Modeling method for hyperspectral medicine component analysis based on graph convolutional neural network

Abstract

The invention provides a modeling method for hyperspectral medical component analysis based on a graph convolutional neural network, relates to the technical field of medical component analysis, and realizes effective correction of nonlinear light intensity and temperature and humidity change influence in spectral data through integrated application of illumination and environment correction technologies. By adopting the organic combination of the multiple regression model and the graph convolutional neural network, the robustness and stability of spectral data are ensured, and the accuracy and efficiency of analysis are greatly improved. Therefore, not only is the dependence of experimental conditions on results reduced, but also the application range of the technology is expanded, so that the detection of medicine components is also reliable under variable environmental conditions, and a solid foundation is provided for subsequent component prediction. Through the scheme, a user can obtain a high-consistency detection result in an unfixed external environment, the detection process is simplified, and the overall detection capability is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of pharmaceutical component analysis, and in particular to a modeling method for graph convolutional neural network hyperspectral pharmaceutical component analysis. Background Art

[0002] In the analysis of hyperspectral pharmaceutical ingredients, light intensity and environmental conditions (such as temperature and humidity) have a significant impact on spectral data. Traditional methods usually rely on placing samples in equipment with constant light, temperature, and humidity. However, this method is not only easily affected by external ambient light, but also increases the complexity of the equipment and limits the flexibility of use scenarios. Especially in the continuous detection of drug samples, subtle changes in light and environment may introduce errors and affect the reliability of data.

[0003] In the prior art, such as the implementation method of a graph convolutional neural network hyperspectral pharmaceutical component analysis published by China Patent Publication No. CN113269196B, on the one hand, the pharmaceutical hyperspectral image data is processed into graph data, which greatly reduces the number of pixels and effectively reduces the amount of data; on the other hand, the characteristic information of the drug is extracted by the graph convolutional neural network model, and the spatial relationship between the visual features in the drug hyperspectral image and the drug components is effectively learned, the representation ability of the classification features of the drug components is improved, and the accuracy of the components and properties of the tested drugs is improved, which can realize non-destructive and rapid detection and analysis of the drug components and quality. Although it can quickly analyze the pharmaceutical components, it still does not take into account the impact of factors such as light, temperature, and humidity on the spectral data of the pharmaceutical components in the model. It is still necessary to use physical isolation to reduce the errors caused by environmental factors before collecting spectral data, which increases the complexity of the equipment and limits the flexibility of the use scenario. Therefore, in order to ensure the accuracy and consistency of the data, it is necessary to develop a technology that can automatically perform illumination and environmental correction in a variable environment, so that the analysis process is simpler and more efficient, and suitable for practical application scenarios.

[0004] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0005] The purpose of the present invention is to provide a modeling method for graph convolutional neural network hyperspectral pharmaceutical component analysis to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] A modeling method for graph convolutional neural network hyperspectral pharmaceutical component analysis, the specific steps include:

[0008] S1: Under the same environmental conditions, target spectral data of target components of drug samples under different light intensities are collected, and baseline spectral data are generated to perform light correction on the target spectral data;

[0009] S2: Analyze the target spectrum data after illumination correction using principal component analysis, and select the bands in the target spectrum data that are most relevant to the target component according to the analysis results and mark them as key bands;

[0010] S3: changing the environmental conditions, collecting the target spectrum data under different environmental conditions and performing illumination correction, constructing a multivariate regression model, wherein the input of the multivariate regression model is different environmental conditions, and the output is the change in the wavelength of the reflected light in the key band of the target spectrum data after illumination correction, and the multivariate regression model is trained and optimized;

[0011] S4: Construct a graph convolutional neural network model, take the key band reflected light wavelength of the target spectral data after illumination correction and environmental correction as input, and take the target components of the corresponding drug samples as output labels to train the graph convolutional neural network model, obtain the key band reflected light wavelength of the target spectral data of the drug to be tested after illumination correction and environmental correction, and input it into the graph convolutional neural network model to generate the target components of the drug to be tested.

[0012] Preferably, the environmental conditions include ambient temperature and ambient humidity, and the logic for generating baseline spectral data to perform illumination correction on target spectral data is:

[0013] Arrange the target spectral data based on the order of acquisition;

[0014] Calculate the intensity difference and spectrum difference between two adjacent sets of target spectrum data, and fit the intensity difference and spectrum difference using a cubic polynomial equation;

[0015] Calculate the light intensity difference between the light intensity corresponding to different target spectral data and the preset reference light intensity, and substitute it into the cubic polynomial equation to generate the corresponding spectral difference;

[0016] The corresponding spectral difference is subtracted from the target spectral data to achieve illumination correction of the target spectral data.

[0017] Preferably, the collected The target spectral data of the group is calibrated as , ,in Indicates The target spectrum data The wavelength of reflected light in each band, For the The light intensity corresponding to the target spectral data of the group, subscript Indicates the index of the target spectrum data, subscript Represents the index of the band in the target spectral data, Indicates the number of bands in the target spectral data. The calculation methods of the intensity difference and spectrum difference between two adjacent sets of target spectral data are:

[0018]

[0019]

[0020]

[0021] In the formula Respectively represent Group and The spectral difference and corresponding light intensity difference between the target spectral data of the group, Indicates Group and The target spectral data of the group The difference in wavelength of reflected light in each band;

[0022] The cubic polynomial equation is expressed as:

[0023]

[0024] In the formula are all polynomial coefficients;

[0025] Calculate the light intensity difference between the light intensity corresponding to different target spectral data and the preset reference light intensity. The calculation method is:

[0026]

[0027] In the formula Indicates The light intensity difference between the light intensity corresponding to the target spectral data and the preset reference light intensity, Indicates the preset reference light intensity;

[0028] Will Substituting into the cubic polynomial equation, replacing To calculate the corresponding spectral difference, and use it to correct the illumination of the target spectral data,

[0029] The wavelength of reflected light after illumination correction is expressed as:

[0030]

[0031] The target spectrum data after illumination correction is expressed as:

[0032]

[0033] In the formula represents the target spectrum data after illumination correction, , Representation and The corresponding spectral difference is .

[0034] Preferably, the logic for screening key bands is:

[0035] Calculate the mean of multiple sets of target spectral data after illumination correction, and then construct the covariance matrix. The calculation methods are as follows:

[0036]

[0037]

[0038] In the formula , Respectively represent the mean of multiple groups of target spectral data after illumination correction, and the covariance matrix of the target spectral data, Indicates the number of target spectral data, the superscript represents transpose;

[0039] Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors. The calculation method of eigenvalue decomposition is:

[0040]

[0041] In the formula Represents the covariance matrix Perform eigenvalue decomposition, , denote the eigenvalue and the corresponding eigenvector respectively, and the subscript Indicates the index of eigenvalues ​​and eigenvectors;

[0042] Sort all eigenvalues ​​from small to large and calculate the corresponding cumulative explained variance. The calculation method is:

[0043]

[0044] In the formula represents the cumulative explained variance;

[0045] choose The band corresponding to the eigenvector of is identified as the key band.

[0046] Preferably,

[0047] The construction logic of the multivariate regression model is:

[0048] Calculate the change in wavelength of reflected light in the key band of the target spectral data after illumination correction , where the subscript is the index of the key band, calculated as:

[0049]

[0050] Then, a multiple regression model is constructed between the change in the wavelength of reflected light in the key bands of the target spectral data after illumination correction and the environmental conditions. The calculation method is:

[0051]

[0052] In the formula , are regression coefficients, represents the model error, , Respectively represent The ambient temperature and humidity corresponding to the target spectral data;

[0053] After the multivariate regression model optimization is completed, , As reference temperature and reference humidity respectively, calculate the change in reflected light wavelength of the key band corresponding to each corrected target spectrum data , calculated as:

[0054]

[0055] based on Get the spectral difference corresponding to the ambient temperature and ambient humidity , ,in Indicates the number of key bands;

[0056] The wavelength of reflected light after environmental correction is expressed as:

[0057]

[0058] In the formula Indicates the wavelength of reflected light after environmental correction;

[0059] The target spectrum data after environmental correction is expressed as:

[0060]

[0061] represents the target spectrum data after environmental correction, .

[0062] Preferably, the graph convolution layer in the graph convolutional neural network model is expressed as:

[0063]

[0064] In the formula Indicates The node characteristics of the layer, , denote the adjacency matrix and degree matrix respectively, Indicates The training matrix of the layer, Represents the activation function.

[0065] Preferably, the graph convolutional neural network model includes two graph convolutional layers, and uses the Relu function as the activation function and the cross entropy loss function. The target spectral data after illumination correction and environment correction is input into the preset graph convolutional neural network model, and the predicted probability corresponding to the target component is output.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] The present invention realizes effective correction of the influence of nonlinear light intensity and temperature and humidity changes in spectral data through the integrated application of illumination and environmental correction technology. The organic combination of multivariate regression model and graph convolutional neural network ensures the robustness and stability of spectral data, greatly improving the accuracy and efficiency of analysis. This not only reduces the dependence of experimental conditions on results, but also expands the scope of application of the technology, making the detection of drug ingredients equally reliable under variable environmental conditions, providing a solid foundation for subsequent ingredient prediction. Through this solution, users can obtain highly consistent test results in a non-fixed external environment, simplifying the detection process and improving the overall detection capability. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a schematic diagram of the overall method flow of the present invention. DETAILED DESCRIPTION

[0069] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0070] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0071] Example:

[0072] See also Figure 1 , the present invention provides a technical solution:

[0073] A modeling method for graph convolutional neural network hyperspectral pharmaceutical component analysis, the specific steps include:

[0074] S1: Under the same environmental conditions, target spectral data of target components of drug samples under different light intensities are collected, and baseline spectral data are generated to perform light correction on the target spectral data.

[0075] It is understandable that changes in light intensity will affect the spectrum of drug samples. In order to eliminate this effect, the existing technology generally places the object to be tested in a specific device and achieves the goal by artificially providing a light source with constant light intensity. However, this method still has certain disadvantages. First, it is easily affected by the light of the external environment. Second, it leads to a narrower scope of application of the entire equipment and makes the steps of the entire detection process more cumbersome, which undoubtedly adds many potential error risks. By correcting the light intensity of the target spectral data, the impact of subtle light changes on the spectral data during the continuous detection of drug samples can be reduced, and the noise introduced by light changes can be reduced to ensure the consistency and reliability of the data during the detection process. At the same time, it also makes the detection process of drug samples no longer have to be in a specific device, such as a black box, thereby expanding the applicable scenarios and making it more conducive to practical applications.

[0076] Environmental conditions include ambient temperature and ambient humidity. The logic for generating baseline spectral data to perform illumination correction on target spectral data is as follows:

[0077] Arrange the target spectral data based on the order of acquisition;

[0078] Calculate the intensity difference and spectrum difference between two adjacent sets of target spectrum data, and fit the intensity difference and spectrum difference using a cubic polynomial equation;

[0079] Calculate the light intensity difference between the light intensity corresponding to different target spectral data and the preset reference light intensity, and substitute it into the cubic polynomial equation to generate the corresponding spectral difference;

[0080] The corresponding spectral difference is subtracted from the target spectral data to achieve illumination correction of the target spectral data.

[0081] The target spectral data is calibrated as , ,in Indicates that the light intensity is At that time, The target spectrum data The wavelength of reflected light in each band, subscript Indicates the index of the target spectrum data, subscript Represents the index of the band in the target spectral data, Indicates the number of bands in the target spectral data. The calculation methods of the intensity difference and spectrum difference between two adjacent sets of target spectral data are:

[0082]

[0083]

[0084]

[0085] In the formula , Respectively represent the intensity difference and spectrum difference between two adjacent sets of target spectral data, Indicates that the light intensity is When the first The difference in wavelength of reflected light in each band.

[0086] The cubic polynomial equation is expressed as:

[0087]

[0088] In the formula ~ represents polynomial coefficients;

[0089] Calculate the light intensity difference between the light intensity corresponding to different target spectral data and the preset reference light intensity. The calculation method is:

[0090]

[0091] In the formula Indicates the light intensity difference between the light intensity corresponding to different target spectral data and the preset reference light intensity. Indicates the preset reference light intensity;

[0092] Will Substituting into the cubic polynomial equation, replacing To calculate the corresponding spectral difference, and use it to correct the illumination of the target spectral data,

[0093] The wavelength of reflected light after illumination correction is expressed as:

[0094]

[0095] The target spectrum data after illumination correction is expressed as:

[0096]

[0097] In the formula represents the target spectrum data after illumination correction, , Representation and The corresponding spectral difference is .

[0098] It can be seen from the expressions of the target spectral data and the spectral difference that both are vectors. By using cubic polynomial fitting to correct the nonlinear relationship between light intensity and spectral response, higher flexibility and adaptability are provided. Compared with the linear model, the cubic polynomial model can better capture complex optical changes, and by calculating the intensity and wavelength differences of adjacent data, the subtle changes between the data can be carefully captured to ensure the accuracy of the correction. This difference calculation makes the correction of spectral data more targeted.

[0099] In this step, by eliminating the influence of light intensity on spectral data, the model can have consistent performance under different lighting conditions, enhancing the robustness and stability of the model and reducing the errors introduced by environmental changes, thus providing a stable starting point for subsequent analysis and greatly reducing the errors introduced by environmental changes. This not only improves the accuracy of feature extraction, but also makes the final component prediction more credible.

[0100] S2: Use the principal component analysis method to analyze the target spectral data after illumination correction, and select the bands in the target spectral data that are most relevant to the target component based on the analysis results and mark them as key bands.

[0101] Principal component analysis is an effective dimensionality reduction method that can extract the most representative features from high-dimensional data. This not only simplifies the data, but also reduces the computational complexity of subsequent analysis. By selecting eigenvectors containing the main variance information, PCA can remove noise and redundant information in the data, so that the analysis focuses on the most important signals in the data.

[0102] The logic for filtering key bands is:

[0103] Calculate the mean of multiple sets of target spectral data after illumination correction, and then construct the covariance matrix. The calculation methods are as follows:

[0104]

[0105]

[0106] In the formula , Respectively represent the mean of multiple groups of target spectral data after illumination correction, and the covariance matrix of the target spectral data, Indicates the number of target spectral data, the superscript Represents the transpose, through the eigenvalue decomposition of the covariance matrix, principal component analysis provides a mathematically interpretable way to understand the variance distribution of the data and select the band that best represents the characteristics of the data.

[0107] Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors. The calculation method of eigenvalue decomposition is:

[0108]

[0109] In the formula Represents the covariance matrix Perform eigenvalue decomposition, , denote the eigenvalue and the corresponding eigenvector respectively, and the subscript Indicates the index of eigenvalues ​​and eigenvectors;

[0110] Sort all eigenvalues ​​from small to large and calculate the corresponding cumulative explained variance. The calculation method is:

[0111]

[0112] In the formula represents the cumulative explained variance;

[0113] choose The band corresponding to the eigenvector of is identified as the key band.

[0114] In this step, principal component analysis is performed based on illumination correction to further ensure that the data input to subsequent models is highly relevant and refined, which improves the reliability of the overall solution. By selecting key bands, this step greatly improves the quality of input data for multivariate regression models and graph convolutional neural network models, allowing these models to capture the characteristic patterns of target components more accurately.

[0115] S3: Change the environmental conditions, collect the target spectral data under different environmental conditions and perform illumination correction, build a multivariate regression model between the change in the wavelength of the reflected light in the key bands of the target spectral data after illumination correction under different environmental conditions and the environmental conditions, and train and optimize the multivariate regression model.

[0116] Similar to light intensity, ambient temperature and humidity will also affect the spectrum of drug samples. By correcting the target spectral data for environmental factors, the impact of changes in ambient temperature and humidity on spectral data can be reduced during the continuous detection of drug samples, ensuring the consistency and reliability of data during the detection process, making it more conducive to practical applications.

[0117] The construction logic of the multiple regression model is:

[0118] Calculate the change in wavelength of reflected light in the key band of the target spectral data after illumination correction , where the subscript is the index of the key band, calculated as:

[0119]

[0120] Then, a multiple regression model is constructed between the change in the wavelength of reflected light in the key bands of the target spectral data after illumination correction and the environmental conditions. The calculation method is:

[0121]

[0122] In the formula , are regression coefficients, represents the model error, , Respectively represent The ambient temperature and humidity corresponding to the target spectral data;

[0123] After the multivariate regression model optimization is completed, , As reference temperature and reference humidity respectively, calculate the change in reflected light wavelength of the key band corresponding to each corrected target spectrum data , calculated as:

[0124]

[0125] The multivariate regression model captures the complex effects of environmental conditions on spectral data, including linear and some nonlinear relationships. It has strong adaptability and can clarify the specific impact of environmental conditions (such as temperature and humidity) on spectral data. This targeted correction method is more detailed and effective than direct overall adjustment.

[0126] based on Get the spectral difference corresponding to the ambient temperature and ambient humidity , ,in Indicates the number of key bands;

[0127] The wavelength of reflected light after environmental correction is expressed as:

[0128]

[0129] In the formula Indicates the wavelength of reflected light after environmental correction;

[0130] The target spectrum data after environmental correction is expressed as:

[0131]

[0132] represents the target spectrum data after environmental correction, .

[0133] In this step, the spectral data after environmental correction is closer to the real data under standard conditions, which improves the reliability of the data and ensures that the spectral data input into the subsequent model is consistent under different environmental conditions, thereby improving the effectiveness of the entire scheme. At the same time, through this correction method, the entire detection system can effectively respond to environmental changes whether in a laboratory environment or in an actual application environment, and the environmental correction method provided in this step can enhance the robustness and applicability of the technical solution.

[0134] S4: Construct a graph convolutional neural network model, couple the multivariate regression model with the graph convolutional neural network model, use the multivariate regression model to perform environmental correction on the reflected light wavelength in the key band, and then use the graph convolutional neural network to process the reflected light wavelength after environmental correction in the key band to generate the predicted probability of the target component.

[0135] The target spectral data after illumination correction under different environmental conditions were divided into training set and validation set in a ratio of 7:3, and were respectively brought into the multivariate regression model for training and optimization.

[0136] The graph convolution layer in the graph convolutional neural network model is represented as:

[0137]

[0138] In the formula Indicates The node characteristics of the layer, , denote the adjacency matrix and degree matrix respectively, Indicates The training matrix of the layer, Represents the activation function.

[0139] The graph convolutional neural network model includes two graph convolutional layers, uses the Relu function as the activation function, and adopts the cross-entropy loss function. The target spectral data after illumination correction and environment correction is input into the preset graph convolutional neural network model, and the predicted probability corresponding to the target component is output.

[0140] The graph convolutional neural network model is good at extracting features from graph structure data. It can effectively capture and extract useful information for the multidimensional features and complex relationships of hyperspectral data. Since it is an existing technology, its specific working principle will not be described in detail. By combining the multivariate regression model with the graph convolutional neural network, the environmental correction ability of the multivariate regression and the powerful feature extraction ability of the graph convolutional neural network model are fully utilized to improve the prediction accuracy, so that the model can effectively learn in complex feature spaces, and can also process large amounts of data more quickly, improve the efficiency of data analysis, and provide more reliable results.

[0141] In summary, the present invention realizes effective correction of the influence of nonlinear light intensity and temperature and humidity changes in spectral data through the integrated application of illumination and environmental correction technology. The organic combination of multivariate regression model and graph convolutional neural network ensures the robustness and stability of spectral data, greatly improving the accuracy and efficiency of analysis. This not only reduces the dependence of experimental conditions on results, but also expands the scope of application of the technology, making the detection of drug ingredients equally reliable under variable environmental conditions, providing a solid foundation for subsequent ingredient prediction. Through this solution, users can obtain highly consistent test results in a non-fixed external environment, simplifying the detection process and improving the overall detection capability.

[0142] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0143] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. Those skilled in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0144] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0145] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A modeling method for hyperspectral pharmaceutical component analysis using a graph convolutional neural network, characterized in that: The specific steps include: S1: Under the same environmental conditions, target spectral data of target components of drug samples under different light intensities are collected, and baseline spectral data are generated to perform light correction on the target spectral data; S2: Analyze the target spectrum data after illumination correction using principal component analysis, and select the bands in the target spectrum data that are most relevant to the target component according to the analysis results and mark them as key bands; S3: changing the environmental conditions, collecting the target spectrum data under different environmental conditions and performing illumination correction, constructing a multivariate regression model, wherein the input of the multivariate regression model is different environmental conditions, and the output is the change in the wavelength of the reflected light in the key band of the target spectrum data after illumination correction, and the multivariate regression model is trained and optimized; S4: Construct a graph convolutional neural network model, take the key band reflected light wavelength of the target spectral data after illumination correction and environmental correction as input, and take the target components of the corresponding drug samples as output labels to train the graph convolutional neural network model, obtain the key band reflected light wavelength of the target spectral data of the drug to be tested after illumination correction and environmental correction, and input it into the graph convolutional neural network model to generate the target components of the drug to be tested.

2. The modeling method of a graph convolutional neural network hyperspectral pharmaceutical component analysis according to claim 1 is characterized by: The environmental conditions include ambient temperature and ambient humidity. The logic for generating baseline spectral data to perform illumination correction on target spectral data is: Arrange the target spectral data based on the order of acquisition; Calculate the intensity difference and spectrum difference between two adjacent sets of target spectrum data, and fit the intensity difference and spectrum difference using a cubic polynomial equation; Calculate the light intensity difference between the light intensity corresponding to different target spectral data and the preset reference light intensity, and substitute it into the cubic polynomial equation to generate the corresponding spectral difference; The corresponding spectral difference is subtracted from the target spectral data to achieve illumination correction of the target spectral data.

3. The modeling method of a graph convolutional neural network hyperspectral pharmaceutical component analysis according to claim 2 is characterized by: The collected The target spectral data of the group is calibrated as , ,in Indicates The target spectrum data The wavelength of reflected light in each band, For the The light intensity corresponding to the target spectral data of the group, subscript Indicates the index of the target spectrum data, subscript Represents the index of the band in the target spectral data, Indicates the number of bands in the target spectral data. The calculation methods of the intensity difference and spectrum difference between two adjacent sets of target spectral data are: In the formula Respectively represent Group and The spectral difference and corresponding light intensity difference between the target spectral data of the group, Indicates Group and The target spectral data of the group The difference in wavelength of reflected light in each band; The cubic polynomial equation is expressed as: In the formula are all polynomial coefficients; Calculate the light intensity difference between the light intensity corresponding to different target spectral data and the preset reference light intensity. The calculation method is: In the formula Indicates The light intensity difference between the light intensity corresponding to the target spectral data and the preset reference light intensity, Indicates the preset reference light intensity; Will Substituting into the cubic polynomial equation, replacing To calculate the corresponding spectral difference, and use it to correct the illumination of the target spectral data, The wavelength of reflected light after illumination correction is expressed as: The target spectrum data after illumination correction is expressed as: In the formula represents the target spectrum data after illumination correction, , Representation and The corresponding spectral difference is .

4. The modeling method of a graph convolutional neural network hyperspectral pharmaceutical component analysis according to claim 3 is characterized by: The logic for filtering key bands is: Calculate the mean of all target spectral data after illumination correction, and then construct the covariance matrix. The calculation methods are: In the formula , They represent the mean of all target spectral data after illumination correction and the covariance matrix of the target spectral data respectively. Indicates the number of all target spectral data collected, with the superscript represents transpose; Perform eigenvalue decomposition on the covariance matrix to obtain the eigenvalues ​​and corresponding eigenvectors. The calculation method of eigenvalue decomposition is: In the formula Represents the covariance matrix Perform eigenvalue decomposition, , Respectively represent The target spectrum data The eigenvalues ​​and corresponding eigenvectors of each band; Sort all eigenvalues ​​from small to large and calculate the corresponding cumulative explained variance. The calculation method is: In the formula represents the cumulative explained variance; choose The band corresponding to the eigenvector of is identified as the key band.

5. The modeling method of a graph convolutional neural network hyperspectral pharmaceutical component analysis according to claim 4 is characterized in that: The construction logic of the multivariate regression model is: Calculate the change in wavelength of reflected light in the key band of the target spectral data after illumination correction , where the subscript is the index of the key band, calculated as: Then, a multiple regression model is constructed between the change in the wavelength of reflected light in the key bands of the target spectral data after illumination correction and the environmental conditions. The calculation method is: In the formula , are regression coefficients, represents the model error, , Respectively represent The ambient temperature and humidity corresponding to the target spectral data; After the multivariate regression model optimization is completed, , As reference temperature and reference humidity respectively, calculate the change in reflected light wavelength of the key band corresponding to each corrected target spectrum data , calculated as: based on Get the spectral difference corresponding to the ambient temperature and ambient humidity , ,in Indicates the number of key bands; The wavelength of reflected light after environmental correction is expressed as: In the formula Indicates the wavelength of reflected light after environmental correction; The target spectrum data after environmental correction is expressed as: represents the target spectrum data after environmental correction, .

6. The modeling method of a graph convolutional neural network hyperspectral pharmaceutical component analysis according to claim 1, characterized in that: The target spectral data after illumination correction under different environmental conditions were divided into training set and validation set in a ratio of 7:3, and were respectively brought into the multivariate regression model for training and optimization.

7. The modeling method of a graph convolutional neural network hyperspectral pharmaceutical component analysis according to claim 5, characterized in that: The graph convolution layer in the graph convolutional neural network model is expressed as: In the formula Indicates The node features of the convolutional layer, , denote the adjacency matrix and degree matrix respectively, Indicates The training matrix of the convolutional layer, Represents the activation function.

8. The modeling method of a graph convolutional neural network hyperspectral pharmaceutical component analysis according to claim 7, characterized in that: The graph convolutional neural network model includes two graph convolutional layers, uses the Relu function as the activation function, and adopts the cross entropy loss function. The target spectral data after illumination correction and environment correction is input into the preset graph convolutional neural network model, and the predicted probability corresponding to the target component is output.

Images