Angelica sinensis adulteration quantitative detection method based on terahertz spectrum and data fusion

Abstract

The invention discloses an angelica sinensis adulteration quantitative detection method based on terahertz spectrum and data fusion, which comprises the following steps: (1) taking radix angelicae pubescentis powder as an adulteration substance, doping radix angelicae pubescentis powder with different concentrations into angelica sinensis powder, and mixing to prepare samples with different adulteration amounts; (2) putting each sample into a terahertz system, and measuring each sample to obtain terahertz absorption spectrum and time-domain spectrum information of each sample; (3) carrying out feature extraction on the terahertz absorption spectrum information of different samples; (4) realizing conversion from terahertz time-domain spectrums of different samples to images by using a Grubm angle difference field, and extracting image feature information; and (5) fusing the terahertz absorption spectrum and the feature information of the Grubrum angle difference field image by adopting a feature level data fusion strategy so as to construct an angelica sinensis quantitative adulteration detection model. According to the method, an efficient and reliable solution is provided for Chinese angelica adulteration detection by utilizing the multi-modal feature information.

Description

Technical Field

[0001] The present invention relates to the technical field of chemical detection, and in particular to a quantitative detection method of angelica adulteration based on terahertz spectroscopy and data fusion. Background Art

[0002] Chinese herbal medicine is an important part of TCM diagnosis and treatment, and is also a key area in drug research. The quality of Chinese herbal medicine directly affects its medical effect in clinical application, so strict and effective quality control and evaluation of Chinese herbal medicine has always been a key hotspot in the medical field. Angelica and Duhuo in traditional Chinese herbal medicine both belong to the Umbelliferae family, and their physical properties are extremely similar. Angelica is used as an effective conditioning agent for the treatment of gynecological diseases and a derived dietary supplement. It is widely used to relieve symptoms such as irregular menstruation and dysmenorrhea in women. Duhuo is usually used to treat rheumatism and joint pain. Angelica has a significant demand in the international pharmaceutical market due to its unique medicinal value, and its price is much higher than Duhuo, which makes Angelica have a high economic value in the market. Unscrupulous merchants usually take advantage of the morphological similarities between the two, mix or replace Angelica with Duhuo for sale in exchange for high profits, which will not only cause a serious crisis of trust in the pharmaceutical industry, but also bring unpredictable safety hazards and health problems to patients. Therefore, it is necessary to conduct pure product testing of Angelica and ensure its effectiveness.

[0003] Conducting adulteration detection in drugs is an important means to maintain the order of the pharmaceutical market. The technologies commonly used for adulteration identification include microscopy, high performance liquid chromatography, gas chromatography-mass spectrometry, immunochromatography, etc. These technologies are not only time-consuming and laborious when applied, but more importantly, they will cause damage to the samples to be tested. Therefore, the development of a fast, reliable, non-destructive technology is of great practical significance for adulteration detection.

[0004] Terahertz waves (0.1-10THz) are a unique electromagnetic radiation between microwaves and infrared waves, with a wavelength of about 3mm to 30μm. In recent years, scientific breakthroughs in radiation sources and key equipment have promoted the stable development of terahertz technology, enabling the effective development of electromagnetic radiation in this special frequency band. The resulting terahertz time-domain spectroscopy is one of the most eye-catching fingerprint spectroscopy technologies, and has shown significant advantages in the detection of condensed phase materials. Many weak interactions and lattice vibrations between molecules can be captured in the terahertz frequency band, and molecular dynamics can be used to effectively analyze the vibration mechanism behind the response, thereby predicting and judging the target molecules. Therefore, terahertz spectroscopy technology is widely used in the fields of agricultural product quality testing, food safety control, etc. However, there is currently no research on the use of terahertz spectroscopy technology for angelica adulteration detection.

[0005] As an emerging analysis strategy, data fusion has shown its unique advantages in the field of spectral applications, and its momentum for improving spectral interpretation capabilities is gradually emerging. There are certain differences in the evolution patterns of substances under different spectral technologies. Therefore, the material information captured by different spectral technologies may be complementary, which provides new data support for solving complex problems and further improves the accuracy of analyzing the composition or characteristics of substances. Spectral technology has also shown extremely high application value in the integration with other detection methods, such as combining electronic nose (E-nose), electronic tongue (E-tongue), machine vision (MV), nuclear magnetic resonance (NMR), etc. However, the above fusion strategies all involve the intervention of other detection technologies, which directly increases the complexity of experimental design and the difficulty of execution, requires effective collaboration of professionals, and increases the cost of detection. Therefore, how to enhance the relevance of spectral information and improve data quality and application effect is a key issue in overcoming the limitations of single spectral technology. Summary of the invention

[0006] In view of the above situation, the purpose of the present invention is to provide a quantitative detection method for adulteration of Angelica sinensis based on terahertz spectroscopy and data fusion. The innovative modeling strategy based on the fusion of terahertz spectroscopy and Gram angle difference field image features can effectively improve the prediction accuracy and stability of the model. By utilizing multimodal feature information, an efficient and reliable solution is provided for the detection of adulteration of Angelica sinensis.

[0007] A quantitative detection method for angelica adulteration based on terahertz spectroscopy and data fusion, comprising the following steps:

[0008] (1) Using angelica sinensis powder as an adulterator, different concentrations of angelica sinensis powder were added to angelica sinensis powder to prepare samples with different adulteration amounts, and the amount of angelica sinensis powder added ranged from 0% to 100%;

[0009] (2) placing each sample into a terahertz system, measuring each sample, and obtaining terahertz absorption spectrum and time domain spectrum information of each sample;

[0010] (3) Extract features of terahertz absorption spectrum information of different samples;

[0011] (4) Using the Gram angle difference field to realize the conversion of terahertz time-domain spectra of different samples into images, and extract the image features;

[0012] (5) A feature-level data fusion strategy was adopted to fuse the characteristic information of the terahertz absorption spectrum and the Gram angle field image, thereby constructing a quantitative adulteration detection model for Angelica sinensis.

[0013] The above-mentioned method for quantitative detection of adulteration in Angelica sinensis based on terahertz spectroscopy and data fusion, wherein the step (1) specifically comprises:

[0014] Before preparing the samples, Angelica sinensis and Angelica dahurica were placed in a constant temperature box at 50°C for dry storage. Before making tablets, the medicine crushed by the grinder was poured into a mortar and fully ground, and the ground powder was passed through a 100-mesh sieve. High-density polyethylene powder (PE) was used as a binder in the experiment. The drug powder and PE powder were placed in a centrifuge tube at a ratio of 2:3, and were fully mixed by a vortex mixer before tableting. Each sample was pressed at a pressure of 12MPa for one minute, and the sample thickness was about 1.2mm. The experiment used Angelica dahurica powder and Angelica sinensis powder to prepare samples with different adulteration ratios. Angelica dahurica powder was used as the adulterant, and a total of 21 adulteration concentration gradients were prepared. The Angelica dahurica powder addition amount ranged from 0% to 100%, with an interval of 5%. Three parallel samples were set for each concentration gradient, totaling 63 mixed samples.

[0015] In the above-mentioned method for quantitative detection of adulteration in Angelica sinensis based on terahertz spectroscopy and data fusion, the step (2) specifically comprises:

[0016] Before the experimental test, the terahertz time-domain spectrometer was fully preheated for 30 minutes, and each measurement spectrum was set to the average value after scanning 1028 times. The scanning frequency range was set to 0.1-7THz. After each sample was placed in the optical cavity, it was waited for 3 minutes before measurement. Two points were measured for each concentration of parallel samples, and each point was measured 10 times. 60 spectral lines were collected for each group of concentrations, and a total of 1260 spectral data were obtained. The absorbance of the sample was measured as the analysis data, and the absorbance A(ω) of the sample was calculated by the following formula:

[0017]

[0018] Among them, A sam (ω) is the sample frequency domain signal amplitude, A ref (ω) is the reference signal amplitude.

[0019] In the above-mentioned method for quantitative detection of adulteration in Angelica sinensis based on terahertz spectroscopy and data fusion, the step (3) specifically comprises:

[0020] The spectral data were first preprocessed using standard normal variable transformation combined with baseline correction. Then, competitive adaptive reweighted sampling, iterative information variable retention and variable iterative space shrinkage methods were used to extract the features of the terahertz absorption spectrum.

[0021] In the above-mentioned method for quantitative detection of adulteration in Angelica sinensis based on terahertz spectroscopy and data fusion, the step (4) specifically comprises:

[0022] The Gram angle difference field is used to encode the terahertz time-domain spectral data within 15-20ps. First, the one-dimensional data is normalized and mapped to the cosine angle in polar coordinates. The timing points are expressed as radial distances r, expressed as follows:

[0023] φ=arccos(x i ), -1≤x i ≤1,x i ∈X

[0024]

[0025] Among them, x i represents the time series, t i represents the timestamp, N is a parameter used to adjust the span of the polar coordinates, and the Gram angle difference field GADF is calculated by the following formula:

[0026]

[0027] The texture features of the GADF image are read from 0°, 45°, 90° and 135° using the gray-level co-occurrence matrix, and the energy, entropy, angular moment and correlation of the image are calculated, and the mean and standard deviation of energy, entropy, angular moment and correlation are output as feature subsets. In addition, the mean, contrast, angular moment and entropy of the image are calculated as feature subsets using gray-level difference statistics.

[0028] (5) Specifically include:

[0029] The feature-level data fusion strategy is adopted to fuse the feature information of the terahertz absorption spectrum and the Gram angle difference field image. A quantitative detection model is established based on the support vector machine regression algorithm. The Kennard-Stone algorithm is used to divide the data into a modeling set and a prediction set in a ratio of 3:1, and this is used as the input data of the SVR model. The penalty factor of the model is set to 4, and the radial basis function parameter is set to 2.5.

[0030] According to the method for quantitative detection of adulteration in Angelica sinensis based on terahertz spectroscopy and data fusion provided by the present invention, terahertz time-domain spectroscopy technology (THz-TDS) is combined with a chemometric method, and Angelica sinensis samples adulterated with Angelica dahurica are taken as research objects. A multimodal feature information fusion method is used to realize quantitative analysis of adulteration in Angelica sinensis. The spectral data with time series characteristics are mapped to a two-dimensional plane through a Grammar angle difference field, an image is created by calculating the angles between points in the sequence, and an image processing algorithm is used to extract image features, which are intermediately fused with spectral features and a regression model is constructed to realize quantitative detection of adulteration in Angelica sinensis. The study found that the CARS-GLCM-GLDS-SVR model established by the fusion method of THz spectrum and Grammar difference field image has excellent performance, and the prediction set correlation coefficient reaches 0.9704. At the same time, a low root mean square error (RMSE=0.0731) and a high residual prediction residual (RPD=4.1429) are achieved, and the accuracy is improved by 8.69% compared with the original spectral model, providing an efficient and reliable solution for Angelica sinensis adulteration detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0032] Figure 1 Terahertz spectra under different optical parameters: (a) absorption spectra of Angelica sinensis and Angelica dahurica; (b) first-order derivative spectrum of Angelica dahurica; (c) time domain spectra of Angelica sinensis and Angelica dahurica;

[0033] Figure 2 is the GADF image of Angelica sinensis;

[0034] Figure 3 It is the GADF image of Angelica dahurica;

[0035] Figure 4 The color difference between adulterated samples with different concentrations and pure Angelica sinensis;

[0036] Figure 5 The color difference heat map of adulterated samples with different concentrations;

[0037] Figure 6 Filter results for CARS-based wavelength variables;

[0038] Figure 7 Filter results for IRIV-based wavelength variables;

[0039] Figure 8 Filter results for VISSA-based wavelength variables;

[0040] Fig. 9The fitting of different models to the data: (a) original spectral model, (b) CARS-SVR model, (c) IRIV-SVR model, (d) VISSA-SVR model, (e) CARS-GLCM-GLDS-SVR model, (f) IRIV-GLCM-GLDS-SVR model, (g) VISSA-GLCM-GLDS-SVR model; (h) fusion model accuracy improvement result; DETAILED DESCRIPTION

[0041] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings. Several embodiments of the present invention are shown in the accompanying drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0042] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0043] The embodiment of the present invention provides a method for quantitatively detecting adulteration of Angelica sinensis based on terahertz spectroscopy and data fusion, comprising the following steps (1) to (5):

[0044] Step (1), using angelica root powder as an adulterator, adding angelica root powder of different concentrations into angelica root powder, and mixing to prepare samples with different adulteration amounts, wherein the amount of angelica root powder added is 0% to 100%.

[0045] In order to eliminate the influence of residual moisture in the sample on the experimental spectrum to the greatest extent, and to dry the experimental materials without damaging their chemical stability, Angelica and Angelicae Pubescentis were placed in a constant temperature box at 50°C for drying and storage before preparing the samples. Before making tablets, the medicine crushed by the pulverizer was poured into a mortar and fully ground, and the ground powder was passed through a 100-mesh sieve to remove larger particles of impurities. In the experiment, high-density polyethylene powder was used as a binder. The drug powder and PE powder were placed in a centrifuge tube at a ratio of 2:3, and the tablets were pressed after being fully mixed by a vortex mixer. Each sample was pressed for one minute under a pressure of 12MPa, and the sample thickness was about 1.2mm. The experiment used Angelicae Pubescentis powder and Angelicae Pubescentis powder to prepare samples with different adulteration ratios. Angelicae Pubescentis powder was used as an adulterant, and a total of 21 adulteration concentration gradients were prepared (the amount of Angelicae Pubescentis powder was from 0% to 100%, with an interval of 5%). Three parallel samples were set for each concentration gradient, totaling 63 mixed samples (each about 200g).

[0046] Step (2), placing each sample into a terahertz system, measuring each sample, and obtaining the terahertz absorption spectrum and time domain spectrum information of each sample.

[0047] The TAS7500 terahertz time-domain spectrometer produced by Advantest of Japan was used to collect data for the experimental samples, and all experimental operations were performed in the transmission module. The test environment meets the temperature of about 25℃±0.5℃, the humidity is less than 10%, and the optical test cavity is continuously swept with flowing dry air throughout the process. In order to obtain more accurate spectral data, the device was fully preheated for 30 minutes before the experimental detection, and each measurement spectrum was set to the average value taken after scanning 1028 times, and the scanning frequency range was set to 0.1-7THz.

[0048] Before the experimental measurement, the optical cavity needs to be fully preheated to ensure that the cavity temperature is around 25℃±0.5℃ and the air humidity in the sample chamber is less than 10%. In a stable and dry test environment, the air in the optical cavity is collected as the measurement background. In order to minimize the disturbance to the cavity environment caused by changing samples, each sample is placed in the optical cavity and waited for 3 minutes before measurement. Two points of parallel samples of each concentration are measured, and each point is measured 10 times. 60 spectral lines are collected for each group of concentration, and a total of 1260 spectral data are obtained. Due to differences in the density of substances, even if the quality is consistent, samples of different concentrations will have different porosities after pressing. The absorbance can effectively reduce the impact of inconsistent sample thickness on the experiment, so the absorbance of the sample is mainly measured as the analysis data. The absorbance A(ω) of the sample is calculated by the following formula:

[0049]

[0050] Among them, Asam (ω) is the sample frequency domain signal amplitude, which is A ref (ω) is the reference signal amplitude.

[0051] Step (3), extracting features from the terahertz absorption spectrum information of different samples.

[0052] Wherein, the step (3) specifically includes:

[0053] Since the sample is formed by pressing powders of different materials, the particles of the materials present different mass, volume, shape and arrangement characteristics at the microscopic level. These microscopic characteristics may change the propagation path of the terahertz wave, and even cause the terahertz wave to scatter and absorb to different degrees when penetrating the sample. Standard normal variable transformation (SNV) is often used to eliminate the differences between samples caused by spectral intensity, and baseline correction is often used to eliminate baseline drift caused by instrument noise or background interference in spectral data. SNV can be used in conjunction with baseline correction to preprocess spectral data.

[0054] Irrelevant information in high-dimensional data usually introduces noise or interference to the model, which may increase the complexity of the model and increase the risk of overfitting. Therefore, feature selection is a key technology to improve model performance. Competitive adaptive reweighted sampling (CARS) dynamically adjusts the weight of features, uses multiple iterations to evaluate the importance of features, gradually selects features with higher weights, and finally outputs the optimal variable set. Iteratively retaining informative variables (IRIV) retains both informative and weak informative variables at each iteration. The bands with strong informative information will continue to drive the model until the uninformative variables and interference variables are completely eliminated from the feature set, and finally the optimal variable set is output through reverse elimination. Variable iterative space shrinkage approach (VISSA) optimizes the generated variable subspace by making full use of weighted binary matrix sampling (WBMS), and the variable space can be gradually optimized as it continues to shrink, thereby calculating the feature set with the highest weight. CARS, IRIV, and VISSA can be used to extract features from terahertz absorption spectra.

[0055] Step (4), using the Gram angle difference field to realize the conversion of the terahertz time domain spectrum into an image, and extracting the image feature information.

[0056] The Gram angle difference field is an effective method for encoding one-dimensional data with time series characteristics into a two-dimensional image. It calculates the cosine values ​​between space vectors and maps these values ​​to the pixels of the two-dimensional image, thereby obtaining an image that can reflect the dynamic and periodic characteristics of the time series data. The present invention uses the Gram angle difference field to encode the terahertz time domain spectroscopy data within 15-20ps, with a total of 2501 data points in this range. First, the one-dimensional data is normalized and mapped to the cosine angle in polar coordinates. The timing points are expressed as radial distances r, expressed as follows:

[0057] φ=arccos(x i ), -1≤x i ≤1,x i ∈X

[0058]

[0059] Among them, x i represents the time series, t i represents the timestamp, N is a parameter used to adjust the span of the polar coordinates, and the Gram angle difference field GADF is calculated by the following formula:

[0060]

[0061] Texture is usually used to describe the spatial distribution and relationship of pixel grayscale in an image, and GADF processing realizes the mapping of one-dimensional spectral data to two-dimensional image pixels, and the evolution pattern of time series signals can be reflected in the image structure. Therefore, the grayscale processing algorithm can be used to capture features from GADF images. Gray-level co-occurrence matrix and gray-level difference statistics are used to describe the spatial relationship of pixel grayscale values ​​in an image and the difference of grayscale values ​​between pixel pairs, respectively. The present invention uses GLCM to read the texture features of GADF images from 0°, 45°, 90° and 135° in sequence, calculates the energy (Energy), entropy (Entropy), angular moment (ASM) and correlation (Correlation) of the image, and outputs the mean and standard deviation of energy, entropy, angular moment and correlation in sequence as feature subsets.

[0062] In addition, GLDS can effectively reveal the degree of closeness of the mutual changes of adjacent pixels in a local area, generate a grayscale difference image by calculating the difference between the grayscale values ​​of pixels, establish a grayscale difference histogram based on these differences, and then calculate various statistical features of the image. The present invention uses GLDS to sequentially calculate the mean, contrast, angular second moment (ASM) and entropy of the image as feature subsets.

[0063] Step (5) adopts a feature-level data fusion strategy to fuse the feature information of the terahertz absorption spectrum and the Gram angle difference field image, thereby constructing a quantitative adulteration detection model for Angelica sinensis.

[0064] Traditional signal processing methods may not be able to fully capture the potential periodicity and trend in time domain signals, resulting in the difficulty in effectively extracting and utilizing key variables in the data. Therefore, the present invention uses GADF transformation to map spectral information with time series characteristics to two-dimensional space, and intuitively displays the differences between different samples through the grayscale changes between image pixels. Compared with the traditional fusion strategy based on different optical parameter spectra and time domain spectra, the use of GADF images shows the application potential of spectral features in different forms of expression, providing a new idea for the construction of fusion models.

[0065] The color difference analysis of samples with different adulteration levels was performed using a colorimeter (CR-10, KONICA MINOLTA, Tokyo, Japan). The color of pure angelica powder was measured as the standard color. The CIELAB color space theory adopted by the International Commission on Illumination in 1976 is considered to be the model closest to human color perception. According to the CIELAB color space theory, the L values ​​of different adulterated samples were calculated in turn. * 、a * 、b * Three color coordinates. The present invention uses the color difference value (dE) to represent the degree of difference between two color samples, and the calculation formula is as follows:

[0066]

[0067] in, Represents the brightness of pure angelica powder. Represents the color range of pure angelica powder from green to red. Represents the color range of pure angelica powder from blue to yellow.

[0068] Support vector regression (SVR) is a derivative algorithm based on support vector machine (SVM). SVR maps the input data to a high-dimensional feature space and calculates an optimal segmentation hyperplane in the space so that as many training samples as possible fall within the boundary bands on both sides of the plane while maximizing the width of the boundary bands. Therefore, the SVR algorithm has significant advantages in improving the generalization ability of the model and processing unbalanced distribution data. The present invention uses the Kennard-Stone (KS) algorithm to divide the data into a modeling set and a prediction set in a ratio of 3:1, and uses this as the input data of the SVR model. The penalty factor c of the model is set to 4, and the radial basis function parameter g is set to 2.5.

[0069] The present invention evaluates the model by calculating the correlation coefficient (r), root mean square error (RMSE), residual prediction residual (RPD) and ΔE (|RMSEP-RMSEC|). High-performance models generally have higher r and RPD (it is generally believed that the model built when RPD>2 has higher reliability and can be used for model analysis), and lower RMSE and ΔE values.

[0070] The experimental results and analysis of the present invention are as follows:

[0071] Figure 1 is the spectrum of each sample under different parameter settings, where Figure 1 (a) shows the absorption spectra of Angelica and Angelica dahurica in the range of 0.5-2.5 THz. The absorbance of both shows an increasing trend with the increase of terahertz frequency, and the absorbance of pure Angelica dahurica is always higher than that of pure Angelica. This difference is mainly due to the differences in the molecular composition and intermolecular interactions of the two substances. The pure Angelica sample has an obvious absorption peak at 1.88 THz, and a weak protrusion can be observed in the pure Angelica dahurica sample at 2.05 THz, but no obvious absorption peak morphology is formed. The present invention performs first-order derivative processing on the spectral data of Angelica dahurica to enhance the recognition of the absorption peak in the spectrum. The processing results are shown in the figure. Figure 1As shown in (b), an obvious trough appears in the first-order derivative image at 2.05 THz. According to the definition of the first-order derivative, the trough in the spectral data corresponds to the peak in the original spectrum. Therefore, although the absorption peak at 2.05 THz is not obvious in the original spectral image, the existence of this trough indicates that Angelica dahurica has potential absorption characteristics at this frequency. Studies have shown that ferulic acid is one of the main medicinal ingredients of Angelica sinensis. It has been proven to have antioxidant, antibacterial and anti-inflammatory effects and has great potential in the food and pharmaceutical fields. Relevant researchers discussed and verified the vibration mode of ferulic acid in the terahertz frequency band through density functional theory calculations and experiments. The results showed that ferulic acid will produce an absorption peak at 1.90 THz in the collective vibration mode between molecules. Coumarin is one of the main medicinal ingredients of Angelica dahurica, which controls the progression of rheumatoid arthritis by inhibiting the proliferation of synovial fibroblasts. In addition, it has been confirmed in existing research reports that coumarin has an absorption peak at 2.06 THz excited by intermolecular vibration. Therefore, the absorption peaks at 1.88 THz and 2.05 THz may be excited by the intermolecular vibration of ferulic acid and coumarin, respectively.

[0072] Figure 1 (c) shows the time domain signals of Angelica and Angelica dahurica in the range of 15-20ps. The response of pure Angelica dahurica to terahertz waves is faster, but the amplitude is slightly lower than that of pure Angelica dahurica. This may be because Angelica dahurica has strong absorption characteristics for terahertz waves in the low frequency band. When it is close to 20ps, the time domain signal of Angelica dahurica still shows weak oscillation, indicating that its response to terahertz radiation has a certain continuity. However, Angelica dahurica has almost no signal in this time domain range, and the response has been significantly attenuated in this range.

[0073] Figure 2 and Figure 3 The GADF images encoded by the time-domain spectral data of the Angelica sinensis sample and the Angelica dahurica sample are shown respectively. Although the overall structure of the two images is similar, there are significant differences in color intensity and spatial position. Especially in the first quadrant (upper right), in the Angelica sinensis image, more pixel values ​​show stronger hue intensity, while the Angelica dahurica image shows relatively low chroma values ​​at the same position. In addition, the main feature area of ​​the Angelica sinensis image is biased towards the central part of the image, while the main feature area of ​​the Angelica dahurica image is biased towards the lower left corner, and the degree of deviation is large. The essence of the GADF algorithm is to map the time series signal to the image pixel through a specific conversion process. Each pixel contains the characteristic relationship of the signal at different time points. Therefore, the changes in the color intensity and spatial position of the main features in the GADF image can reflect the subtle differences in the sample time domain signal, and these differences will provide favorable information support for the image algorithm in the feature extraction stage.

[0074] Figure 4The color difference between adulterated samples with different concentrations and pure Angelica sinensis is shown. As the adulteration amount increases, the color difference value gradually becomes larger. Based on the extensive research and application accumulation in the field of color science, it is generally considered that dE≥2 is the threshold at which the human eye can significantly perceive the color difference, and when dE≤1, the color difference is difficult to distinguish by the human eye. Therefore, it is difficult to directly visually identify whether the drug is pure Angelica sinensis powder at low adulteration concentrations. At medium adulteration concentrations (1<dE<2), to accurately determine whether Angelica sinensis is adulterated, professional personnel are needed for differentiation. The color difference coordinates show that as the incorporation of Heracleum hemsleyanum powder increases, the brightness difference (dL) significantly decreases, indicating that the overall sample becomes darker as the adulteration ratio increases. The red-green color difference (da) increases, indicating that the adulterated sample gradually tends to be red. The yellow-blue color difference (db) increases, showing that the adulterated sample gradually tends to be yellow.

[0075] Figure 5 is the color difference heat map of adulterated samples with different concentrations. It can be seen that the change in color difference is the result of the combined action of colors. The differentiation of medium and low concentration adulterated samples is difficult to judge by the naked eye.

[0076] Figure 6 is the wavelength variable screening result based on CARS. CARS enhances the robustness of model validation and feature selection through Monte Carlo sampling technology. In this invention, 1000 times of Monte Carlo cross-validation are used to determine the optimal features and principal components. From Figure 6 it can be obtained that after 300 samplings, the number of wavelength decreases sharply, and the remaining spectral variables will be further refined. When the number of samplings reaches 610 times, the root mean square error of cross-validation (RMSECV) reaches the minimum value of 0.093, indicating that the model has the best prediction performance in cross-validation at this time. After screening, there are 13 feature variables, accounting for 4.94% of the total number of variables.

[0077] The IRIV algorithm uses the exponential decay function (EDF) and binary matrix sampling (BMS) technology to optimize spectral feature selection. A total of 500 times of binary matrix sampling are carried out, and 30 times of exponential decay function runs are used to iteratively optimize feature selection, and the model performance is evaluated in each iteration. Finally, while retaining the optimal number of features, the 5-fold cross-validation method is used to further evaluate the generalization ability and performance of the model to determine the best feature subset. Figure 7 shows the variation relationship between the number of features and the number of iterations. After 6 iterations, the number of features decreases from 263 to 19, and then the RMSECV is calculated for the remaining feature variable combinations through the backward elimination strategy. When the RMSECV is the smallest, the backward elimination is completed, and finally 17 feature variables are determined, accounting for 6.46% of the total number of variables.

[0078] In VISSA feature selection, the variable subset generated by weighted binary sampling (WBMS) was set to 1000, the initial sampling weight was 0.5, the proportion of the subset model was 5%, and the 5-fold cross-validation method was used to establish the PLS model. The optimal feature subset was determined based on the minimum value of RMSECV. Figure 8 The relationship between the number of features and RMSECV is shown. As the number of feature variables increases, the value of RMSECV shows a trend of first a sharp drop, then a slight increase, and finally a stable trend. VISSA finally selects 10 feature variables at the minimum RMSECV value. At this time, RMSECV is 0.134, and the feature subset accounts for 3.8% of the total number of variables. Table 1 shows the wavelength screening of different feature extraction strategies.

[0079] Table 1 Results of optimal feature variables selected by CARS, IRIV and VISSA methods

[0080]

[0081]

[0082] Table 2 shows the modeling effects of the full spectrum and spectral feature set. It is worth noting that the RPD of all models is greater than 2, which proves that the SVR algorithm has strong adaptability and robustness, and can maintain stable performance on data sets of different qualities. The results show that feature screening can effectively remove redundant information and noise in spectral data, and improve model accuracy by retaining key information. The regression models constructed by spectral data processed by CARS, IRIV, and VISSA are better than the models constructed by original spectral data in performance. Among them, the CARS-SVR model has the best comprehensive performance, with a correlation coefficient of 0.9315, which is 0.048 higher than the original spectral model. At the same time, compared with IRIV-SVR and VISSA-SVR, the RMSE (0.1101) and ΔE (0.0041) of the CARS-SVR model are lower, indicating that the model has obvious advantages in prediction accuracy and error control. At this time, the spectral data input to the model only accounts for 4.94% of the total data.

[0083] Table 2 Modeling results of full spectrum and optimal spectrum variables

[0084]

[0085] Fig. 9 It shows how well different models fit the data. Fig. 9 (a) is the model fitting result diagram based on full spectrum data modeling. Fig. 9(b), (c), and (d) show the model fitting results of modeling with different feature subsets. This paper further evaluates the model fitting effect and prediction ability by calculating the 95% confidence interval and 95% prediction interval of different models. The 95% confidence interval is used to represent the credible interval of the regression line estimate, and the 95% prediction interval represents the prediction range of the data point. In this study, whether it is full spectrum or characteristic spectrum modeling, the model shows a narrow 95% confidence interval, and the 95% prediction interval performs well, which further confirms that the SVR model has high accuracy and stability in parameter estimation. Among them, the residual of the full spectrum model is relatively large in the low and high concentration ranges, while the residual is relatively small in the intermediate concentration range. This may be because there is a strong boundary effect in the extreme concentration area, which makes the spectral characteristics in this area not completely match the linear relationship assumed by the model, thereby affecting the model's prediction accuracy for extreme values ​​and increasing the residual. In the model constructed using the feature subset extracted by CARS, IRIV, and VISSA, some concentration residuals are large, but the overall residuals are more uniform compared to full spectrum modeling. Even after feature extraction, the model may still not be able to fully capture the nonlinear effects at certain concentrations. Samples of a certain concentration may be more sensitive to some weak but important spectral features, but these effects may be weakened or simplified in data dimensionality reduction, resulting in large prediction errors for some concentrations.

[0086] Table 3 shows the modeling effect based on the fusion of THz spectrum and GADF image data. The present invention uses different feature extraction strategies to combine GLCM and GLDS image processing algorithms in turn to construct a fusion model based on spectral features and image features. The results show that the data set after fusion of image features shows significant advantages in the model. Compared with the single feature spectrum model, the correlation of all fusion models has been greatly improved, the RMSE value has been effectively reduced, and the RPD value has been kept above 2, and the reliability and superiority of the fusion model are obvious. It is worth noting that compared with the spectral feature fusion of GLCM or GLDS features alone, the performance of the model that simultaneously fuses GLCM and GLDS features is significantly better than the model of single feature fusion. It shows that the multimodal feature fusion strategy can more effectively utilize the important information in the data. Through the complementarity of GLCM and GLDS, the texture details of the GADF image are fully retained, so that the model can more accurately capture complex nonlinear relationships, thereby enhancing the prediction ability of the model. Among them, the comprehensive performance of the CARS-GLCM-GLDS-SVR model is the most outstanding, with a correlation of 0.9704, RMSE and RPD of 0.0731 and 4.1429 respectively. At the same time, the accuracy of the CARS-SVR model was improved by 3.89% after fusing the GLCM and GLDS features, which was significantly higher than that of the IRIV-SVR and VISSA-SVR models.

[0087] Table 3 Modeling results of fusion of terahertz spectrum and GADF image data

[0088]

[0089] Fig. 9 (e), (f), and (g) show the model fitting results of the simultaneous fusion of GLCM and GLDS features with different feature extraction strategies. The deep fusion of spectral features and image features effectively reduces the residual of the regression model, and the 95% prediction interval of the model is significantly reduced. Although modeling with a feature subset of the spectrum can effectively improve the relevance of the model, the 95% prediction interval of the characteristic spectrum model has not been significantly reduced compared with the fitting results of the original spectral data, so the prediction ability of the model is still limited. However, after the simultaneous fusion of GLCM and GLDS features, the 95% prediction interval of the model is significantly narrowed, indicating that the introduced image features provide effective discriminant information for the judgment of the target concentration, so that the model can more accurately capture the laws in the data, thereby increasing the confidence of the predicted value and reducing the residual. The residuals of the IRIV-GLCM-GLDS-SVR model and the VISSA-GLCM-GLDS-SVR model are obvious in some concentration areas, and the CARS-GLCM-GLDS-SVR model has the highest fitting performance.

[0090] like Fig. 9As shown in (h), the fusion model based on feature extraction of CARS, IRIV and VISSA improved the accuracy by 8.69%, 6.61% and 6.62% respectively compared with the original spectral model, among which the CARS model had the largest improvement, with the RMSE value of the prediction set reduced from 0.7806 to 0.0731 and the RPD value increased from 2.1387 to 4.1429. It can be seen that the use of appropriate feature selection algorithms can effectively improve the prediction accuracy and generalization ability of the model for the target variable. In summary, according to the quantitative detection method of angelica adulteration based on terahertz spectroscopy and data fusion provided by the present invention, terahertz time-domain spectroscopy technology (THz-TDS) is combined with a chemometric method, and angelica samples adulterated with Angelica dahurica are taken as the research object. A multimodal feature information fusion method is used to realize the quantitative analysis of angelica adulteration, and the spectral data with time series characteristics are mapped to a two-dimensional plane through the Gram angle difference field, and an image is created by calculating the angle between each point in the sequence, and an image processing algorithm is used to extract image features, which are intermediately fused with spectral features and a regression model is constructed to realize the quantitative detection of angelica adulteration. The study found that the CARS-GLCM-GLDS-SVR model established by the fusion method of THz spectrum and Gram difference field image has excellent performance, and the prediction set correlation coefficient reaches 0.9704. At the same time, a low root mean square error (RMSE=0.0731) and a high residual prediction residual (RPD=4.1429) are achieved, and compared with the original spectral model, the accuracy is improved by 8.69%, which provides an efficient and reliable solution for angelica adulteration detection.

[0091] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the patent of the present invention. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A quantitative detection method for adulteration of Angelica sinensis based on terahertz spectroscopy and data fusion, characterized in that: The following steps are involved: (1) Using angelica root powder as an adulterator, different concentrations of angelica root powder were added to angelica root powder to prepare samples with different adulteration amounts, with the angelica root powder addition amount ranging from 0% to 100%; (2) placing each sample into a terahertz system, measuring each sample, and obtaining the terahertz absorption spectrum and time domain spectrum information of each sample; (3) Feature extraction of terahertz absorption spectrum information of different samples; (4) Using the Gram angle difference field to realize the conversion of terahertz time-domain spectra of different samples into images, and extract the image feature information; (5) A feature-level data fusion strategy was adopted to fuse the characteristic information of the terahertz absorption spectrum and the Gram angle difference field image, thereby constructing a quantitative adulteration detection model for Angelica sinensis.

2. The method for quantitative detection of angelica adulteration based on terahertz spectroscopy and data fusion according to claim 1, characterized in that: The step (1) specifically includes: Before preparing the samples, Angelica sinensis and Duhuo were placed in a constant temperature box at 50°C for drying and storage. Before making the tablets, the medicine crushed by the grinder was poured into a mortar and fully ground, and the ground powder was passed through a 100-mesh sieve. High-density polyethylene powder (PE) was used as a binder in the experiment. The drug powder and PE powder were placed in a centrifuge tube at a ratio of 2:3, and were fully mixed by a vortex mixer before tableting. Each sample was pressed at a pressure of 12MPa for one minute, and the sample thickness was about 1.2mm. The experiment used Duhuo powder and Angelica sinensis powder to prepare samples with different adulteration ratios. Duhuo powder was used as the adulterant, and a total of 21 adulteration concentration gradients were prepared. The Duhuo powder addition amount ranged from 0% to 100%, with an interval of 5%. Three parallel samples were set for each concentration gradient, totaling 63 mixed samples.

3. The method for quantitative detection of adulteration in Angelica sinensis based on terahertz spectroscopy and data fusion according to claim 2, characterized in that: The step (2) specifically includes: Before the experimental test, the terahertz time-domain spectrometer was fully preheated for 30 minutes, and each measurement spectrum was set to the average value after scanning 1028 times. The scanning frequency range was set to 0.1-7THz. After each sample was placed in the optical cavity, it was waited for 3 minutes before measurement. Two points of parallel samples of each concentration were measured, and each point was measured 10 times. 60 spectral lines were collected for each group of concentrations, and a total of 1260 spectral data were obtained. The absorbance of the sample was measured as the analytical data. The absorbance A(ω) of the sample was calculated by the following formula: Among them, A sam (ω) is the sample frequency domain signal amplitude, A ref (ω) is the reference signal amplitude.

4. The method for quantitative detection of adulteration in Angelica sinensis based on terahertz spectroscopy and data fusion according to claim 1, characterized in that: The step (3) specifically includes: The spectral data were first preprocessed using standard normal variable transformation combined with baseline correction. Then, competitive adaptive reweighted sampling, iterative information variable retention and variable iterative space shrinkage were used to extract the features of the terahertz absorption spectrum.

5. The method for quantitative detection of adulteration in Angelica sinensis based on terahertz spectroscopy and data fusion according to claim 1, characterized in that: The step (4) specifically includes: The Gram angle difference field is used to encode the terahertz time domain spectroscopy data within 15-20ps. First, the one-dimensional data is normalized and mapped to the cosine angle in polar coordinates. The time points are expressed as radial distance r, which is expressed as follows: φ=arccos(x i ),-1≤x i ≤1,x i ∈X Among them, x i represents the time series, t i represents the timestamp, N is a parameter used to adjust the span of the polar coordinates, and the Gram angle difference field GADF is calculated by the following formula: The gray-level co-occurrence matrix is ​​used to read the texture features of the GADF image at 0°, 45°, 90°, and 135°. The energy, entropy, angular moment, and correlation of the image are calculated, and the mean and standard deviation of energy, entropy, angular moment, and correlation are output as feature subsets. In addition, the mean, contrast, angular moment, and entropy of the image are calculated as feature subsets using gray-level difference statistics.

6. The method for quantitative detection of adulteration in Angelica sinensis based on terahertz spectroscopy and data fusion according to claim 1, characterized in that: The step (5) specifically includes: adopting a feature-level data fusion strategy to fuse the feature information of the terahertz absorption spectrum and the Gram angle difference field image. A quantitative detection model is established using a support vector machine regression algorithm, wherein the data is divided into a modeling set and a prediction set in a ratio of 3:1 using the Kennard-Stone algorithm, and the data is used as the input data of the SVR model. The penalty factor of the model is set to 4, and the radial basis function parameter is set to 2.5.

Images