A method and system for measuring and calculating the chlorophyll content of rice under cadmium stress

Abstract

The application discloses a kind of cadmium stress under rice chlorophyll content estimation method and system, method includes the following steps: collecting the spectral image of different cadmium concentration under different varieties different growth stages under rice leaf and measuring chlorophyll content;Extract the spectral reflectance of rice leaf spectral image;Pretreatment is carried out;According to different varieties and different growth stages, spectral characteristic band screening is carried out respectively;According to proportion, it is divided into training set and test set;Chlorophyll content prediction model is established;On training set, the prediction model is trained respectively, and after training, the prediction model is tested using test set respectively;Performance verification is carried out during training and testing process, and the final prediction model is obtained;The spectral image of the rice leaf to be measured is input into the final prediction model to predict chlorophyll content.The present application is based on hyperspectral technology and machine learning to quantitatively measure the chlorophyll content of rice under Cd stress, to realize the non-destructive, rapid and accurate online estimation of chlorophyll content under Cd stress.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of chlorophyll content measurement, and particularly relates to a method and system for measuring chlorophyll content of rice under cadmium stress. BACKGROUND

[0002] Heavy metal cadmium (Cd) is one of the most toxic heavy metals, has strong migration, is toxic to the growth and development of crops, is easily absorbed and accumulated by plants, and then directly or indirectly affects human health through the food chain. In many crops, such as rapeseed, pea, rice, corn, barley, mung bean and wheat, if exposed to Cd environment for a short or long time, Cd toxicity will inhibit photosynthesis activity, mainly by inhibiting the biosynthesis of chlorophyll in plants; especially rice, which is considered to be a high-efficiency cadmium accumulation body compared with other crops; therefore, studying the chlorophyll content of rice leaves can provide a basis for the degree of cadmium stress of rice and provide information for future precision agriculture.

[0003] In the traditional method, farmers observe the color change of rice leaves by subjective experience to judge the harm degree of Cd to rice; however, this depends on manpower and experience accumulation, and the evaluation result is easily affected by subjective factors, and there is no strict standard; researchers mainly determine the chlorophyll content and other physiological and biochemical indexes of leaves and the content of heavy metals in soil by biochemical analysis to directly judge the harm degree of Cd to rice; however, these chemical analysis methods are time-consuming and laborious, and cannot meet the intelligent requirements of modern agricultural monitoring and production. Therefore, it is urgent to propose a rapid and accurate method for measuring the chlorophyll content of rice under cadmium stress. SUMMARY

[0004] The main purpose of the present application is to overcome the shortcomings and deficiencies of the prior art, provide a method and system for measuring chlorophyll content of rice under cadmium stress, and based on hyperspectral technology and machine learning, the chlorophyll content of rice under Cd stress is quantitatively measured, the non-destructive, rapid and accurate online estimation of chlorophyll content under Cd stress is realized, which provides support for intelligent judgment of the Cd stress degree of rice, and provides a new theoretical basis and technical means for rice growth monitoring, and has the advantages of high calculation efficiency and high accuracy.

[0005] The first purpose of the present application is to provide a method for measuring chlorophyll content of rice under cadmium stress, comprising the following steps:

[0006] Using a hyperspectral instrument to collect spectral images of rice leaves of different varieties and different growth periods under different cadmium concentrations and measure the chlorophyll content;

[0007] Extracting the spectral reflectance of the spectral image of the rice leaves by the mean value method;

[0008] Pretreat the spectral image in combination with spectral reflectance;

[0009] Screen the spectral characteristic bands of the pretreated spectral image according to different varieties and different growth periods respectively;

[0010] Divide the screened spectral characteristic bands and the corresponding chlorophyll content into a training set and a test set in proportion;

[0011] Establish a chlorophyll content prediction model based on a convolutional neural network, a partial least squares regression method and a support vector machine respectively;

[0012] Train the prediction model on the training set, and test the prediction model after training using the test set; performance verification is performed during the training and testing processes to obtain a final prediction model;

[0013] Input the spectral image of the rice leaf to be measured into the final prediction model to predict the chlorophyll content.

[0014] As a preferred technical solution, the spectral image of the rice leaf under different cadmium concentrations, different varieties and different growth periods is collected, and the chlorophyll content is measured, specifically:

[0015] Select several rice seedlings of different varieties under the same conditions;

[0016] Prepare culture solutions with different cadmium concentrations to cultivate rice seedlings of different varieties;

[0017] Use a hyperspectral instrument to collect spectral images of rice leaves of different varieties and different growth periods, and use hyperspectral technology to measure the corresponding chlorophyll content;

[0018] The number of spectral images collected from the rice leaves of different varieties and different growth periods is equal.

[0019] As a preferred technical solution, the hyperspectral instrument uses an ASD FieldSpec4 portable hyperspectral instrument, the spectral range is 350-2500nm, and the spectral resolution is 3nm;

[0020] The hyperspectral instrument is corrected using a calibration whiteboard before collecting spectral images.

[0021] As a preferred technical solution, the spectral reflectance is extracted using View Spec Pro software: import the spectral image into View Spec Pro software for preview, select multiple spectral curves in the spectral image for merging; and the spectral reflectance is extracted by taking the mean value of the merged data;

[0022] The pretreatment adopts a multivariate scatter correction algorithm MSC, a smoothing algorithm S-G and a first derivative 1-D.

[0023] As a preferred technical solution, the spectral feature band screening is respectively performed by using the uninformative variable elimination method and the competitive adaptive reweighted sampling method.

[0024] The pretreated spectral images are classified according to different varieties, the spectral feature band screening is performed on the spectral images classified according to the varieties by using the uninformative variable elimination method and the competitive adaptive reweighted sampling method respectively, and iteration is performed for 20 times to obtain UVE variety band data sets and CARS variety band data sets.

[0025] The pretreated spectral images are classified according to different growth periods, the spectral feature band screening is performed on the spectral images classified according to the growth periods by using the uninformative variable elimination method and the competitive adaptive reweighted sampling method respectively, and iteration is performed for 20 times to obtain UVE growth period band data sets and CARS growth period band data sets.

[0026] As a preferred technical solution, the UVE variety band data sets and the corresponding chlorophyll content are divided into a UVE variety training set and a UVE variety test set according to a proportion; the CARS variety band data sets and the corresponding chlorophyll content are divided into a CARS variety training set and a CARS variety test set according to a proportion; the UVE growth period band data sets and the corresponding chlorophyll content are divided into a UVE growth period training set and a UVE growth period test set according to a proportion; and the CARS growth period band data sets and the corresponding chlorophyll content are divided into a CARS growth period training set and a CARS growth period test set according to a proportion.

[0027] As a preferred technical solution, the chlorophyll content prediction model comprises a CNN model, a PLS model and a SVM model.

[0028] The CNN model is constructed based on a convolutional neural network and comprises a convolutional layer, a pooling layer and a fully connected layer; the convolutional layer has a size of 3*3 and is used for extracting data features; the pooling layer is used for greatly reducing the parameter magnitude and preventing model overfitting; and the fully connected layer is used for outputting a prediction result.

[0029] The PLS model is constructed based on a partial least squares method.

[0030] The SVM model is constructed based on a support vector machine, a hinge loss function is used to calculate an empirical risk, and a regularization term is added to a solution system to optimize a structural risk; and a Gaussian kernel function is used as the kernel function of the SVM model.

[0031] As a preferred technical scheme, the UVE variety training set, the CARS variety training set, the UVE growth period training set and the CARS growth period training set are normalized and input into a CNN model, a PLS model and a SVM model respectively for training.

[0032] The iteration number of the CNN model is 400, and the learning rate is 0.001; the regularization parameter of the SVM model is 1, and the tolerance is 10 -3 ; the number of reserved principal components of the PLS model is 2, and the tolerance is 10 -6 ;

[0033] After training, the UVE variety test set, the CARS variety test set, the UVE growth period test set and the CARS growth period test set are normalized and input into the CNN model, the PLS model and the SVM model respectively for testing.

[0034] As a preferred technical scheme, the performance verification is performed by using a determination coefficient and a root mean square error.

[0035] The determination coefficient represents the fitting degree of the predicted chlorophyll content and the measured chlorophyll content, and the closer the determination coefficient is to 1, the higher the accuracy of the prediction model is; the determination coefficient R 2 , and the formula is:

[0036]

[0037] On the contrary, the root mean square error reflects the deviation degree of the predicted chlorophyll content and the measured chlorophyll content, and the lower the root mean square error is, the closer to 0, the higher the accuracy of the prediction model is; the root mean square error RMSE, and the formula is:

[0038]

[0039] wherein, is the chlorophyll content of the i-th sample predicted by the prediction model, is the average value of the measured chlorophyll content; y i is the chlorophyll content measured by the i-th sample, and n is the sample number.

[0040] The second object of the application is to provide a system for measuring and calculating the chlorophyll content of rice under cadmium stress, which is applied to the method for measuring and calculating the chlorophyll content of rice under cadmium stress.

[0041] The spectral acquisition module is used for acquiring spectral images of leaves of different varieties of rice under different cadmium concentrations and measuring the chlorophyll content using a hyperspectral instrument.

[0042] The reflectance extraction module is used for extracting the spectral reflectance of the spectral image of the rice leaf through a mean value method;

[0043] The preprocessing module is used for preprocessing the spectral image in combination with the spectral reflectance;

[0044] The characteristic waveband screening module is used for screening the spectral characteristic wavebands of the preprocessed spectral image according to different varieties and different growth periods respectively;

[0045] The data division module is used for dividing the screened spectral characteristic wavebands and the corresponding chlorophyll content into a training set and a test set in proportion;

[0046] The model establishment module is used for establishing a chlorophyll content prediction model based on a convolutional neural network, a partial least squares regression method and a support vector machine respectively;

[0047] The model training module is used for training the prediction model on the training set, testing the prediction model after training using the test set, and obtaining a final prediction model through performance verification during the training and testing processes;

[0048] The content prediction module is used for inputting the spectral image of the to-be-measured rice leaf into the final prediction model to predict the chlorophyll content.

[0049] Compared with the prior art, the present application has the following advantages and beneficial effects:

[0050] 1. The present application quantitatively calculates the chlorophyll content of rice under Cd stress based on hyperspectral technology, realizes non-destructive, rapid and accurate online estimation of the chlorophyll content of rice under Cd stress, and provides support for intelligent judgment of the Cd stress degree of rice and provides a new theoretical basis and technical means for rice growth monitoring.

[0051] 2. According to the rich spectral information provided by the hyperspectral technology, the present application uses two methods of UVE and CARS for characteristic waveband screening, reduces the data volume, improves the operation speed and accuracy, and at the same time, constructs a prediction model of the chlorophyll content of rice through machine learning, uses the powerful feature extraction capability of machine learning to extract spectral information related to the chlorophyll content, and predicts the chlorophyll content of rice of different varieties and different growth periods, so as to realize automatic chlorophyll content calculation with high accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to make the technical solutions in the embodiments of the present application clearer, the accompanying drawings needed in the embodiment description will be briefly introduced. Obviously, the accompanying drawings in the following description only constitute some embodiments of the present application, and all other drawings obtained by those skilled in the art without any creative effort based on these drawings also belong to the protection scope of the present application.

[0053] Figure 1 The flow chart of the method for measuring and calculating the chlorophyll content of rice under cadmium stress in the embodiment of the present application.

[0054] Figure 2 The spectrum of the original spectral reflectance and the spectrum of the spectral reflectance after pretreatment in the embodiment of the present application.

[0055] Figure 3 The chlorophyll content graph of different varieties at different growth stages and the spectral characteristic band graph of the screened rice leaf under different Cd concentrations in the embodiment of the present application.

[0056] Figure 4 The chlorophyll content graph of different varieties at different growth stages and the spectral characteristic band graph of the screened rice leaf under different Cd concentrations in the embodiment of the present application.

[0057] Figure 5 The structure diagram of the system for measuring and calculating the chlorophyll content of rice under cadmium stress in the embodiment of the present application. DETAILED DESCRIPTION

[0058] In order to make the technical solutions in the embodiments of the present application clearer, the accompanying drawings needed in the embodiment description will be briefly introduced. Obviously, the accompanying drawings in the following description only constitute some embodiments of the present application, and all other drawings obtained by those skilled in the art without any creative effort based on these drawings also belong to the protection scope of the present application.

[0059] In the present application, the phrase "embodiment" means that the specific features, structures or characteristics described in combination with the embodiment can be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily mean the same embodiment, nor is it an independent or alternative embodiment to other embodiments. Those skilled in the art explicitly and implicitly understand that the embodiments described in the present application can be combined with other embodiments.

[0060] As shown in Figure 1 The embodiment provides a method for measuring and calculating the chlorophyll content of rice under cadmium stress, which comprises the following steps:

[0061] S1, using a hyperspectral instrument to collect spectral images of rice leaves of different varieties and different growth stages under different cadmium concentrations and measure the chlorophyll content;

[0062] Specifically, step S1 is:

[0063] Under the same conditions, several rice seedlings of different varieties; configure culture solution with different cadmium concentrations to cultivate rice seedlings of different varieties; use a hyperspectral instrument to collect spectral images of rice leaves of different varieties and different growth stages, and use hyperspectral technology to measure the corresponding chlorophyll content; wherein the number of spectral images of rice leaves of different varieties and different growth stages is equal.

[0064] Specifically, in this embodiment, high cadmium storage rice varieties (grain cadmium content 0.94 mg / kg, number 968) and low cadmium storage rice varieties (grain cadmium content 0.19 mg / kg, number 786) are selected as test materials and planted in Guangdong area; when the rice seedlings grow to the three-leaf-one-heart period, uniform seedlings are selected and transferred to the Kimura B nutrient solution water culture barrel containing different Cd concentration stress (0 mg / L, 0.01 mg / L, 0.05 mg / L, 0.25 mg / L, 1 mg / L, 2 mg / L, 4 mg / L, 5 mg / L), CdCl2·2.5H2O is used to provide, and HCl or NaOH is used to adjust the pH value to 5.6±0.1, the seedlings are fixed with sponge and planting basket, four rice plants per barrel, the water culture barrel volume is 5L, each concentration is set as three parallel random arrangements, during the culture, the rice seedlings grow in the greenhouse at 25-30℃ under natural light, the nutrient solution is replaced every 3 days; to verify the cadmium stress of important growth stages of rice, the flag leaves and the second leaves from the bottom are sampled at the tillering stage (breeding for 50 days) and the heading stage (breeding for 80 days), respectively, four plants per barrel as one sample, three parallel samples are collected for each concentration, and stored in a 4℃ refrigerator for laboratory measurement of chlorophyll content, dry matter and other physiological and biochemical indicators and spectral reflectance.

[0065] The experiment uses ASDFieldSpec4 portable hyperspectral instrument developed by ASD (analytical spectral devices) company of the United States, which measures the spectral range of 350-2500 nm, and the spectral resolution is 3 nm; the collected rice leaves are placed in the laboratory for measurement, and the calibration white board is used for correction during measurement; 160 spectral data are obtained, which are divided according to different growth stages, 67 data at the tillering stage and 93 data at the heading stage; according to different varieties, 96 data of high cadmium storage variety and 64 data of low cadmium storage variety.

[0066] S2, extracting the spectral reflectance of the spectral image of the rice leaf by the mean value method;

[0067] S3, pre-process the spectral image by combining spectral reflectivity;

[0068] Specifically, the spectral reflectivity of each rice leaf is obtained by taking the mean value method by using View Spec Pro software: first, import the spectral image into the View Spec Pro software for preview, select multiple spectral curves in the spectral image for merging; the spectral reflectivity is obtained by taking the mean value of the merged data. Due to the influence of hyperspectral collection instruments or environmental factors, the original spectrum of the rice leaf has problems such as scattering effect, random noise and system noise, which will weaken the spectral signal of the chlorophyll of the rice leaf, and is not conducive to the establishment of the regression model. Therefore, the present application combines three pretreatment algorithms of multi-source scattering correction algorithm MSC, smoothing algorithm S-G and first derivative 1-D to pretreat the original spectral data of the rice leaf:

[0069] In the embodiment, the View Spec Pro software selects 10 spectral curves from each spectral image for merging, and obtains the spectral reflectivity by taking the average value of the 10 spectral curves.

[0070] In order to eliminate the pseudo-spectrum or defective spectrum in the data matrix, the MSC algorithm is used to make each spectrum closer to some 'ideal' spectrum; in order to obtain the best estimate value of the spectral data points and effectively reduce the random noise of the average reflectance spectrum, the S-G is used to 'fit' each point within a certain width window of the single-point spectral data; in order to remove the baseline offset process and separate the overlapping spectral peaks, the 1-D is used to enhance the small amount of information in the spectrum, and the difference between two subsequent spectral data points is estimated. The comparison chart of the processing result and the original extracted spectral reflectivity is shown in Figure 2 .

[0071] S4, spectral feature band screening is performed on the pretreated spectral image according to different varieties and different growth periods;

[0072] Specifically, the present application respectively uses the uninformative variable elimination method UVE and the competitive adaptive reweighted sampling method CARS for spectral feature band screening; wherein, the pretreated spectral image is classified according to different varieties, the uninformative variable elimination method and the competitive adaptive reweighted sampling method are used for spectral feature band screening on the spectral image classified by varieties, and iteration is performed for 20 times to obtain the UVE variety band data set and the CARS variety band data set.

[0073] The pretreated spectral image is classified according to different growth periods, the uninformative variable elimination method and the competitive adaptive reweighted sampling method are used for spectral feature band screening on the spectral image classified by growth periods, and iteration is performed for 20 times to obtain the UVE growth period band data set and the CARS growth period band data set.

[0074] As shown in Figure 3 According to different growth periods: tillering stage and heading stage; 43 characteristic wavelengths are screened out by using the CARS method in the tillering stage, and 1016 characteristic wavelengths are screened out by using the UVE method; 33 characteristic wavelengths are screened out by using the CARS method in the heading stage, and 77 characteristic wavelengths are screened out by using the UVE method; according to different varieties: high cadmium storage varieties and low cadmium storage varieties; 42 characteristic wavelengths are screened out by using the CARS method for high cadmium storage varieties, and 367 characteristic wavelengths are screened out by using the UVE method; 28 characteristic wavelengths are screened out by using the CARS method for low cadmium storage varieties, and 25 characteristic wavelengths are screened out by using the UVE method; among them, the UVE method screens out the maximum number of characteristic wavelengths 1016 in the tillering stage; the UVE method screens out the minimum number of characteristic wavelengths 25 for low cadmium storage varieties; according to the screening results of the two methods, the fitting degree of the no-information variable removal method is obviously more complete than that of the competing adaptive reweighted sampling method, and the required information can be included to improve the modeling accuracy; the positions of the screened characteristic wavelengths are mainly concentrated near 550nm-700nm and the near-infrared short wave band.

[0075] S5, the screened spectral characteristic wave band and the corresponding chlorophyll content are divided into a training set and a test set in proportion;

[0076] The UVE variety wave band data set and the corresponding chlorophyll content are divided into a UVE variety training set and a UVE variety test set in proportion; the CARS variety wave band data set and the corresponding chlorophyll content are divided into a CARS variety training set and a CARS variety test set in proportion; the UVE growth period wave band data set and the corresponding chlorophyll content are divided into a UVE growth period training set and a UVE growth period test set in proportion; the CARS growth period wave band data set and the corresponding chlorophyll content are divided into a CARS growth period training set and a CARS growth period test set in proportion.

[0077] In this embodiment, the division is carried out in a proportion of 4:1.

[0078] S6, a chlorophyll content prediction model is respectively established based on a convolutional neural network, a partial least squares regression method and a support vector machine;

[0079] The present application utilizes a convolutional neural network (CNN), a support vector machine (SVM) and a partial least squares regression (PLS) to construct a prediction model between the spectral reflectivity of rice leaves and the chlorophyll content thereof, including a CNN model, a PLS model and an SVM model.

[0080] The CNN model is constructed based on a convolutional neural network and comprises a convolutional layer, a pooling layer and a fully connected layer. The convolutional layer has a size of 3*3 and is used to extract data features. The pooling layer is used to greatly reduce the parameter magnitude and prevent model overfitting. The fully connected layer is one layer and is used to output a prediction result.

[0081] The PLS model is constructed based on a partial least squares method. The partial least squares method is a mathematical optimization technique that finds the best function match for a set of data by minimizing the sum of squares of errors.

[0082] The SVM model is constructed based on a support vector machine. The hinge loss function is used to calculate the empirical risk, and a regularization term is added to the solution system to optimize the structural risk. The SVM model is a classifier with sparsity and robustness. The kernel function of the SVM model is a Gaussian kernel function.

[0083] S7. The prediction model is trained on the training set, and after training, the prediction model is tested using the test set. During the training set test, performance verification is performed to obtain the final prediction model.

[0084] Specifically, the UVE variety training set, the CARS variety training set, the UVE growth period training set and the CARS growth period training set are normalized and then input into the CNN model, the PLS model and the SVM model for training.

[0085] The number of iterations of the CNN model is 400, and the learning rate is 0.001. The regularization parameter of the SVM model is 1, and the tolerance is 10 -3 . The number of retained principal components of the PLS model is 2, and the tolerance is 10 -6 .

[0086] After training, the UVE variety test set, the CARS variety test set, the UVE growth period test set and the CARS growth period test set are normalized and then input into the CNN model, the PLS model and the SVM model for testing.

[0087] When training and testing the prediction model, the coefficient of determination (R 2 ) and the root mean square error (RMSE) are used to verify the effect of the prediction model. The coefficient of determination indicates the fitting degree of the predicted chlorophyll content and the actually measured chlorophyll content. The closer the coefficient of determination is to 1, the higher the accuracy of the prediction model. The formula is:

[0088]

[0089] Conversely, the root mean square error (RMSE) reflects the degree of deviation between the predicted chlorophyll content and the actual measured chlorophyll content. The lower the RMSE, the closer it is to 0, the higher the accuracy of the prediction model. The formula is:

[0090]

[0091] in, To predict the chlorophyll content of the i-th sample using the prediction model, To measure the average chlorophyll content; y i Let n be the chlorophyll content measured for the i-th sample, and n be the number of samples.

[0092] In this embodiment, partial least squares regression (PLS), support vector machine (SVM), and convolutional neural network (CNN) models were combined with spectra selected by UVE and CARS feature wavelengths to analyze the chlorophyll content of rice leaves at different growth stages (tillering and heading stages) and different rice varieties (high and low cadmium storage). Analysis of the constructed prediction models and accuracy (see Table 1 below) shows that the spectra of rice leaves at the tillering stage have good fit to the chlorophyll content predictions of different feature band selection methods combined with different models, and also exhibit good accuracy (e.g., Figure 4 ), R 2 The lowest value was 0.5481; the performance of UVE combined with the three models in rice leaf spectroscopy at the heading stage was better than that of CARS combined with the three models; the model accuracy R on the test set was... 2 Specifically, the order is: UVE-SVM > UVE-PLS > UVE-CNN > CARS-PLS > CARS-CNN > CARS-SVM; UVE combined with the three models has the best effect in predicting chlorophyll content in high cadmium-storing varieties, with the model accuracy R on the test set being [value missing]. 2 They are 0.7528, 0.7072, and 0.7128 respectively (e.g., Figure 4 However, CARS, which combines three models, performed relatively poorly in predicting chlorophyll content in cadmium-rich varieties, with a model accuracy R0 on the test set. 2 The values ​​were 0.1794, 0.1846, and 0.2592, respectively. In summary, the three models established by screening the sensitive feature bands using the UVE method—Partial Least Squares Regression (PLS), Support Vector Machine (SVM), and Convolutional Neural Network (CNN)—showed the best accuracy and stability in predicting the chlorophyll content of leaves in high-cadmium-storage rice varieties. The tillering stage had a better predictive effect than the heading stage.

[0093] Table 1 Results of the spectral-based model for estimating chlorophyll content in rice leaves

[0094]

[0095]

[0096] S8, input the spectral image of the to-be-measured rice leaf into the final prediction model to perform chlorophyll content prediction.

[0097] Specifically, the final prediction model selects a suitable prediction model for prediction based on the variety and growth period of the to-be-measured rice leaf, to obtain the chlorophyll content.

[0098] The method for measuring and calculating the chlorophyll content of rice under cadmium stress provided by the application can not only quantitatively predict the chlorophyll content of rice under Cd stress, but also provide theoretical data support for effective judgment of the Cd stress degree. The determination of the Cd stress degree and the influence of Cd stress on physiological and biochemical indicators such as chlorophyll of plants in the traditional way often depends on empirical judgment and laboratory biochemical detection, which is time-consuming, labor-intensive and low in detection efficiency. In the application, the hyperspectral technology is combined with the machine learning method to be applied to the judgment of the Cd stress degree and the determination of the chlorophyll content. Different models are selected according to different varieties and different growth periods of rice to perform prediction, so as to realize the nondestructive online measurement and calculation of the chlorophyll content of rice under Cd stress, and the method has the advantages of high efficiency and high accuracy.

[0099] It should be noted that, for the foregoing method embodiments, in order to facilitate simple description, they are all described as a series of action combinations, but those skilled in the art should know that the application is not limited by the action sequence described, because according to the application, certain steps can be performed in other sequences or simultaneously.

[0100] Based on the same idea as the method for measuring and calculating the chlorophyll content of rice under cadmium stress in the above embodiment, the application further provides a system for measuring and calculating the chlorophyll content of rice under cadmium stress, which can be used to execute the method for measuring and calculating the chlorophyll content of rice under cadmium stress. In order to facilitate illustration, only the parts related to the embodiments of the application are shown in the structural schematic diagram of the system for measuring and calculating the chlorophyll content of rice under cadmium stress, and those skilled in the art can understand that the illustrated structure does not constitute a limitation on the device, and the device can include more or fewer components than the illustrated components, or some components can be combined, or different components can be arranged.

[0101] As shown in Figure 3 Another embodiment of the application provides a system for measuring and calculating the chlorophyll content of rice under cadmium stress, which includes a spectral acquisition module, a reflectivity extraction module, a preprocessing module, a feature waveband screening module, a data division module, a model establishment module, a model training module and a content prediction module.

[0102] The spectral acquisition module is used to acquire spectral images of leaf blades of different varieties of rice under different cadmium concentrations and measure the chlorophyll content using a hyperspectral instrument.

[0103] The reflectivity extraction module is configured to extract the spectral reflectivity of the spectral image of the rice leaf by a mean value method;

[0104] The preprocessing module is configured to preprocess the spectral image in combination with the spectral reflectivity;

[0105] The characteristic waveband screening module is configured to screen the spectral characteristic waveband of the preprocessed spectral image according to different varieties and different growth periods respectively;

[0106] The data division module is configured to divide the screened spectral characteristic waveband and the corresponding chlorophyll content into a training set and a test set in proportion;

[0107] The model establishment module is configured to establish a chlorophyll content prediction model based on a convolutional neural network, a partial least squares regression method and a support vector machine respectively;

[0108] The model training module is configured to train the prediction model on the training set, and test the prediction model after training using the test set; performance verification is performed during the training and testing processes, and a final prediction model is obtained;

[0109] The content prediction module is configured to input the spectral image of the to-be-measured rice leaf into the final prediction model to predict the chlorophyll content.

[0110] It should be noted that the one kind of cadmium stress under rice chlorophyll content measuring system of the application corresponds to the one kind of cadmium stress under rice chlorophyll content measuring method of the application, the technical features and advantages described in the embodiment of the one kind of cadmium stress under rice chlorophyll content measuring method are applicable to the embodiment of the one kind of cadmium stress under rice chlorophyll content measuring system, and the specific content can be referred to the description in the method embodiment of the application, which will not be described here again, and hereby declared.

[0111] In addition, in the embodiment of the one kind of cadmium stress under rice chlorophyll content measuring system of the above embodiment, the logical division of each program module is only an example, and in actual application, the above functions can be completed by different program modules according to the needs, for example, the configuration requirements of the corresponding hardware or the convenience of software implementation, that is, the internal structure of the one kind of cadmium stress under rice chlorophyll content measuring system is divided into different program modules to complete all or part of the functions described above.

[0112] The technical features of the above embodiments can be combined arbitrarily, and in order to make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combination of the technical features does not exist contradictory, it should be considered as the scope of the present application.

[0113] The above embodiments are the preferred embodiments of the present application, but the embodiments of the present application are not limited to the above embodiments, and any changes, modifications, substitutions, combinations, simplifications, etc. made without departing from the spirit and principles of the present application should be equivalent replacement manners and should be included in the protection scope of the present application.

Claims

1. A method for estimating the chlorophyll content of rice under cadmium stress, characterized by, The method comprises the following steps: Collecting spectral images of rice leaves of different varieties and different growth periods under different cadmium concentrations and measuring the chlorophyll content; Extracting the spectral reflectance of the spectral images of the rice leaves by the mean value method; Pretreating the spectral images in combination with the spectral reflectance; Screening the spectral characteristic bands of the pretreated spectral images according to different varieties and different growth periods; Dividing the screened spectral characteristic bands and the corresponding chlorophyll content into a training set and a test set in proportion; Establishing chlorophyll content prediction models based on a convolutional neural network, a partial least squares regression method and a support vector machine respectively; Training the prediction models on the training set and testing the prediction models on the test set after the training; Verifying the performance during the training and the testing to obtain a final prediction model; Inputting the spectral images of the rice leaves to be measured into the final prediction model to predict the chlorophyll content; The spectral characteristic band screening is performed by the uninformative variable elimination method and the competitive adaptive reweighted sampling method respectively; Classifying the pretreated spectral images according to different varieties, screening the spectral characteristic bands of the spectral images classified according to the varieties by the uninformative variable elimination method and the competitive adaptive reweighted sampling method respectively, and iterating 20 times to obtain UVE variety band data sets and CARS variety band data sets; Classifying the pretreated spectral images according to different growth periods, screening the spectral characteristic bands of the spectral images classified according to the growth periods by the uninformative variable elimination method and the competitive adaptive reweighted sampling method respectively, and iterating 20 times to obtain UVE growth period band data sets and CARS growth period band data sets.

2. The method for measuring and calculating the content of chlorophyll in rice under cadmium stress according to claim 1, characterized in that, The method for collecting spectral images of rice leaves of different varieties and different growth periods under different cadmium concentrations and measuring the chlorophyll content comprises the following steps: Selecting rice seedlings of different varieties under the same conditions; Configuring culture solutions with different cadmium concentrations to cultivate the rice seedlings of different varieties; Collecting spectral images of the rice leaves of different varieties and different growth periods by using a hyperspectral instrument and measuring the corresponding chlorophyll content by using hyperspectral technology; The number of the spectral images collected from the rice leaves of different varieties and different growth periods is equal.

3. The method according to claim 2, wherein, The hyperspectral instrument is an ASD FieldSpec4 portable hyperspectral instrument, the spectral range is 350-2500nm, and the spectral resolution is 3nm; The hyperspectral instrument is calibrated by using a calibration whiteboard before collecting spectral images.

4. The method according to claim 1, wherein, The spectral reflectance is extracted by using View Spec Pro software: importing the spectral images into the View Spec Pro software for preview, selecting a plurality of spectral curves in the spectral images for merging, and extracting the spectral reflectance by taking the mean value of the merged data; The pretreatment is performed by using a multivariate scatter correction algorithm MSC, a smoothing algorithm S-G and a first derivative 1-D.

5. The method for measuring the content of chlorophyll in rice under cadmium stress according to claim 1, characterized in that, Dividing the UVE variety band data sets and the corresponding chlorophyll content into a UVE variety training set and a UVE variety test set in proportion; The CARS variety waveband dataset and the corresponding chlorophyll content are divided into a CARS variety training set and a CARS variety test set according to a proportion; the UVE growth period waveband dataset and the corresponding chlorophyll content are divided into a UVE growth period training set and a UVE growth period test set according to a proportion; The CARS growth period waveband dataset and the corresponding chlorophyll content are divided into a CARS growth period training set and a CARS growth period test set according to a proportion.

6. The method according to claim 5, wherein, The chlorophyll content prediction model comprises a CNN model, a PLS model and a SVM model; The CNN model is constructed based on a convolutional neural network and comprises a convolutional layer, a pooling layer and a fully connected layer; the convolutional layer has a size of 3*3 and is used for extracting data features; the pooling layer is used for greatly reducing the parameter order of magnitude and preventing model overfitting; and the fully connected layer is used for outputting a prediction result; The PLS model is constructed based on a partial least squares method; The SVM model is constructed based on a support vector machine, uses a hinge loss function to calculate an empirical risk, and adds a regularization term to a solution system to optimize a structural risk; and a Gaussian kernel function is used as the kernel function of the SVM model.

7. The method according to claim 6, wherein, After normalization, the UVE variety training set, the CARS variety training set, the UVE growth period training set and the CARS growth period training set are input into the CNN model, the PLS model and the SVM model for training; Wherein, the iteration number of the CNN model is 400, the learning rate is 0.001; the regularization parameter of the SVM model is 1, and the tolerance is 10 -3 ; the number of reserved principal components of the PLS model is 2, and the tolerance is 10 -6 ; After training, the UVE variety test set, the CARS variety test set, the UVE growth period test set and the CARS growth period test set are normalized and input into the CNN model, the PLS model and the SVM model for testing.

8. The method according to claim 7, wherein the method is characterized by, The performance verification is performed by using a determination coefficient and a root mean square error; Wherein the determination coefficient represents the fitting degree of the predicted chlorophyll content and the measured chlorophyll content, and the closer the determination coefficient is to 1, the higher the accuracy of the prediction model is; the determination coefficient R 2 , the formula is: ； On the contrary, the root mean square error reflects the deviation of the predicted chlorophyll content from the measured chlorophyll content; the lower the root mean square error, the closer to 0, and the higher the accuracy of the prediction model; the root mean square error RMSE is calculated according to the following formula: ， wherein, is the predicted chlorophyll content of the i-th sample by the prediction model, i is the average of the measured chlorophyll contents of the samples, is the average of the measured chlorophyll contents of the samples, y i is the measured chlorophyll content of the i-th sample, i is the measured chlorophyll content of the i-th sample, n is the number of samples.

9. A system for measuring chlorophyll content in rice under cadmium stress, characterized in that, The system is applied to the method for measuring and calculating the chlorophyll content of rice under cadmium stress according to any one of claims 1-8, and the system comprises a spectrum acquisition module, a reflectivity extraction module, a preprocessing module, a characteristic waveband screening module, a data division module, a model establishment module, a model training module and a content prediction module; The spectrum acquisition module is used for acquiring spectral images of leaves of different varieties of rice under different cadmium concentrations and measuring the chlorophyll content of the leaves by using a hyperspectral instrument; The reflectivity extraction module is used for extracting the spectral reflectivity of the spectral images of the leaves of rice by using a mean value method; The preprocessing module is used for preprocessing the spectral images in combination with the spectral reflectivity; The characteristic waveband screening module is used for screening spectral characteristic wavebands of the preprocessed spectral images according to different varieties and different growth periods; The data division module is used for dividing the screened spectral characteristic wavebands and the corresponding chlorophyll content into a training set and a test set according to a proportion; The model establishment module is used for establishing chlorophyll content prediction models based on a convolutional neural network, a partial least squares regression method and a support vector machine, respectively. The model training module is configured to train the prediction model on a training set, and test the prediction model after training using a test set; Performance verification is performed during the training and testing processes to obtain a final prediction model; The content prediction module is configured to input a spectrum image of a to-be-tested rice leaf into the final prediction model to predict the chlorophyll content.

Images