Method for accurately evaluating plant leaf color by combining spectrum with RBFNN

Abstract

The invention discloses a method for accurately evaluating plant leaf color by combining a spectrum with RBFNN (Radial Basis Function Neural Network), relates to the technical field of plant leaf color evaluation, and solves the urgent problem of how to accurately evaluate the plant leaf color in the botany research field. According to the method, the evaluation model of the plant leaf color grade and the spectral data is established through the physiological test, the plant leaf color can be more rapidly, simply and accurately evaluated by using the method, and the plant leaf color classification grade is obtained. Spectrum is a sensitive tool for nondestructive determination of leaf color grade, and the method is applied to scientific research of plant leaf color phenotype survey, provides help for positioning leaf color gene by plant genome data, and greatly assists molecular plant breeding; meanwhile, monitoring is effective and accurate, errors caused by manual grading can be well avoided, accurate grading data are obtained, the accurate evaluation effect on the leaf color of the same plant is remarkable, the method can be effectively applied to scientific research of plant leaf color phenotype investigation, and scientificity and accuracy of test data are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of plant leaf color evaluation, and specifically to a method for accurately evaluating plant leaf color by combining spectrum with RBFNN. Background Art

[0002] In the field of botanical research, the ability to master the genome has far exceeded the ability to measure its impact on phenotype. In response to this unbalanced situation, more and more investment has been made in plant phenotypic research in recent years. With the rapid development of optical measurement methods, it has been applied to plant phenotypic trait analysis due to its advantages such as fast speed, non-destructiveness, and long-term tracking and monitoring. The radial basis function neural network (RBFNN) performs well in dealing with classification and regression problems. Its simple structure, strong generalization ability, and fast speed make it an effective neural network model. More and more studies have applied this technology to plant phenotypic research and application fields such as plant physiological and biochemical research, seed vitality assessment, photosynthesis and respiration, root phenotype, pest and disease and abiotic stress research, yield prediction, and agricultural product quality testing.

[0003] Among the methods for measuring and classifying plant leaf color phenotypes, traditional methods for measuring and classifying plant leaf color phenotypes mainly include observation method, leaf color diagnosis method and image analysis method;

[0004] 1. Observation method:

[0005] Observation is one of the earliest and most commonly used classification methods, which is based on the morphological characteristics of plants. By observing and comparing the visible characteristics of plants, such as plant height, stem thickness, leaf shape, flower color, etc., the phenotypic differences of plants can be determined. This method usually requires manual measurement and recording, which is difficult to meet the requirements of high-throughput processing and accuracy.

[0006] 2. Leaf color diagnosis method:

[0007] Leaf color diagnosis is a method of simulating the shades of plant leaf colors to make a series of color cards as a comparison standard for measuring leaf color, and comparing them with the leaf color of the plant to be tested to determine the nutritional status of the plant. There are certain human factors in using this method. The comparison and judgment of leaf color mainly depends on the visual judgment of the observer. Different observers may have inconsistent judgment results due to personal visual differences or different interpretations of color cards.

[0008] 3. Determination of chlorophyll content and SPAD value:

[0009] The health of plants is assessed by measuring the chlorophyll content in leaves. Commonly used methods include ultraviolet spectrophotometry, fluorescence analysis, in vivo chlorophyll meter method, photoacoustic spectroscopy and high performance liquid chromatography. These methods can measure the chlorophyll content in leaves non-destructively, thus reflecting the growth status of plants. The disadvantages of chlorophyll content determination mainly lie in the variability of the measurement results and susceptibility to multiple factors, as well as the limitations of the measurement methods.

[0010] SPAD value determination is a method for quickly measuring chlorophyll content based on the principle of photoelectric non-destructive testing. By measuring the absorption of light in the red light region (peak wavelength 650nm) and the near-infrared region (peak wavelength 940nm) by the leaves, the relative content of chlorophyll is calculated and expressed as a SPAD value. This method does not damage the leaves and can quickly obtain chlorophyll content information. This method is affected by the illumination of light radiation, which means that under different lighting conditions, the measured SPAD value may deviate.

[0011] In the field of botanical research, a pressing issue is how to achieve accurate assessment of plant leaf color. Current assessment methods are subject to inherent subjective judgment and technical limitations, making it difficult to capture the subtle and critical differences in leaf color. These traditional methods not only lack accuracy in color description, but also fail to reliably verify their results, and are even more difficult to effectively connect with increasingly sophisticated genomic data. Given that plant functional genomics is developing at an unprecedented pace, we urgently need an innovative quantitative evaluation system for leaf color phenotypes that can not only accurately distinguish changes in leaf color, but also ensure the rigor and reliability of experimental data, laying a solid foundation for scientific research. Summary of the invention

[0012] In view of the shortcomings of the prior art, the present invention provides a method for accurately evaluating plant leaf color by combining spectrum with RBFNN, which solves an urgent problem in the field of botanical research, that is, how to achieve accurate evaluation of plant leaf color.

[0013] To achieve the above objectives, the present invention is implemented through the following technical solutions: a method for accurately evaluating plant leaf color by combining spectrum with RBFNN, comprising:

[0014] Step 1: Collect standard sample spectral data and use SNV to preprocess the spectral data;

[0015] Step 2: Divide the standard samples into graded variables and construct the RBFNN discriminant model;

[0016] Step 3: After constructing the RBFNN discriminant model, perform model permutation test and verification;

[0017] Step 4: Collect the spectral data of the samples to be evaluated and import them into the model for grade determination; input the spectral data of the samples to be evaluated into the RBFNN model to obtain the leaf color grade of the plants to be evaluated.

[0018] As a further solution of the present invention: In step 1, the specific method of collecting the standard sample spectrum data is:

[0019] The leaf color was determined according to the Royal Horticultural Society (RHS) plant standard colorimetric card, and standard grade leaf samples were collected according to leaf color classification. After collecting the plant leaves, the surface dust and stains were wiped off and the leaves were flattened. The leaves were placed in the center of the scanning platform for spectral data collection, and the spectral reflectance in the 320nm to 1100nm band was collected.

[0020] As a further solution of the present invention: SNV preprocesses the spectral data in the following manner:

[0021] SNV preprocessing reduces the instrument and environmental noise in the spectral data, weakens and eliminates the effects of sample unevenness, baseline drift, high-frequency noise, and stray light on the spectral signal. It standardizes the spectral data of each sample so that the spectral data of different samples have the same mean and unit variance, eliminating the differences between samples.

[0022] As a further solution of the present invention: in step 2, the specific process of establishing the plant leaf color RBFNN discrimination model is as follows:

[0023] According to the characteristics of plant leaf color depth, the plants are divided into leaf color grades, and the graded variable values ​​are determined for the samples; at this stage, the center of the hidden layer neurons needs to be determined; the training sample set is clustered using the k-means algorithm, and the obtained cluster center is used as the center of the radial basis function; after determining the center of the radial basis function, the next step is to calculate the neuron connection weights between the hidden layer and the output layer; this is done using the least squares method (OLS) or the gradient descent method; through the above two stages of training, RBFNN can learn an effective representation of the input data, and can perform prediction or classification tasks based on the training data, and establish an RBFNN discrimination model.

[0024] As a further solution of the present invention: in step three, it includes:

[0025] Validate the RBFNN model using a permutation test; collect data for training the RBFNN model and collect corresponding real data for validating the model; train the RBFNN model using the collected data to obtain the model's predicted output; calculate the difference between the model's predicted output and the actual output, randomly rearrange the labels of the actual data, and then recalculate the difference statistic using these permuted data; repeat this process multiple times to generate a null distribution; compare the difference statistic calculated from the original data with the null distribution obtained through the permutation operation to evaluate the performance of the model;

[0026] As a further solution of the present invention: After evaluating the performance of the model, observe R 2 The coefficient of determination of the numerical discriminant model is used to evaluate the accuracy of the model prediction.

[0027] As a further embodiment of the present invention: R 2 The specific numerical discrimination model is:

[0028] R 2 The neural network is an extension of the radial basis function neural network; unlike the traditional neural network, R 2 Neural networks are particularly suitable for processing data with nonlinear characteristics; the R value, also known as the coefficient of determination, is an indicator used in statistics to measure the goodness of fit of a model; when the R value approaches 1, it means that the model can fit the data perfectly without any error.

[0029] The present invention provides a method for accurately evaluating plant leaf color by combining spectrum with RBFNN. Compared with the prior art, it has the following beneficial effects:

[0030] (1) The present invention establishes an evaluation model of plant leaf color grade and spectral data through physiological experiments. This method can be used to more quickly and easily accurately evaluate plant leaf color and obtain plant leaf color classification grades. Spectra are sensitive tools for non-destructive determination of leaf color grades. This method is applied to scientific research on plant leaf color phenotype investigations, provides assistance for locating leaf color genes with plant genome data, and greatly assists molecular plant breeding;

[0031] (2) The present invention establishes a method for accurately evaluating plant leaf color based on spectral imaging technology combined with the RBFNN model, and uses spectral data to quickly and accurately evaluate the plant leaf color level. Compared with traditional measurement methods, the present invention can quickly and easily distinguish plant leaf colors accurately, ensuring the scientificity and accuracy of the test data.

[0032] (3) The monitoring method of the present invention is effective and accurate, and can well avoid the errors caused by manual grading, thereby obtaining accurate grading data. It has a significant effect on the precise evaluation of leaf color of the same plant species, and can be effectively applied to the investigation of plant leaf color phenotypes in scientific research to ensure the scientificity and accuracy of the test data. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The present invention will be further described below in conjunction with the accompanying drawings.

[0034] Figure 1 This is a technical implementation flow chart of a method for accurately evaluating plant leaf color by combining spectrum with RBFNN in the present invention;

[0035] Figure 2 This is an example of a method for accurately evaluating plant leaf color by combining spectroscopy with RBFNN in the present invention, and a model for accurately evaluating plant leaf color grade samples;

[0036] Figure 3 This is the original spectrum of the collected samples of the method for accurately evaluating plant leaf color by combining spectrum with RBFNN in the present invention;

[0037] Figure 4 This is a spectrum data diagram after SNV processing of a method for accurately evaluating plant leaf color by combining spectrum with RBFNN in the present invention;

[0038] Figure 5 This is a comparison chart of the actual value and predicted value results of a method for accurately evaluating plant leaf color by combining spectrum with RBFNN in the present invention;

[0039] Figure 6 This is a fitting diagram of the true value and predicted value results of a method for accurately evaluating plant leaf color by combining spectrum with RBFNN in the present invention. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the 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.

[0041] In the following embodiments, the present invention uses a 1100 portable surface spectrometer (PSR-1100 for short), the spectral range of PSR-1100 is 320-1100nm (ultraviolet-visible-near infrared), the PSR-1100 measurement system is composed of a spectrometer and an optical fiber, and is a passive spectrometer for measuring reflectivity, including a 512 linear array detector, a fixed holographic grating as a dispersion element, and a 25° optical fiber probe; in order to avoid external interference, the device is installed in a sealed space, and a halogen lamp (35W 230V) is installed at an appropriate position;

[0042] Embodiment 1

[0043] See also Figure 1-Figure 6 ,The present invention provides a method for accurately evaluating plant leaf color by combining spectrum with RBFNN, comprising:

[0044] Step 1: Collect standard sample spectral data;

[0045] Specifically, the leaf color is determined according to the Royal Garden Society (RHS) plant standard colorimetric card, and standard grade leaf samples are collected according to leaf color classification; after collecting plant leaves, the surface dust and stains are wiped off and flattened, and the leaves are placed in the center of the scanning platform for spectral data collection, and the spectral reflectance in the 320nm to 1100nm band is collected;

[0046] The original spectral data has a large amount of noise. In order to reduce the instrument and environmental noise in the data, weaken and eliminate the influence of sample unevenness, baseline drift, high-frequency noise, stray light, etc. on the spectral signal, the present invention uses SNV to preprocess the spectral data; SNV preprocessing is a commonly used spectral preprocessing method, which is mainly used to eliminate the multiplicative scattering effect and baseline drift in spectral data. SNV preprocessing standardizes the spectral data of each sample so that the spectral data of different samples have the same mean and unit variance, thereby eliminating the differences between samples and improving the accuracy of data analysis;

[0047] Step 2: Divide the standard samples into graded variables and construct the RBFNN discriminant model;

[0048] The specific process of establishing the plant leaf color RBFNN discrimination model is as follows:

[0049] According to the characteristics of plant leaf color depth, the plants are divided into leaf color grades, and the graded variable values ​​are determined for the samples; at this stage, the center of the hidden layer neurons needs to be determined; the training sample set is clustered using the k-means algorithm, and the obtained cluster center is used as the center of the radial basis function; after determining the center of the radial basis function, the next step is to calculate the neuron connection weights between the hidden layer and the output layer; this is done by the least squares method (OLS) or the gradient descent method; through the above two stages of training, RBFNN can learn the effective representation of the input data, and can perform prediction or classification tasks based on the training data, and establish an RBFNN discrimination model;

[0050] Step 3: After constructing the RBFNN discriminant model, perform model permutation test and verification;

[0051] Specifically, the RBFNN model is validated using a permutation test; data used to train the RBFNN model is collected, and corresponding actual data is collected for validation of the model; the RBFNN model is trained using the collected data to obtain the model's predicted output; the difference between the model's predicted output and the actual output is calculated, the labels of the actual data are randomly rearranged, and then the difference statistics are recalculated using these permuted data; this process is repeated multiple times to generate a null distribution; the difference statistics calculated from the original data are compared with the null distribution obtained through the permutation operation to evaluate the performance of the model; and then R is observed on this basis. 2 The coefficient of determination of the numerical discriminant model is used to evaluate the accuracy of the model prediction;

[0052] Step 4: Collect the spectral data of the samples to be evaluated and import them into the model for grade determination; input the spectral data of the samples to be evaluated into the RBFNN model to obtain the leaf color grade of the plants to be evaluated;

[0053] Embodiment 2

[0054] In the specific implementation process of this embodiment, based on the first embodiment, and different from the first embodiment, this embodiment specifically describes a method for accurately evaluating plant leaf color by combining spectroscopy with RBFNN by combining RBFNN:

[0055] See also Figure 1 :Leaf color is classified according to the colorimetric card, and the spectral reflectance of the standard sample is collected by the spectrometer according to the classification; then the spectral data is preprocessed by SNV; according to the leaf color grade of the standard sample, the classification variable value of the sample is determined; through the training of center selection and weight calculation, RBFNN can learn the effective representation of the input data, and can perform classification tasks according to the training data to establish the RBFNN discrimination model; then the RBFNN model permutation test is performed; finally, the spectral data of the sample to be evaluated is input, and the leaf color grade result of the plant sample to be evaluated is output;

[0056] It should be noted that when investigating plant leaf color phenotypes, leaf colors are often similar and the differences are small, which makes it impossible to quantify them manually and fail to meet the needs of accurate investigation; Figure 2 As shown, some Chinese cabbage samples from the 442F2 population have little difference in leaf color and the same chlorophyll content, but they belong to different leaf color categories when visually inspected;

[0057] The following is an example of accurately evaluating the leaf color of Chinese cabbage. Figure 2-6 The present invention is further described in detail with specific embodiments:

[0058] Step 1: If Figure 3 As shown, the leaf color classification of this test was determined according to the Royal Garden Society (RHS) plant standard colorimetric card, and 330 standard grade leaf samples were collected according to the leaf color classification, and the scores were recorded through visual inspection; the Chinese cabbage leaves were collected and flattened, and the dust and stains on the surface were wiped off, and the leaves were placed in the center of the scanning platform for spectral data collection, and the spectral reflectance in the 320nm to 1100nm band was collected;

[0059] The second step is, Figure 4 As shown in the figure, in order to reduce the instrument and environmental noise in the data, weaken and eliminate the influence of sample inhomogeneity, baseline drift, high-frequency noise, stray light, etc. on the spectral signal, it is necessary to preprocess the spectral data; SNV is used to preprocess the spectral data; SNV preprocessing standardizes the spectral data of each sample so that the spectral data of different samples have the same mean and unit variance, thereby eliminating the differences between samples and improving the accuracy of data analysis;

[0060] The third step is Figure 5-6 As shown in the figure, the establishment and permutation test process of the RBFNN discriminant model of Chinese cabbage leaf color: according to the classification results of the British Royal Garden Society RHS plant standard colorimetric card, Chinese cabbage is divided into four categories: light green, oily green, dark green, and dark green, and the classification variable values ​​1, 2, 3, and 4 are assigned to the samples; the training sample set is clustered using the K-means algorithm, and the obtained cluster center is used as the center of the radial basis function; the center of the radial basis function is determined, and then the neuron connection weights between the hidden layer and the output layer are calculated; this is performed by the least squares method (OLS) or the gradient descent method; through these two stages of training, RBFNN can learn the effective representation of the input data, and can perform prediction or classification tasks based on the training data, and establish the RBFNN discriminant model;

[0061] Next, we collect data for training the RBFNN model and collect corresponding real data for validating the model; we use the collected data to train the RBFNN model and get the model’s predicted output; we calculate the difference between the model’s predicted output and the actual output, randomly rearrange the labels of the actual data, and then recalculate the difference statistic using these permuted data; we repeat this process multiple times to generate a null distribution; we compare the difference statistic calculated from the original data with the null distribution obtained through the permutation operation to evaluate the performance of the model; and then we observe the R 2 The coefficient of determination of the numerical discriminant model is used to evaluate the accuracy of the model prediction;

[0062] like Figure 5-6 As shown, R 2 The neural network is an extension of the radial basis function (RBF) neural network. Unlike traditional neural networks, R 2 Neural networks are particularly suitable for processing data with nonlinear characteristics. The core idea is to use RBF as neurons in the hidden layer to map the input space to a new space so that the data can be linearly separated in this new space. The R value, also known as the coefficient of determination, is an indicator used in statistics to measure the goodness of fit of a model. When the R value approaches 1, it means that the model can fit the data perfectly without any error. For neural networks, this means that the performance of the network has reached an ideal level and can accurately learn and predict the relationship between input and output. The R values ​​of the training set and the test set obtained by the RBFNN model are as follows: 2 It is 0.99999, which means the data fit is good;

[0063] Step 4: Input the spectrum of the sample to be evaluated into the RBFNN model to obtain the classification of the Chinese cabbage leaf color to be evaluated, and conduct technical verification. The verification results are shown in Table 1 below, which are the classification results of 100 Chinese cabbage leaf samples to be evaluated based on the RBFNN model;

[0064] Table 1 Classification results of 100 Chinese cabbage leaf samples to be evaluated based on RBFNN model

[0065]

[0066] The established leaf sample RBFNN model has a good effect on the grade classification of 100 Chinese cabbage leaf samples to be evaluated; except for one sample category misjudgment, the recognition rate was 90.90%, and the other classifications reached 100% recognition rate; the overall recognition rate was 97.72%, indicating that the multi-spectral combined with RBFNN discriminant analysis method can effectively evaluate Chinese cabbage with different leaf colors, providing an effective method for rapid and accurate evaluation of different Chinese cabbage leaf colors;

[0067] Embodiment 3

[0068] The specific implementation process of this embodiment includes the entire implementation process of the above two groups of embodiments.

[0069] Some of the data in the above formulas are dimensionless and numerically calculated. Meanwhile, the contents not described in detail in this specification belong to the prior art known to those skilled in the art.

[0070] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A method for accurately evaluating plant leaf color by combining spectroscopy with RBFNN, characterized in that: include: Step 1: Collect standard sample spectral data and use SNV to preprocess the spectral data; Step 2: Divide the standard samples into graded variables and construct the RBFNN discriminant model; Step 3: After constructing the RBFNN discriminant model, perform model permutation test and verification; Step 4: Collect the spectral data of the samples to be evaluated and import them into the model for grade determination; input the spectral data of the samples to be evaluated into the RBFNN model to obtain the leaf color grade of the plants to be evaluated.

2. The method for accurately evaluating plant leaf color by combining spectrum with RBFNN according to claim 1, characterized in that: In step 1, the specific method of collecting standard sample spectrum data is: The leaf color was determined according to the Royal Horticultural Society (RHS) plant standard colorimetric card, and standard grade leaf samples were collected according to leaf color classification. After collecting the plant leaves, the surface dust and stains were wiped off and the leaves were flattened. The leaves were placed in the center of the scanning platform for spectral data collection, and the spectral reflectance in the 320nm to 1100nm band was collected.

3. The method for accurately evaluating plant leaf color by combining spectrum with RBFNN according to claim 2, characterized in that: SNV preprocesses spectral data as follows: SNV preprocessing reduces the instrument and environmental noise in the spectral data, weakens and eliminates the effects of sample unevenness, baseline drift, high-frequency noise, and stray light on the spectral signal. It standardizes the spectral data of each sample so that the spectral data of different samples have the same mean and unit variance, eliminating the differences between samples.

4. The method for accurately evaluating plant leaf color by combining spectrum with RBFNN according to claim 1, characterized in that: In step 2, the specific process of establishing the plant leaf color RBFNN discrimination model is as follows: According to the characteristics of plant leaf color depth, the plants are divided into leaf color grades, and the graded variable values ​​are determined for the samples; at this stage, the center of the hidden layer neurons needs to be determined; the training sample set is clustered using the k-means algorithm, and the obtained cluster center is used as the center of the radial basis function; after determining the center of the radial basis function, the next step is to calculate the neuron connection weights between the hidden layer and the output layer; this is done using the least squares method (OLS) or the gradient descent method; through the above two stages of training, RBFNN can learn an effective representation of the input data, and can perform prediction or classification tasks based on the training data, and establish an RBFNN discrimination model.

5. The method for accurately evaluating plant leaf color by combining spectrum with RBFNN according to claim 1, characterized in that: In step three, include: The RBFNN model is validated using a permutation test; data is collected for training the RBFNN model and corresponding real data is collected for validating the model; the RBFNN model is trained using the collected data to obtain the model's predicted output; the difference between the model's predicted output and the actual output is calculated, the labels of the actual data are randomly rearranged, and then the difference statistics are recalculated using these permuted data; this process is repeated multiple times to generate a null distribution; the difference statistics calculated on the original data are compared with the null distribution obtained through the permutation operation to evaluate the performance of the model.

6. The method for accurately evaluating plant leaf color by combining spectrum with RBFNN according to claim 5, characterized in that: After evaluating the performance of the model, we then looked at the R 2 The coefficient of determination of the numerical discriminant model is used to evaluate the accuracy of the model prediction.

7. The method for accurately evaluating plant leaf color by combining spectrum with RBFNN according to claim 6, characterized in that: R 2 The specific numerical discriminant model is: R 2 The neural network is an extension of the radial basis function neural network; unlike the traditional neural network, R 2 Neural networks are particularly suitable for processing data with nonlinear characteristics; the R value, also known as the coefficient of determination, is an indicator used in statistics to measure the goodness of fit of a model; when the R value approaches 1, it means that the model can fit the data perfectly without any error.

Images