feeling of the spectral data information of plant growth such as plant growth status and chlorophyll content changes, thereby improving the monitoring effect of plant growth status. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments, which constitute a part of the present invention. The exemplary embodiments of the present invention and their descriptions explain the present invention and do not constitute improper limitations on the present invention. In the drawings:

[0053] Figure 1 A flow chart of a sunlight-induced chlorophyll fluorescence multispectral imaging method of the present invention;

[0054] Figure 2 It is a schematic diagram of plant image distribution data segmentation of the present invention;

[0055] Figure 3A A plant image distribution data diagram in the imaging method of the present invention;

[0056] Figure 3B This is a visualized spectral data diagram obtained in the imaging method of the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the specific embodiments of the present invention and the corresponding drawings. In the description of the present invention, it should be noted that the term "or" is usually used in the sense of including "and / or", unless the content clearly indicates otherwise.

[0058] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or a magnetic connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal connection of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances. In addition, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance. In the description of the present invention, the meaning of "multiple" is at least two, such as two, three or more, etc., unless otherwise clearly and specifically limited.

[0059] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0060] In order to solve the problems existing in the prior art, the embodiments of the present application provide a sunlight-induced chlorophyll fluorescence multi-spectral imaging method and imaging system.

[0061] Example 1

[0062] This embodiment provides a sunlight-induced chlorophyll fluorescence multispectral imaging method, such as Figure 1 As shown, it includes the following steps:

[0063] Acquire target plant data and background data at a target wavelength; wherein the target plant data includes plant image distribution data and plant image intensity data, and the background data includes background image intensity data;

[0064] Calculate the plant growth spectrum data set through the plant image intensity data and the background image intensity data;

[0065] The classification model is trained with the unlabeled sample set and the preset labeled sample set in the plant image distribution data to obtain the classification model; the labeled sample data set contains at least two different categories of labeled samples, wherein the unlabeled sample extraction of the plant image distribution data is carried out in the following manner: 1. Image acquisition: collecting the data set containing the plant image; 2. Image preprocessing: grayscale conversion, filtering, edge detection; 3. Feature extraction: shape features, color features (color histogram), etc.; 4. Target detection: such as Haar feature classifier, HOUGH transform, SVM, YOLO, SSD, R-CNN recognition 5. Image segmentation: extract the target plant from the background. 6. Data enhancement: enhance the diversity of the data set by rotation, scaling, cropping and other methods, or adopt the relevant sample extraction method in the prior art. Input each unlabeled sample in the unlabeled sample data set into the classification model to obtain the classification result data set. Input each unlabeled sample of the category to be analyzed (i.e., the parameter type to be analyzed, such as disease, parameters reflecting plant growth conditions, etc., i.e., the category corresponding to the labeled sample) in each plant growth spectrum data set and the classification result data set into the mapping network for correlation analysis.

[0066] If the analysis result is relevant, the plant growth spectral data is assigned to the plant image distribution data area corresponding to the unlabeled sample; specifically, the mapping network model can refer to existing literature, such as "Classification of Hyperspectral Images Based on Multiclass Spatial-Spectral Generative Adversarial Networks", which assigns different colors to the plant growth spectral data on the plant image distribution data according to the data intensity (i.e., the numerical size) to obtain a visual spectral data image.

[0067] The sunlight-induced chlorophyll fluorescence multispectral imaging method of the present invention is based on fusing the spectral data of plant growth with the plant image distribution data to form a visualized spectral data image, which can provide an intuitive understanding of the spectral data information of plant growth such as the growth status of plants and changes in chlorophyll content. Figure 3A / 3B shows that the monitoring effect of plant growth status is improved, and problems that occur during plant growth can be discovered in time. For example, in the early prevention and control of wheat stripe rust, the plant image distribution data itself can determine the specific leaf part where the stripe rust breaks out; however, early stripe rust is difficult to identify only through image data; spectral data information can analyze the spectral differences between different disease levels and diseased areas and healthy areas. Changes in the reflectivity of spectra in specific bands, such as the near-infrared region, are related to the severity of wheat stripe rust and can be used for early diagnosis and prevention. Early stripe rust cannot be seen only through image data, but the fused data can mark the location of early stripe rust on the plant.

[0068] In some preferred embodiments, the step of training the classification model using the unlabeled sample set and the preset labeled sample set in the plant image distribution data includes:

[0069] The simple linear iterative clustering method (SLIC algorithm) is used to segment the plant image distribution data into superpixel groups containing multiple superpixel blocks of different sizes and shapes, and the labeled samples are randomly assigned to the superpixel blocks, and at least one pixel block does not contain a labeled sample; Figure 2 As shown in the figure, corn plants are divided into four regions, γ1, γ2, γ3, and γ4, by the SLIC algorithm. The green plants represent labeled samples, and the gray plants represent unlabeled samples. The γ2, γ3, and γ4 regions all contain labeled samples, and the segmented region of γ1 does not contain any samples; a1, a2, a3, a4, and a5 represent labeled samples of category a, and b1, b2, and b3 represent labeled samples of category b.

[0070] Specifically, the simple linear iterative clustering method is used to segment the plant image distribution data into superpixel groups including multiple superpixel blocks of different sizes and shapes, including the following steps:

[0071] Initialize cluster center points: Assuming that the pixel resolution of the acquired target image is N, first initialize it to be divided into M rectangular areas of similar size. At this time, the number of pixels in each segmented area is approximately N / M, and the distance between the center points of two adjacent areas is The cluster center is determined by searching for the minimum value of the image grayscale in the 3×3 area within each segmented area.

[0072] Mark each pixel in the image and calculate the distance between each pixel (x, y) and all adjacent cluster centers (xc, yc) within a certain range (2S×2S) (Where dlab is the distance between the pixel point (x, y) and the cluster center (xc, yc) in the Lab color space, dxy is the distance between the pixel point (x, y) and the cluster center (xc, yc) in the Euclidean geometric space, and m is the color distance constant) Select Dmin with the smallest distance as the cluster center of the pixel point (x, y).

[0073] The calculation is repeated until the cluster center of each pixel does not change.

[0074] According to whether the superpixel blocks include labeled samples, the multiple superpixel blocks are divided into a labeled group and an unlabeled group;

[0075] Propagate the category corresponding to the largest number of labeled samples in each superpixel block to the unlabeled samples in the same superpixel block to obtain a first propagation sample group;

[0076] Calculate the similarity between each unlabeled sample in the superpixel block in the unlabeled group and all categories, and propagate the category corresponding to the maximum similarity to each unlabeled sample to obtain the second propagation sample group; specifically, for the superpixel segmentation area γ1 containing only unlabeled samples, use the superpixel-level nearest neighbor algorithm to assign labels, and calculate the similarity between each unlabeled supersample in γ1 and each labeled sample category (such as the average spectral similarity or feature vector matching between known sample categories), that is, the similarity α between each unlabeled sample in γ1 and the a-class label (a1, a2, a3, a4, a5) and the similarity β with the b-class label (b1, b2, b3). By comparing the size of α and β, it is determined whether the sample in area γ1 propagates the a-class label or the b-class label.

[0077] Each sample in the first propagation sample group and the second propagation sample group is classified according to category to obtain multiple sample sets of different categories, and unlabeled samples marked by propagation are equally selected from each sample set to form an unlabeled sample set, and the classification model is trained with the unlabeled sample set and the labeled sample set.

[0078] In some preferred embodiments, the step of training the classification model with the unlabeled sample set and the labeled sample set includes:

[0079] Construct a two-stream network model, which includes a supervised stream and an unsupervised stream;

[0080] By processing the labeled samples through the supervision flow, the supervised cross entropy loss function is calculated

[0081] Among them, H(·) is the cross entropy equation, Nx is the total number of labeled samples, ω(·) corresponds to the unlabeled samples with different perturbations implemented by the supervision flow, and x irepresents a labeled sample from a labeled set, where i is the index of the sample. In the loss function, x i Used to calculate supervised loss, i.e. cross entropy loss L_X; y i is with x i The associated label, that is, x i The true category. When calculating the supervision loss L_X, y i Used to compare with the class probability P(y i |ω(x i )) for comparison.

[0082] The unsupervised loss function is calculated by processing the unlabeled samples in the unlabeled sample set through the unsupervised flow

[0083] Where H(·) is the cross entropy equation, Nu is the total unlabeled samples, Ω(·) corresponds to the unlabeled samples with different perturbations implemented in the unsupervised flow; u n represents an unlabeled sample from an unlabeled set, where n is the index of the sample, N U is the total number of unlabeled samples. n Used to calculate unsupervised loss, that is, unsupervised cross entropy loss L_U;

[0084] The final loss function L = L X +L U The training endpoint is determined by minimization, and a two-stream network for classification is obtained as the classification model.

[0085] A two-stream network model is adopted. In the two-stream structure, one stream is a supervised stream and the other is an unsupervised stream. The two streams share the same encoder to represent features, but use different classifiers to produce different label estimates. In this way, the network can simultaneously use labeled and unlabeled samples to learn more discriminative feature representations, and obtain good training results based on fewer labeled samples.

[0086] In some preferred embodiments, the step of processing the unlabeled samples in the unlabeled sample set by the unsupervised flow further includes:

[0087] Calculate the predicted probability of unlabeled samples Select the threshold corresponding to the category of the unlabeled sample and compare the predicted probability with the threshold. If the predicted probability is greater than the threshold, the unlabeled sample can be used to calculate the unsupervised loss parameter L U .

[0088] In some preferred embodiments, each unlabeled sample in the unlabeled sample set is processed by the unsupervised stream after being processed by the consistency regularization. Based on the dual-stream structure with consistency regularization, for the same unlabeled sample, under the condition of adding different disturbance interference items, similar prediction results can be obtained, thereby improving the accuracy of the classification of unlabeled samples.

[0089] In some preferred embodiments, the step of selecting a threshold value corresponding to the category of the unlabeled sample includes:

[0090] Determine the number of categories corresponding to the unlabeled samples. For each category ∈, count the classification probability σ in all unlabeled samples. η Unlabeled sample data larger than the initial threshold §0;

[0091]

[0092] Normalize the unlabeled sample data to obtain normalized data

[0093] Constructing nonlinear weight mapping function

[0094] The threshold is calculated by the following formula:

[0095]

[0096] Designing an adaptive threshold classification method based on unlabeled samples of different categories can effectively alleviate the problems of overfitting of simple classification problems and excessive prediction bias of complex classification problems. For example, categories a and b should be screened using different thresholds §a and §b respectively to prevent underfitting or overfitting.

[0097] In some preferred embodiments, the nonlinear weight mapping function It is determined by quadratic parabola fitting.

[0098] in,

[0099] Here, δ is obtained by minimizing L U The value is determined, and δ∈(0,1].

[0100] In some preferred embodiments, the step of calculating and obtaining a plant growth spectrum data set by using plant image intensity data and background image intensity data comprises:

[0101] The plant image intensity data and the background image intensity data are binarized, normalized and averaged. The fluorescence intensity information is obtained based on the conversion relationship between the system-calibrated light power density and the actual intensity. The light energy utilization rate is obtained through the fluorescence intensity information, and the plant growth spectrum data is calculated based on the light energy utilization rate.

[0102] In some preferred embodiments, the plant growth spectral data include primary productivity, photosynthesis rate, chlorophyll quantum yield, chlorophyll fluorescence value, canopy total stimulated chlorophyll fluorescence value, and drought stress parameters.

[0103] Specifically, the calculation method is as follows:

[0104] Extract the spectral data of the target crop canopy; after binarization, normalization, and averaging of the acquired image intensity information, extract the fluorescence intensity information I from the acquired image information according to the conversion relationship between the system-calibrated light power density and the actual intensity. 686 ,I 740 (target plant), I 686bg and I 740bg (background data), the specific fluorescence intensity can be obtained by the following method:

[0105] The CCD camera and the target wavelength filter are used as the imaging system to obtain plant image information.

[0106] Select a filter with a wavelength of 686nm and take a photo at a distance of 8-10 meters from a calibrated light source to obtain the light intensity I of the obtained photo. 2(t) , change the light intensity I of the calibration light source 1(t) , obtain multiple sets of corresponding data pairs of calibration light source and photo light intensity, and obtain the calibration curve η(t)=I 2(t) / I 1(t) ;

[0107] The target plant is photographed with the same imaging system to obtain the light intensity I of the obtained photo. 3(t) , then the fluorescence intensity information (i.e., fluorescence intensity) I 686 =I 3(t) ×A / (K×η(t)), where A is the filter peak area, K is the shutter time, and η(t) is I 3(t) The corresponding η value.

[0108] The calculation of light energy utilization parameters can be measured based on the environment around the actual installation site. The specific calculation formula is as follows:

[0109] ε=ε0×T(t)×SAg×W(t),

[0110] Among them, ε is the light energy utilization rate, ε0 is the maximum light energy utilization rate, T is the temperature, SAg is the limiting effect of site age on the maximum light energy utilization rate, W is the surface water index, and t is the time;

[0111] After the instrument is installed, the light energy utilization rate is calibrated by the fluorescence intensity ratio parameter:

[0112] ε=-a0*(I 740 / I 740bg )+a1*(I 686 / I 686bg )+a2;

[0113] Among them, ε is the light energy utilization rate;

[0114] a0, a1, a2 - calibration coefficients of the relative ratio of light energy utilization rate to fluorescence intensity, which need to be calibrated according to the on-site installation environment conditions.

[0115] The process of obtaining the parameters a0, a1, and a2: Calculating the relative fluorescence intensity I R , using the observation light intensity I / background light intensity I obtained by the imaging system bg The light energy utilization ε and the different wavelengths λ acquired by the imaging system i Relative fluorescence intensity I Ri , the parameter a is determined by linear fitting i and b,

[0116]

[0117] ε is the light energy utilization rate, a unitless proportional coefficient, i is the linear fitting proportional coefficient, b is the linear fitting constant parameter;

[0118] Based on the light energy utilization rate, the primary productivity, photosynthesis rate, chlorophyll quantum yield, chlorophyll fluorescence, and total canopy excited chlorophyll fluorescence value of corn plants were further analyzed.

[0119] a) Primary productivity = light energy utilization rate ε◇photosynthetically active radiation PAR (λ),

[0120] b) Leaf photosynthesis rate Pn = light energy utilization rate ε◇ photosynthetically active radiation PAR (λ) (umol◇m -2 ◇s -1 ), quantum yield

[0121] Among them, Ne is the total number of electrons transferred by the reaction, Np is the number of incident photons, and v is the reaction rate (mol◇s -1 ), NA Avogadro constant 6.02*10^23mol -1, K is the number of electrons transferred in the reaction, h is Planck's constant 6.62*10^-34J*s, c is the speed of light 3◇10^8m / s, I is the light power density (W◇m -2 ), A is the incident light area (m^2), λ is the wavelength nm;

[0122] c) Chlorophyll quantum yield AQY YLS =AQY◇η(t),

[0123] Total Excited Chlorophyll Fluorescence (SIF) tot (λ)=PAR(λ)◇f APAR ◇(AQY YLS / I R ),

[0124] Where PAR is the photosynthetically active radiation at the top of the canopy, f APAR is the vegetation photosynthetically active radiation absorption coefficient;

[0125] d) Chlorophyll fluorescence value

[0126] Where ρ is the canopy reflectance, K is the SIF canopy extinction coefficient, θ is the solar zenith angle, and LAI is the leaf area index (m 2 m -2 ), Ω is the aggregation index representing the degree to which the real spatial distribution of vegetation canopy leaves deviates from the random distribution, and its range is 0-1;

[0127] e) Total stimulated chlorophyll fluorescence value (SIF) of the canopy total (λ)=SIF(λ)*S ccd / f 2 , where S ccd is the pixel area, f is the focal length of the lens;

[0128] f) Drought stress: Under severe drought stress, I 686 / I 740 The relative ratio increased

[0129] Example 2

[0130] This embodiment provides an imaging system, which includes:

[0131] An acquisition module, used to acquire target plant data and background data under a target wavelength; wherein the target plant data includes plant image distribution data and plant image intensity data, and the background data includes background image intensity data;

[0132] A first calculation module, used for calculating a plant growth spectrum data set through plant image intensity data and background image intensity data;

[0133] A training module is used to train a classification model using an unlabeled sample set and a preset labeled sample set in the plant image distribution data to obtain a classification model; the labeled sample data set includes at least two labeled samples of different categories;

[0134] A classification module is used to input each unlabeled sample in the unlabeled sample data set into the classification model to obtain a classification result data set;

[0135] The analysis module is used to input each unlabeled sample of the category to be analyzed in each plant growth spectrum data set and the classification result data set into the mapping network for correlation analysis.

[0136] If the analysis result is related, then the plant growth spectrum data is assigned to the plant image distribution data area corresponding to the unlabeled sample;

[0137] The visualization module is used to assign different colors to the plant growth spectrum data on the plant image distribution data according to the data intensity to obtain a visualized spectrum data image.

[0138] The imaging system of the present invention forms a visualized spectral data image based on the fusion of the spectral data of plant growth and the plant image distribution data. It can provide an intuitive perception of the spectral data information of plant growth such as the growth status of plants and changes in chlorophyll content, thereby improving the monitoring effect of the growth status of plants.

[0139] The embodiments of the present invention are described above in conjunction with the accompanying drawings, but the present invention is not limited to the above-mentioned specific implementation modes, which are merely illustrative rather than restrictive. Under the guidance of the present invention, ordinary technicians in this field can also make many forms without departing from the scope of protection of the present invention and the claims, all of which are within the protection of the present invention.

Claims

1. A method for sunlight-induced chlorophyll fluorescence multispectral imaging, It is characterized in that The steps include: Acquire target plant data and background data at a target wavelength; wherein the target plant data includes plant image distribution data and plant image intensity data, and the background data includes background image intensity data; Calculating a plant growth spectrum data set through the plant image intensity data and the background image intensity data at different target wavelengths; A classification model is trained using an unlabeled sample set and a preset labeled sample set in the plant image distribution data to obtain a classification model; the labeled sample data set includes at least two labeled samples of different categories; Inputting each unlabeled sample in the unlabeled sample data set into the classification model to obtain a classification result data set; Inputting each of the plant growth spectrum data sets and each unlabeled sample of the category to be analyzed in the classification result data set into the mapping network for correlation analysis, If the analysis result is related, then the plant growth spectrum data is assigned to the plant image distribution data area corresponding to the unlabeled sample; The plant growth spectrum data on the plant image distribution data are assigned different colors according to data intensity to obtain a visualized spectrum data image.

2. The sunlight-induced chlorophyll fluorescence multispectral imaging method according to claim 1, It is characterized in that The step of training the classification model using the unlabeled sample set and the preset labeled sample set in the plant image distribution data includes: A simple linear iterative clustering method is adopted to divide the plant image distribution data into a superpixel group including a plurality of superpixel blocks of different sizes and shapes, and the labeled samples are randomly assigned to the superpixel blocks, and at least one of the pixel blocks does not contain the labeled sample; According to whether the super pixel block includes a labeled sample, dividing the plurality of super pixel blocks into a labeled group and an unlabeled group; Propagate the category corresponding to the largest number of labeled samples in each superpixel block to the unlabeled samples in the same superpixel block to obtain a first propagation sample group; Calculating the similarity between each unlabeled sample in the superpixel block in the unlabeled group and all the categories, and propagating the category corresponding to the maximum similarity value to each unlabeled sample to obtain a second propagation sample group; The samples in the first propagation sample group and the second propagation sample group are classified according to categories to obtain a plurality of sample sets of different categories, and unlabeled samples labeled by propagation are selected in equal amounts from each of the sample sets to form an unlabeled sample set, and the classification model is trained using the unlabeled sample set and the labeled sample set.

3. The sunlight-induced chlorophyll fluorescence multispectral imaging method according to claim 2, It is characterized in that The step of training the classification model using the unlabeled sample set and the labeled sample set comprises: Constructing a two-stream network model, wherein the two-stream network model includes a supervised stream and an unsupervised stream; By processing the labeled samples through the supervision flow, the supervised cross entropy loss function is calculated Where H(·) is the cross entropy equation, Nx is the total number of labeled samples, and ω(·) corresponds to the unlabeled samples with different perturbations implemented by the supervision flow; The unsupervised loss function is calculated by processing the unlabeled samples in the unlabeled sample set through the unsupervised flow. Among them, H(·) is the cross entropy equation, Nu is the total unlabeled samples, and Ω(·) corresponds to the unlabeled samples with different perturbation versions implemented by the unsupervised flow; The final loss function L = L X +L U The training endpoint is determined by minimization, and a two-stream network for classification is obtained as the classification model.

4. The method for sunlight-induced chlorophyll fluorescence multispectral imaging according to claim 3, It is characterized in that Before the step of processing the unlabeled samples in the unlabeled sample set by the unsupervised flow, the step further includes: Calculate the predicted probability of the unlabeled sample Select a threshold corresponding to the category of the unlabeled sample, compare the predicted probability with the threshold, and if the predicted probability is greater than the threshold, the unlabeled sample can be used to calculate the unsupervised loss parameter L U .

5. The sunlight-induced chlorophyll fluorescence multispectral imaging method according to claim 3, It is characterized in that Each of the unlabeled samples in the unlabeled sample set is processed by the unsupervised flow after being processed by consistency regularization.

6. The sunlight-induced chlorophyll fluorescence multispectral imaging method according to claim 4, It is characterized in that The step of selecting a threshold value corresponding to the category of the unlabeled sample comprises: Determine the number of categories corresponding to the unlabeled samples, and for each category η, count the classification probabilities σ in all unlabeled samples η Greater than the initial threshold§ 0 Unlabeled sample data; The unlabeled sample data is normalized to obtain normalized data θ η =σ η / max(σ η ), θ η ∈[0,1]; Constructing nonlinear weight mapping function The threshold is calculated by the following formula:

7. The sunlight-induced chlorophyll fluorescence multispectral imaging method according to claim 6, It is characterized in that The nonlinear weight mapping function It is determined by quadratic parabola fitting. in, Here, δ is obtained by minimizing L U The value is determined, and δ∈(0,1].

8. The sunlight-induced chlorophyll fluorescence multispectral imaging method according to claim 1, It is characterized in that The step of calculating and obtaining a plant growth spectrum data set by using the plant image intensity data and the background image intensity data at different target wavelengths comprises: The plant image intensity data and the background image intensity data are binarized, normalized and averaged, and fluorescence intensity information is obtained based on the conversion relationship between the system-calibrated light power density and the actual intensity. The light energy utilization rate is obtained through the fluorescence intensity information, and the plant growth spectrum data is calculated based on the light energy utilization rate.

9. The sunlight-induced chlorophyll fluorescence multispectral imaging method according to claim 1, It is characterized in that The plant growth spectrum data include primary productivity, photosynthesis rate, chlorophyll quantum yield, chlorophyll fluorescence value, canopy total excited chlorophyll fluorescence value, and drought stress parameters.

10. An imaging system, It is characterized in that include: An acquisition module, used for acquiring target plant data and background data at a target wavelength; The target plant data includes plant image distribution data and plant image intensity data, and the background data includes background image intensity data; A first calculation module, used for calculating a plant growth spectrum data set through the plant image intensity data and the background image intensity data at different target wavelengths; A training module, used for training a classification model using an unlabeled sample set and a preset labeled sample set in the plant image distribution data to obtain a classification model; the labeled sample data set includes at least two labeled samples of different categories; A classification module, used for inputting each unlabeled sample in the unlabeled sample data set into the classification model to obtain a classification result data set; An analysis module is used to input each unlabeled sample of the category to be analyzed in each of the plant growth spectrum data sets and the classification result data sets into a mapping network for correlation analysis, If the analysis result is related, then the plant growth spectrum data is assigned to the plant image distribution data area corresponding to the unlabeled sample; The visualization module is used to assign different colors to the plant growth spectrum data on the plant image distribution data according to data intensity to obtain a visualized spectrum data image.

Citation Information

Patent Citations

  • Plant physiology monitoring method based on chlorophyll fluorescence and multispectral image

    CN105548113A

  • Plant drought stress diagnostic method and device based on chlorophyll fluorescence imaging technology

    CN106546567A

Cited By

  • Snapshot sunlight-induced chlorophyll fluorescence imaging method and system

    CN120404677A