Method for discriminating producing area of genuine medicinal material based on multi-modal molecular spectral characteristic recombination

Abstract

The invention belongs to the field of quality control of traditional Chinese medicinal materials, and particularly relates to a genuine medicinal material producing area distinguishing method based on multi-mode molecular spectrum characteristic recombination. Accurate identification of the producing area of the traditional Chinese medicine decoction pieces / powder is achieved by collecting the near infrared spectrum, the middle infrared spectrum and the terahertz spectrum of the sample, meanwhile, the recombination characteristic spectrum of the traditional Chinese medicine decoction pieces / powder in the near infrared spectrum, the middle infrared spectrum and the terahertz wave band can be visually displayed, and the method has the advantages of being higher in identification pertinence, higher in accuracy and wider in application range.

Description

Technical Field

[0001] The present invention belongs to the field of quality control of traditional Chinese medicines, and in particular relates to a method for identifying the origin of authentic medicinal materials based on multimodal molecular spectral feature reconstruction. Background Art

[0002] Traditional Chinese medicine (TCM) is an essential material foundation for treating diseases, regulating the body, and maintaining health. They boast a wide range of therapeutic applications, high safety, minimal toxicity and side effects, customizable formulations, and a rich cultural heritage. The type and authenticity of TCM ingredients are crucial for their efficacy. The composition of different TCMs varies significantly, and many look similar, making them difficult to accurately identify with the naked eye. The composition of the same TCM from different origins also varies, and some are affected by geological and environmental contamination. Therefore, to ensure the reliability of the efficacy of TCM ingredients, accurate identification of their type and origin is crucial for the development of TCM.

[0003] Traditional methods for identifying the origin of Chinese medicinal materials include phenotypic identification, gas chromatography, thin-layer chromatography, and high-performance liquid chromatography. Phytochemical identification relies on subjective experience. While simple and cost-effective, accuracy is difficult to guarantee. Chemical analysis, while capable of providing detailed compositional information, suffers from complex sample preparation, potential sample damage, long analysis cycles, low efficiency, and the potential for chemical contamination.

[0004] Spectral analysis technology is highly regarded in the quality assessment of traditional Chinese medicines due to its high efficiency, accuracy, and cost-effectiveness. Currently, most methods rely solely on a single spectral technique to assess the quality of traditional Chinese medicines. However, in practical applications, single spectral technology has limitations such as susceptibility to interference, insufficient ability to handle complex samples, and low detection and analysis accuracy in some samples. It cannot fully reveal the complex chemical composition and structure of traditional Chinese medicines. These shortcomings greatly limit the application of spectral technology in the field of identification of the origin of traditional Chinese medicines. Compared with single spectral methods, multispectral technology can provide more comprehensive chemical composition and structural information, and significantly improve identification accuracy by combining multi-band data. In addition, multispectral technology effectively reduces environmental interference and enhances the stability and reliability of detection.

[0005] However, most existing multispectral technologies target the characteristics of a single component and provide limited information. The interactions between different substances in actual complex samples will affect the spectral response, which is different from the response of a single substance. The above methods cannot fully reveal the complex chemical composition and structure of Chinese medicinal materials. The lack of multidimensional data support also means that the recognition accuracy of a single spectral technology cannot meet the requirements of higher-precision identification, and some technologies may only be effective for specific types of samples. Summary of the Invention

[0006] Based on the above-mentioned defects, the purpose of this application is to provide a method for identifying the origin of authentic medicinal materials based on the recombination of multimodal molecular spectral features. This method does not require spectral preprocessing and directly realizes the accurate identification of the origin of Chinese medicinal material slices / powders by collecting near-infrared, mid-infrared and terahertz spectra of the samples. At the same time, it can visualize the recombined characteristic spectra of Chinese medicinal material slices / powders in the near-infrared spectrum, mid-infrared spectrum and terahertz bands, and has the characteristics of stronger identification targeting, higher accuracy and wider adaptability.

[0007] To this end, the technical solution of the present application provides a method for identifying the origin of authentic medicinal materials based on multimodal molecular spectral feature reconstruction, comprising the following steps:

[0008] Pre-processing the target Chinese medicinal materials to make them into powder or flakes, wherein the target Chinese medicinal materials are produced from no less than four sources;

[0009] The near-infrared, mid-infrared and terahertz spectra of the crushed target Chinese medicinal materials were collected. Abnormal spectra were eliminated using the Mahalanobis distance threshold, and then the spectral ranges of the near-infrared, mid-infrared and terahertz were corrected by judging the spectral signal-to-noise ratio.

[0010] The corrected spectra were trained using a convolutional neural network model to extract and reconstruct the spectral features, and the reconstructed features were visualized using class activation mapping.

[0011] Build models and evaluate and debug model performance.

[0012] Furthermore, the crushed particle size is sieved through a 200-mesh sieve.

[0013] Furthermore, in the training of the convolutional neural network model, there are 3 convolution layers, the convolution kernel size is 3, the number of filters is 128, 256, and 512 respectively; and the step size is 2.

[0014] Furthermore, in the training of the convolutional neural network model, the number of training times is 800-2000 times, and the ratio of the training set to the test set is 8:2 or 7:3.

[0015] Beneficial effects of the present invention: This application realizes accurate identification of the origin of Chinese medicinal material slices / powders by collecting near-infrared, mid-infrared and terahertz spectra of samples, and can visually display the recombined characteristic spectra of Chinese medicinal material slices / powders in the near-infrared spectrum, mid-infrared spectrum and terahertz band, with the characteristics of stronger identification targeting, higher accuracy and wider adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is the abnormal spectrum elimination diagram based on Mahalanobis distance;

[0017] Figure 2 This is a set of near-infrared spectra of Chuanxiong slices;

[0018] Figure 3 This is a graph showing the accuracy change of the model validation set;

[0019] Figure 4 The accuracy of the validation set for training the model 1000 times;

[0020] Figure 5 is the recombinant characteristic spectrum of origin 1;

[0021] Figure 6 This is the recombinant characteristic spectrum of origin 2;

[0022] Figure 7 This is the recombinant characteristic spectrum of origin 3;

[0023] Figure 8 This is the recombinant characteristic spectrum of origin 4;

[0024] Figure 9 This is the recombinant characteristic spectrum of origin 5;

[0025] Figure 10 This is the recombinant characteristic spectrum of origin 6. DETAILED DESCRIPTION

[0026] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0027] Unless otherwise specifically stated, the relative arrangement of the parts and steps, numerical expressions and numerical values ​​set forth in these embodiments do not limit the scope of the present application. At the same time, it should be understood that, for ease of description, the sizes of the various parts shown in the drawings are not drawn according to actual proportional relationships. The techniques, methods and equipment known to those of ordinary skill in the relevant art may not be discussed in detail, but where appropriate, the techniques, methods and equipment should be considered as part of the authorization specification. In all examples shown and discussed here, any specific values ​​should be interpreted as being merely exemplary and not as limitations. Therefore, other examples of the exemplary embodiments may have different values. It should be noted that similar numbers and letters represent similar items in the following figures, and therefore, once an item is defined in one figure, it does not need to be further discussed in subsequent figures.

[0028] The experimental methods in the following examples are all conventional methods unless otherwise specified. The experimental materials used in the following examples were all purchased from commercial channels, and quality control measures were implemented during the experimental process to ensure the quality of the data results.

[0029] The first embodiment of the present application discloses a method for identifying the origin of authentic medicinal materials based on multimodal molecular spectral feature recombination, comprising the following steps:

[0030] Grinding the target Chinese medicinal materials, where the target Chinese medicinal materials come from no less than four sources;

[0031] The near-infrared, mid-infrared and terahertz spectra of the crushed target Chinese medicinal materials were collected. Abnormal spectra were eliminated using the Mahalanobis distance threshold, and then the spectral ranges of the near-infrared, mid-infrared and terahertz were corrected by judging the spectral signal-to-noise ratio.

[0032] The corrected spectra were trained using a convolutional neural network model to extract and reconstruct the spectral features, and the reconstructed features were visualized using class activation mapping.

[0033] Build models and evaluate and debug model performance.

[0034] In this embodiment, the multimodal molecular spectroscopy technique can collect 3.3-12820cm -1 Compared with the collection of single spectral information, the spectral information within the range, that is, the information within the near-infrared, mid-infrared and terahertz bands, can obtain more comprehensive information. The convolutional layer in the convolutional neural network is used to capture the key part of the local features. Each convolutional layer is responsible for extracting a specific feature. At the same time, as the number of convolutional layers increases, more and more in-depth features can be extracted layer by layer, and important bands can be gradually strengthened and irrelevant noise can be weakened. The activation function can capture more complex features. Nonlinear activation is the key to learning and simulating complex functions. The pooling layer is used to reduce the spatial size of the feature map, thereby reducing the number of parameters and the amount of calculation.

[0035] In this embodiment, the near-infrared spectrum in the multimodal molecular spectroscopy technique refers to the near-infrared spectrum with a wave number of 12820-4000 cm -1 The spectral region mainly reflects the frequency harmonics and combination frequency vibrations of hydrogen-containing functional groups (such as OH, NH and CH) in molecules. It is mostly used for the detection of organic substances. The mid-infrared spectrum refers to the wave number between 4000-400cm -1 The spectral region mainly reflects the detailed information of molecular vibration, especially the fundamental frequency vibration of molecular groups, which is of high value for the detection of various chemical bonds and minerals. Terahertz spectroscopy refers to the 333-3.3cm -1Terahertz waves can penetrate many non-metallic materials and cover the vibrations of organic and inorganic macromolecules, including intermolecular vibrations and collective molecular vibrations. They are also extremely sensitive to water molecules. Therefore, by combining near-infrared, mid-infrared, and terahertz spectroscopy and using them in a complementary manner, we can provide more comprehensive molecular-level spectral information, including chemical bonds, functional groups, and intermolecular-to-molecular information, enabling more accurate identification and analysis of various substances.

[0036] In this embodiment, a convolutional neural network model is combined with a convolutional neural network model of a class activation mapping algorithm. By visualizing the feature map of the last convolutional layer of the convolutional neural network model, the spectrum that contributes most to the identification of Chinese herbal medicine slices / powders from different origins, i.e., the characteristic spectrum, is extracted.

[0037] In this implementation, the class activation map is generated by combining the feature map of the last convolutional layer in the convolutional neural network with the weights of the fully connected layer. The map highlights the input image areas that have the greatest impact on the model's classification decision, thereby providing an intuitive explanation of the model's decision-making process.

[0038] In a further embodiment, the crushed particle size is to pass through a 200-mesh sieve; the elimination of abnormal spectra is to eliminate spectra with low signal-to-noise; in the training using the convolutional neural network model, the convolution layer is 3 layers, the convolution kernel size is 3, the number of filters is 128, 256, and 512 respectively; the step size is 2; the number of training times is 1000 times, and the ratio of training set to test set is 7:3.

[0039] The following will take Chuanxiong slices as an example to explain the technical solution and technical effects of this application in more detail in combination with specific embodiments and drawings.

[0040] Example 1 Identification of the Origin of Chuanxiong Pieces

[0041] S1. Purchase Chuanxiong Rhizome from six different production areas, process the purchased Chuanxiong Rhizome into 4mm thick slices, remove surface impurities, and store in a cool, dry place. Randomly select 50 slices of Chuanxiong Rhizome with uniform thickness and no holes on the surface from each production area.

[0042] S2. Set the spectrometer's scanning range, resolution, number of scans, number of scan repetitions, and other parameters to collect near-infrared, mid-infrared, and terahertz spectra of the Chuanxiong slices.

[0043] S3. Combine Mahalanobis distance to eliminate abnormal spectra, such as Figure 1 As shown in , the spectra with Mahalanobis distance greater than 2 are removed. At the same time, the spectral ranges of near infrared, mid-infrared and terahertz are corrected, such as near infrared, as shown in Figure 2 As shown, since the collected near-infrared spectrum is greater than 10000 cm -1 and less than 4000cm-1 The signal-to-noise ratio is low, so the corrected near-infrared spectrum range is 10000-4000cm -1 Similarly, the corrected mid-infrared spectrum range is 4000-400cm -1 , the terahertz band range is 333-3.3cm -1 .

[0044] S4. Set the model's convolution layer to 3 layers, the convolution kernel size to 3, the number of filters to 128, 256, and 512, the step size to 2, the loss rate, the learning rate, the number of training times to 1000, and the ratio of the training set to the test set to 7:3. The model will randomly reselect the training set and test set for each training.

[0045] S5. Figure 5-10 As shown, the characteristic spectra after feature recombinant extraction were obtained from the near-infrared spectrum, mid-infrared spectrum and terahertz band of Chuanxiong slices from 6 production areas.

[0046] S6. Figure 3 As shown, the horizontal axis represents the number of runs, and the vertical axis represents the accuracy of the validation set. The accuracy of the model increases with the increase in the number of model training times, as shown in Figure 4 As shown, the vertical axis represents the actual origin of the Chuanxiong slices in the validation set, the horizontal axis represents the origin of the Chuanxiong slices judged by the model, and the diagonal numbers represent the number of correct judgments of the origin. When the number of training times is 1000, the validation set is 90, the number of correct judgments in the validation set is 86, and the accuracy of the model is 95.56%.

[0047] The above examples demonstrate that the accuracy of the identification method of the present application for determining the origin of authentic medicinal materials is higher than 95%, thereby improving the identification accuracy.

[0048] The above embodiments are only used to illustrate the technical solutions of the present invention, but the scope of protection of the present invention is not limited thereto. The embodiments of the present invention and the features in the embodiments can be arbitrarily combined with each other without conflict. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present invention, which should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for identifying the origin of authentic medicinal materials based on multimodal molecular spectral feature recombination, characterized in that: The following steps are involved: Pre-processing the target Chinese medicinal materials to make them into powder or flake form, wherein the target Chinese medicinal materials are produced from no less than 4 places of origin; The near-infrared, mid-infrared and terahertz spectra of the crushed target Chinese medicinal materials were collected. The abnormal spectra were eliminated using the Mahalanobis distance threshold, and then the spectral ranges of the near-infrared, mid-infrared and terahertz were corrected by judging the spectral signal-to-noise ratio. The corrected spectra were trained using a convolutional neural network model, and the features of each spectrum were extracted and reorganized respectively. The reorganized features were visualized using class activation mapping. Build models and evaluate and debug model performance.

2. The identification method according to claim 1, characterized in that: The crushed particle size is through a 200 mesh sieve.

3. The identification method according to claim 1, characterized in that: In the training of the convolutional neural network model, there are 3 convolution layers, the convolution kernel size is 3, the number of filters is 128, 256, and 512 respectively; and the step size is 2.

4. The identification method according to claim 1, characterized in that: In the training using the convolutional neural network model, the number of training times is 800-2000 times, and the ratio of the training set to the test set is 8:2 or 7:3.

Images