Angelicae sinensis radix adulteration quantitative detection method based on terahertz spectroscopy and data fusion

Abstract

Disclosed in the present invention is an Angelicae sinensis radix adulteration quantitative detection method based on terahertz spectroscopy and data fusion. The method comprises the following steps: (1) using Angelicae pubescentis radix powder as an adulterant, doping Angelicae pubescentis radix powder of different concentrations into Angelicae sinensis radix powder, and performing mixing to prepare samples having different adulteration levels; (2) putting each sample into a terahertz system, and measuring each sample to obtain terahertz absorption spectral information and time-domain spectral information of each sample; (3) performing feature extraction on the terahertz absorption spectral information of different samples; (4) using a Gramian angular difference field to convert terahertz time-domain spectra of different samples into images, and extracting image feature information; and (5) using a feature-level data fusion strategy to fuse feature information of terahertz absorption spectra with feature information of Gramian angular difference field images, so as to construct an Angelicae sinensis radix adulteration quantitative detection model. The present invention provides an efficient and reliable solution for Angelicae sinensis radix adulteration detection by using multi-modal feature information.

Description

A method for detecting the doping level of Angelica sinensis based on terahertz spectroscopy and data fusion Technical Field

[0001] This invention relates to the field of chemical detection technology, and in particular to a method for detecting the doping amount of Angelica sinensis based on terahertz spectroscopy and data fusion. Background Technology

[0002] Traditional Chinese medicine (TCM) herbs are an important component of TCM diagnosis and treatment, and a key area of ​​drug research. The quality of TCM herbs directly affects their clinical efficacy; therefore, strict and effective quality control and evaluation of TCM herbs has always been a crucial focus in the pharmaceutical field. Angelica sinensis and Angelica pubescens, both belonging to the Apiaceae family, share highly similar physical properties. Angelica sinensis is widely used as an effective remedy for gynecological diseases and as a dietary supplement to alleviate symptoms such as irregular menstruation and dysmenorrhea in women. Angelica pubescens is typically used to treat rheumatism and joint pain. Due to its unique medicinal value, Angelica sinensis has significant demand in the international pharmaceutical market, and its price is much higher than that of Angelica pubescens, giving it high economic value. Unscrupulous merchants often exploit the morphological similarity between the two, mixing or substituting Angelica sinensis and Angelica pubescens for each other to reap high profits. This not only causes a serious crisis of trust in the pharmaceutical industry but also poses unpredictable safety and health risks to patients. Therefore, it is necessary to conduct pure product testing of Angelica sinensis and ensure its effectiveness.

[0003] Detecting adulteration in pharmaceuticals is a crucial means of maintaining order in the pharmaceutical market. Commonly used techniques for adulteration identification include microscopy, high-performance liquid chromatography (HPLC), gas chromatography-mass spectrometry (GC-MS), and immunochromatography. However, these techniques are not only time-consuming and labor-intensive, but more importantly, they can damage the samples. Therefore, developing a rapid, reliable, and non-destructive technique for adulteration detection is of significant practical importance.

[0004] Terahertz waves (0.1-10 THz) are a unique segment of electromagnetic radiation located between microwaves and infrared waves, with wavelengths ranging from approximately 3 mm to 30 μm. In recent years, scientific breakthroughs in radiation sources and key equipment have promoted the stable development of terahertz technology, enabling the effective exploitation of electromagnetic radiation in this special frequency band. Terahertz time-domain spectroscopy, which has emerged as a result, is one of the most compelling fingerprint spectroscopy techniques, demonstrating significant advantages in the detection of condensed phase substances. Many weak interactions and lattice vibrations between molecules can be captured in the terahertz frequency band. Molecular dynamics can be used to effectively resolve the vibrational mechanisms behind the responses, thereby predicting and identifying target molecules. Therefore, terahertz spectroscopy is widely used in agricultural product quality testing and food safety control. However, there is currently no research using terahertz spectroscopy for the detection of adulteration in Angelica sinensis (Dang Gui).

[0005] Data fusion, as an emerging analytical strategy, has demonstrated its unique advantages in the field of spectral applications, and its momentum in improving spectral interpretation capabilities is gradually emerging. The evolutionary patterns of matter differ under different spectral techniques; therefore, the information captured using different spectral techniques may be complementary. This provides new data support for solving complex problems and further improves the accuracy of analyzing the composition or properties of substances. Spectroscopic techniques have also shown great application value in fusion with other detection methods, such as combining electronic noses (E-nose), electronic tongues (E-tongue), machine vision (MV), and nuclear magnetic resonance (NMR). However, these fusion strategies all involve the intervention of other detection technologies, directly increasing the complexity of experimental design and execution, requiring effective collaboration among professionals, and increasing detection costs. Therefore, how to enhance the correlation of spectral information and improve data quality and application effectiveness is a key issue in overcoming the limitations of single-spectral techniques. Summary of the Invention

[0006] In view of the above, the purpose of this invention is to provide a method for detecting adulteration in 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, it provides an efficient and reliable solution for detecting adulteration in Angelica sinensis.

[0007] A method for detecting the doping amount of Angelica sinensis based on terahertz spectroscopy and data fusion includes the following steps:

[0008] (1) Using Angelica pubescens powder as an adulterant, different concentrations of Angelica pubescens powder were mixed into Angelica sinensis powder to prepare samples with different adulterant amounts. The amount of Angelica pubescens powder added ranged from 0% to 100%.

[0009] (2) Place each sample into the terahertz system and measure each sample to obtain the terahertz absorption spectrum and time-domain spectrum information of each sample.

[0010] (3) Feature extraction of terahertz absorption spectrum information of different samples;

[0011] (4) The Gram angle difference field is used to convert the terahertz time-domain spectra of different samples into images and extract the image features.

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

[0013] The above-mentioned method for detecting the doping amount of Angelica sinensis based on terahertz spectroscopy and data fusion, wherein step (1) specifically includes:

[0014] Before preparing the samples, Angelica sinensis and Angelica pubescens were dried and stored in a constant temperature oven at 50℃. Before tableting, the pulverized drugs were ground thoroughly in a mortar and the 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 thoroughly mixed by a vortex mixer before tableting. Each sample was pressed at a pressure of 12 MPa for one minute, and the sample thickness was about 1.2 mm. The experiment used Angelica pubescens powder and Angelica sinensis powder to prepare samples with different adulteration ratios. Using Angelica pubescens powder as the adulterant, a total of 21 adulteration concentration gradients were prepared, with the amount of Angelica pubescens powder ranging from 0% to 100% at 5% intervals. Three parallel samples were set for each concentration gradient, for a total of 63 mixed samples.

[0015] The above-mentioned method for detecting the doping amount of Angelica sinensis based on terahertz spectroscopy and data fusion, wherein step (2) specifically includes:

[0016] Before the experiment, the terahertz time-domain spectrometer was preheated for 30 minutes. Each measurement was set to the average of 1028 scans, with a frequency range of 0.1-7 THz. After placing each sample into the optical cavity, a 3-minute interval was allowed before measurement. Two measurements were taken from each parallel sample of each concentration, with each measurement repeated 10 times. Sixty spectral lines were collected from each concentration group, resulting in a total of 1260 spectral data points. The absorbance of the measured samples was used as the analytical data. The absorbance A(ω) of the sample was calculated using the following formula:

[0017] Among them, A sam (ω) represents the amplitude of the sample frequency domain signal, A ref (ω) is the amplitude of the reference signal.

[0018] The above-mentioned method for detecting the doping amount of Angelica sinensis based on terahertz spectroscopy and data fusion, wherein step (3) specifically includes:

[0019] First, standard normal variable transformation combined with baseline correction is used to preprocess the spectral data. Then, competitive adaptive reweighted sampling, iterative retention of information variables, and variable iterative space shrinkage method are used to extract features from the terahertz absorption spectrum.

[0020] The above-mentioned method for detecting the doping amount of Angelica sinensis based on terahertz spectroscopy and data fusion, wherein step (4) specifically includes:

[0021] Image encoding of terahertz time-domain spectral data within the 15-20 ps range was performed using Gram angle difference fields. First, the one-dimensional data was normalized and then mapped to cosine angles in polar coordinates. The time series points are represented by radial distance r, as expressed below: φ=arccos(x i -1≤x i ≤1,x i ∈X

[0022] Where, x i Represents a time series, t i Representing the timestamp, N is a parameter used to adjust the span of the polar coordinates. The Gram angular difference field (GADF) is calculated using the following formula:

[0023] Texture features of the GADF image are read sequentially from 0°, 45°, 90°, and 135° using the gray-level co-occurrence matrix. The image's energy, entropy, arcsecond moment, and correlation are calculated, and the mean and standard deviation of these parameters are output as feature subsets. Furthermore, the image's mean, contrast, arcsecond moment, and entropy are calculated sequentially using gray-level difference statistics as feature subsets.

[0024] (5) Specifically includes:

[0025] A feature-level data fusion strategy was adopted to fuse the feature information of terahertz absorption spectra and Gram angular difference field images. A quantitative detection model was established using a support vector machine-based regression algorithm. The Kennard-Stone algorithm was used to divide the data into a modeling set and a prediction set at a ratio of 3:1, which were then used as the input data for the SVR model. The penalty factor of the model was set to 4, and the radial basis function parameter was set to 2.5.

[0026] The present invention provides a method for detecting adulteration quantification in Angelica sinensis based on terahertz spectroscopy and data fusion. This method employs terahertz time-domain spectroscopy (THz-TDS) combined with chemometrics, using Angelica sinensis samples adulterated with Angelica pubescens as the research object. Through multimodal feature information fusion, adulteration quantification of Angelica sinensis is analyzed. The spectral data with time-series characteristics are mapped to a two-dimensional plane using Gram difference field. An image is created by calculating the angles between points in the sequence, and image processing algorithms are used to extract image features. These features are then fused with spectral features to construct a regression model, thus achieving adulteration quantification of Angelica sinensis. The study found that the CARS-GLCM-GLDS-SVR model established using the fusion method of THz spectroscopy and Gram difference field images exhibits excellent performance, with a prediction set correlation coefficient reaching 0.9704. It also achieves low root mean square error (RMSE = 0.0731) and high residual prediction residual (RPD = 4.1429), and the accuracy is improved by 8.69% compared to the original spectral model. This provides an efficient and reliable solution for detecting adulteration in Angelica sinensis. Attached Figure Description

[0027] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0028] Figure 1 shows the terahertz spectra under different optical parameters: (a) absorption spectra of Angelica sinensis and Angelica pubescens; (b) first derivative spectrum of Angelica pubescens; (c) time-domain spectra of Angelica sinensis and Angelica pubescens.

[0029] Figure 2 shows the GADF image of Angelica sinensis;

[0030] Figure 3 shows the GADF image of Angelica pubescens;

[0031] Figure 4 shows the color difference between adulterated samples of different concentrations and pure Angelica sinensis;

[0032] Figure 5 shows the color difference thermograms of adulterated samples with different concentrations;

[0033] Figure 6 shows the wavelength variable screening results based on CARS;

[0034] Figure 7 shows the wavelength variable selection results based on IRIV;

[0035] Figure 8 shows the wavelength variable screening results based on VISSA;

[0036] Figure 9 shows the fitting results 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) accuracy improvement results of the fusion model. Detailed Implementation

[0037] To make the objectives, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Several embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of the present invention will be thorough and complete.

[0038] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0039] This invention provides a method for detecting the doping amount of Angelica sinensis based on terahertz spectroscopy and data fusion, comprising the following steps (1) to (5):

[0040] Step (1): Using Angelica pubescens powder as an adulterant, different concentrations of Angelica pubescens powder were mixed into Angelica sinensis powder to prepare samples with different adulterant amounts. The amount of Angelica pubescens powder added ranged from 0% to 100%.

[0041] To minimize the impact of residual moisture in the samples on the experimental spectra and to dry the experimental materials without compromising their chemical stability, Angelica sinensis and Angelica pubescens were dried in a constant temperature oven at 50℃ before sample preparation and then stored. Before tableting, the pulverized drugs were thoroughly ground in a mortar and pestle, and the powder was then passed through a 100-mesh sieve to remove larger particles. High-density polyethylene powder was used as a binder in the experiment. The drug powder and PE powder were placed in centrifuge tubes at a ratio of 2:3 and thoroughly mixed using a vortex mixer before tableting. Each sample was pressed at a pressure of 12 MPa for one minute, with a sample thickness of approximately 1.2 mm. Samples with different adulteration ratios were prepared by mixing Angelica pubescens powder and Angelica sinensis powder. Using Angelica pubescens powder as the adulterant, a total of 21 adulteration concentration gradients were prepared (Angelica pubescens powder adulteration amount from 0% to 100%, with 5% intervals). Three parallel samples were set for each concentration gradient, for a total of 63 mixed samples (each approximately 200 g).

[0042] Step (2): Place each sample into the terahertz system and measure each sample to obtain the terahertz absorption spectrum and time-domain spectrum information of each sample.

[0043] Data acquisition of experimental samples was performed using a TAS7500 terahertz time-domain spectrometer manufactured by Advantest Corporation of Japan. All experimental operations were conducted within the transmission module. The test environment met the requirements of a temperature of approximately 25℃ ± 0.5℃ and humidity of less than 10%, with continuous flow of dry air sweeping the optical test cavity throughout the entire process. To obtain more accurate spectral data, the device was preheated for 30 minutes before the experiment, and each measurement spectral line was set to be the average value after 1028 scans, with the scan frequency range set to 0.1-7 THz.

[0044] Before the experiment, the optical cavity must be fully preheated to ensure a cavity temperature of approximately 25℃±0.5℃ and a sample chamber humidity of less than 10%. In a stable and dry testing environment, air from inside the optical cavity is collected as the measurement background. To minimize disturbance to the cavity environment during sample replacement, each sample is placed in the optical cavity and measured after 3 minutes. Two measurements are taken for each concentration of parallel samples, with each measurement repeated 10 times. 60 spectral lines are collected for each concentration group, resulting in a total of 1260 spectral data points. Due to differences in material density, even with consistent mass, samples of different concentrations will exhibit different porosities after compression. Absorbance can effectively reduce the impact of inconsistent sample thickness on the experiment; therefore, the absorbance of the samples is primarily measured as the analytical data. The absorbance A(ω) of the sample is calculated using the following formula:

[0045] Among them, A sam (ω) represents the amplitude of the sample frequency domain signal, denoted as A. ref (ω) is the amplitude of the reference signal.

[0046] Step (3) involves feature extraction of terahertz absorption spectral information from different samples.

[0047] Specifically, step (3) includes:

[0048] Because the samples are formed by pressing powders of different materials, the particles exhibit varying masses, volumes, shapes, and arrangement characteristics at the microscopic level. These microscopic characteristics may alter the propagation path of terahertz waves, and even cause varying degrees of scattering and absorption when the terahertz waves penetrate the sample. Standard normal variable transformation (SNV) is commonly used to eliminate differences in spectral intensity between samples, while baseline correction is commonly used to eliminate baseline drift in spectral data caused by instrument noise or background interference. SNV combined with baseline correction can be used for preprocessing spectral data.

[0049] Irrelevant information in high-dimensional data often introduces noise or interference into the model, potentially increasing model complexity and the risk of overfitting. Therefore, feature selection is a key technique for improving model performance. Competitive adaptive reweighted sampling (CARS) dynamically adjusts feature weights, using multiple iterations to evaluate feature importance and progressively select features with higher weights, ultimately outputting the optimal variable set. Iteratively retaining informative variables (IRIV) retains both highly informative and less informative variables in each iteration. Highly informative bands continue to drive the model until non-informative and interfering variables are completely removed from the feature set, finally outputting the optimal variable set through reverse elimination. The Variable iterative space shrinkage approach (VISSA) fully utilizes weighted binary matrix sampling (WBMS) to optimize the generated variable subspace, and the variable space can be progressively optimized as it shrinks, thereby calculating the set of features with the highest weights. Features can be extracted from terahertz absorption spectra using CARS, IRIV, and VISSA.

[0050] Step (4) uses the Gram angular difference field to realize the conversion of terahertz time-domain spectrum to image and extracts image feature information.

[0051] Gram difference field is an effective method for encoding one-dimensional data with temporal characteristics into two-dimensional images. It calculates the cosine values ​​between spatial vectors and maps these values ​​to pixels in a two-dimensional image, thus obtaining an image that reflects the dynamic and periodic characteristics of the temporal data. This invention utilizes Gram difference field to encode terahertz time-domain spectral data within the range of 15-20 ps, ​​totaling 2501 data points. First, the one-dimensional data is normalized and then mapped to cosine angles in polar coordinates. The time series points are represented by radial distance r, as expressed below: φ=arccos(x i -1≤x i ≤1,x i ∈X

[0052] Where, x i Represents a time series, t i Representing the timestamp, N is a parameter used to adjust the span of the polar coordinates. The Gram angular difference field (GADF) is calculated using the following formula:

[0053] Texture is typically used to describe the spatial distribution and relationship of pixel gray levels in an image. GADF processing achieves the mapping from one-dimensional spectral data to two-dimensional image pixels, and can reflect the evolution pattern of temporal signals in the image structure. Therefore, gray-level processing algorithms can effectively capture features from GADF images. The gray-level co-occurrence matrix and gray-level difference statistics are used to describe the spatial relationship of pixel gray values ​​and the difference in gray values ​​between pixel pairs, respectively. This invention uses GLCM to read the texture features of the GADF image sequentially from 0°, 45°, 90°, and 135°, calculates the image's energy, entropy, arcsecond moment (ASM), and correlation, and outputs the mean and standard deviation of energy, entropy, ASM, and correlation as feature subsets.

[0054] Furthermore, GLDS can effectively reveal the degree of mutual variation between adjacent pixels within local regions. By calculating the differences between pixel grayscale values ​​to generate a grayscale difference image, a grayscale difference histogram is built based on these differences, and then various statistical features of the image are calculated. This invention utilizes GLDS to sequentially calculate the image's mean, contrast, arc-second moment (ASM), and entropy as feature subsets.

[0055] Step (5) adopts a feature-level data fusion strategy to fuse the feature information of terahertz absorption spectrum and Gram angle difference field image, thereby constructing a quantitative adulteration detection model for Angelica sinensis.

[0056] Traditional signal processing methods may struggle to fully capture the potential periodicity and trends in time-domain signals, making it difficult to effectively extract and utilize key variables from the data. Therefore, this invention employs GADF transformation to map spectral information with temporal characteristics into a two-dimensional space, visually demonstrating the differences between different samples through grayscale variations between image pixels. Compared to traditional fusion strategies based on spectra with different optical parameters and time-domain spectra, the utilization of GADF images demonstrates the application potential of spectral features in different representations, providing a new approach for constructing fusion models.

[0057] Color difference analysis was performed on samples with different adulteration levels using a colorimeter (CR-10, KONICA MINOLTA, Tokyo, Japan). This invention uses the color of pure Angelica sinensis powder as the standard color. The CIELAB color space theory, adopted by the International Commission on Illumination in 1976, is considered the model closest to human color perception. Based on the CIELAB color space theory, the L values ​​of different adulterated samples were calculated sequentially. * a * b * Three color coordinates. This invention uses the color difference value (dE) to represent the degree of difference between two color samples, and the calculation formula is as follows:

[0058] in, The brightness of pure angelica powder This represents the color range of pure angelica powder, from green to red. This represents the color range of pure angelica powder, from blue to yellow.

[0059] 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 computes an optimal separating hyperplane within this space, ensuring that as many training samples as possible fall within the boundary bands on both sides of the hyperplane, while maximizing the width of the boundary bands. Therefore, the SVR algorithm has significant advantages in improving model generalization ability and handling imbalanced data. This invention uses the Kennard-Stone (KS) algorithm to divide the data into a modeling set and a prediction set at a 3:1 ratio, and uses this as the input data for the SVR model. The model's penalty factor c is set to 4, and the radial basis function parameter g is set to 2.5.

[0060] This invention evaluates models by calculating the correlation coefficient (r), root mean square error (RMSE), residual prediction (RPD), and ΔE (|RMSEP - RMSEC|). High-performance models generally have high r and RPD (it is generally believed that models with RPD > 2 have high reliability and can be used for model analysis), and low RMSE and ΔE values.

[0061] The experimental results and analysis of this invention are as follows:

[0062] Figure 1 shows the spectra of various samples under different parameter settings. Figure 1(a) shows the absorption spectra of Angelica sinensis and Angelica pubescens in the range of 0.5-2.5 THz. The absorbance of both increased with increasing terahertz frequency, and the absorbance of pure Angelica pubescens was consistently higher than that of pure Angelica sinensis. This difference is mainly due to the differences in molecular composition and intermolecular interactions between the two substances. The pure Angelica sinensis sample showed a significant absorption peak at 1.88 THz, while the pure Angelica pubescens sample showed a weak spike at 2.05 THz, but without a distinct absorption peak shape. This invention performed first-derivative processing on the spectral data of Angelica pubescens to enhance the identification of absorption peaks in the spectrum. The processing result is shown in Figure 1(b), where a distinct trough appeared in the first-derivative image at 2.05 THz. According to the definition of the first 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 presence of this trough indicates that Angelica pubescens has potential absorption characteristics at this frequency. Studies have shown that ferulic acid is one of the main medicinal components of Angelica pubescens and has been proven to have antioxidant, antibacterial, and anti-inflammatory effects, showing great potential in the food and pharmaceutical fields. Related researchers discussed and verified the vibrational modes of ferulic acid in the terahertz band through density functional theory calculations and experiments. The results showed that ferulic acid produces an absorption peak at 1.90 THz under the collective vibrational mode of intermolecular molecules. Coumarin is one of the main medicinal components of Angelica pubescens, controlling the progression of rheumatoid arthritis by inhibiting the proliferation of synovial fibroblasts. Furthermore, previous studies have confirmed that coumarin has an absorption peak at 2.06 THz excited by intermolecular vibrations. Therefore, the absorption peaks at 1.88 THz and 2.05 THz may be excited by the intermolecular vibrations of ferulic acid and coumarin, respectively.

[0063] Figure 1(c) shows the time-domain signals of Angelica sinensis and Angelica pubescens in the 15-20 ps range. Pure Angelica pubescens responds faster to terahertz waves, but its amplitude is slightly lower than that of pure Angelica sinensis. This may be due to the strong absorption characteristics of Angelica pubescens to terahertz waves in the low-frequency band. Approaching 20 ps, ​​the time-domain signal of Angelica sinensis still shows weak oscillations, indicating that its response to terahertz radiation has a certain continuity. Angelica pubescens, however, shows almost no signal in this time-domain range, and its response has significantly attenuated within this range.

[0064] Figures 2 and 3 respectively show the GADF images encoded from the time-domain spectral data of Angelica sinensis samples and Heracleum hemsleyanum samples. Although the overall structures of the two images are 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 intensities, while the Heracleum hemsleyanum image shows relatively lower chromaticity values at the same position. In addition, the main feature area of the Angelica sinensis image tends to be in the central part of the image, while the main feature area of the Heracleum hemsleyanum image tends to be in the lower left corner, and the degree of deviation is relatively large. The essence of the GADF algorithm is to map time-series signals to image pixels through a specific conversion process. Each pixel contains the characteristic relationships of the signal at different time points. Therefore, the changes in color intensity and the spatial position of the main features in the GADF image can reflect the subtle differences in the time-domain signals of the samples, and these differences will provide favorable information support for the image algorithm in the feature extraction stage.

[0065] Figure 4 shows the color differences between adulterated samples with different concentrations and pure Angelica sinensis. As the adulteration amount increases, the color difference value gradually becomes larger. Based on extensive research and application accumulation in the field of color science, it is generally considered that dE≥2 is the threshold for the human eye to clearly perceive color differences, and when dE≤1, it is difficult to distinguish color differences 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 judge whether Angelica sinensis is adulterated, professional personnel are needed for distinction. The color difference coordinates show that as the Heracleum hemsleyanum powder is incorporated, the brightness difference (dL) decreases significantly, 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, and the yellow-blue color difference (db) increases, showing that the adulterated sample gradually tends to be yellow.

[0066] 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 effect of colors. It is difficult to judge the distinction of medium and low concentration adulterated samples by the naked eye.

[0067] 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. It can be concluded from Figure 6 that after 300 samplings, the number of wavelengths 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.

[0068] The IRIV algorithm employs an exponentially decreasing function (EDF) and binary matrix sampling (BMS) to optimize spectral feature selection. A total of 500 binary matrix sampling operations were performed, and the exponentially decreasing function was run 30 times to iteratively optimize feature selection, evaluating model performance in each iteration. Finally, while preserving the optimal number of features, 5-fold cross-validation was used to further evaluate the model's generalization ability and performance to determine the best feature subset. Figure 7 shows the relationship between the number of features and the number of iterations. After 6 iterations, the number of features decreased from 263 to 19. Then, an inverse elimination strategy was used to calculate the RMSECV of the remaining feature variable combinations. When the RMSECV was minimized, the inverse elimination was completed, ultimately determining 17 feature variables, accounting for 6.46% of the total variables.

[0069] In VISSA feature selection, the subset of variables generated by Weighted Binary Sampling (WBMS) was set to 1000, with an initial sampling weight of 0.5. The proportion of the subset model was 5%, and a PLS model was built using 5-fold cross-validation. The optimal feature subset was determined based on the minimum RMSECV. Figure 8 shows the relationship between the number of features and RMSECV. As the number of feature variables increases, the RMSECV value shows a trend of first decreasing sharply, then increasing slightly, and finally stabilizing. VISSA ultimately selected 10 feature variables at the minimum RMSECV value, at which point the RMSECV was 0.134, and the feature subset accounted for 3.8% of the total number of variables. Table 1 shows the wavelength selection for different feature extraction strategies.

[0070] Table 1. Results of optimal feature variable selection using CARS, IRIV, and VISSA methods.

[0071] Table 2 shows the modeling results for the full spectrum and spectral feature sets. Notably, the RPD of all models is greater than 2, demonstrating the strong adaptability and robustness of the SVR algorithm, maintaining stable performance on datasets of varying quality. The results indicate that feature selection can effectively remove redundant information and noise from spectral data and improve model accuracy by retaining key information. Regression models constructed from spectral data processed by CARS, IRIV, and VISSA all outperform models constructed from the original spectral data. Among them, the CARS-SVR model exhibits the best overall performance, with a correlation coefficient of 0.9315, an improvement of 0.048 compared to the original spectral model. Furthermore, compared to IRIV-SVR and VISSA-SVR, the CARS-SVR model has lower RMSE (0.1101) and ΔE (0.0041), indicating a significant advantage in prediction accuracy and error control, even though the input spectral data only accounts for 4.94% of the total data.

[0072] Table 2 Modeling results for full spectrum and optimal spectral variables

[0073] Figure 9 illustrates the fitting results of different models to the data. Figure 9(a) shows the model fitting results based on full-spectrum data modeling, while Figures 9(b), (c), and (d) show the model fitting results using different feature subsets. This paper further evaluates the model's fitting effect and predictive ability by calculating the 95% confidence interval and 95% prediction interval for different models. The 95% confidence interval represents the confidence interval for the regression line estimation, and the 95% prediction interval represents the prediction range for the data points. In this study, both full-spectrum and feature-spectrum modeling exhibited narrow 95% confidence intervals and good performance in the 95% prediction interval, further confirming the high accuracy and stability of the SVR model in parameter estimation. Specifically, the full-spectrum model showed relatively large residuals in the low and high concentration ranges, while the residuals were relatively small in the intermediate concentration range. This may be because strong boundary effects exist in extreme concentration regions, causing the spectral characteristics in these regions to not perfectly match the linear relationship assumed by the model, thus affecting the model's prediction accuracy for extreme values ​​and increasing the residuals. In models built using feature subsets extracted by CARS, IRIV, and VISSA, some concentration residuals are large, but the overall residuals are more uniform compared to full-spectrum modeling. Even with feature extraction, the model may still not be able to fully capture nonlinear effects at certain concentrations. Samples at specific concentrations may be more sensitive to certain weak but important spectral features, but these effects may be weakened or simplified during data dimensionality reduction, leading to larger prediction errors for some concentrations.

[0074] Table 3 shows the modeling results based on the fusion of THz spectral and GADF image data. This invention utilizes different feature extraction strategies, sequentially combining GLCM and GLDS image processing algorithms, to construct a fusion model based on spectral and image features. The results show that the dataset after fusing image features exhibits significant advantages in the model. Compared to single-feature spectral models, the correlation of all fusion models is significantly improved, the RMSE value is effectively reduced, and the RPD value remains above 2, demonstrating the reliability and superiority of the fusion model. Notably, compared to fusing spectral features with GLCM or GLDS features alone, the model that simultaneously fuses GLCM and GLDS features significantly outperforms the model fused with a single feature. This indicates that the multimodal feature fusion strategy can more effectively utilize important information in the data. Through the complementarity of GLCM and GLDS, the texture details of the GADF image are fully preserved, enabling the model to more accurately capture complex nonlinear relationships, thereby enhancing the model's predictive ability. Among them, the CARS-GLCM-GLDS-SVR model shows the most outstanding overall performance, with a correlation of 0.9704, RMSE of 0.0731, and RPD of 4.1429. Meanwhile, the CARS-SVR model improved accuracy by 3.89% after fusing GLCM and GLDS features, which is significantly higher than the improvement of the IRIV-SVR and VISSA-SVR models.

[0075] Table 3 Modeling results of terahertz spectrum fusion with GADF image data

[0076] Figures 9(e), (f), and (g) show the model fitting results of different feature extraction strategies that simultaneously fuse GLCM and GLDS features. Deep fusion of spectral and image features effectively reduces the residuals of the regression model, and the 95% prediction interval of the model is significantly reduced. Although modeling using a subset of spectral features can effectively improve the model's relevance, compared with the fitting results of the original spectral data, the 95% prediction interval of the feature spectral model is not significantly reduced, therefore the model's predictive ability remains limited. However, by simultaneously fusing GLCM and GLDS features, the 95% prediction interval of the model is significantly narrowed, indicating that the introduced image features provide effective discriminative information for judging the target concentration, enabling the model to more accurately capture the patterns in the data, thereby increasing the confidence of the predicted values ​​and reducing the residuals. The IRIV-GLCM-GLDS-SVR model and the VISSA-GLCM-GLDS-SVR model show significant residuals in some concentration regions, while the CARS-GLCM-GLDS-SVR model has the highest fitting performance.

[0077] As shown in Figure 9(h), the fusion models based on feature extraction from CARS, IRIV, and VISSA improved the accuracy by 8.69%, 6.61%, and 6.62% respectively compared to the original spectral model. The CARS model showed the largest improvement, with the RMSE value of the prediction set decreasing from 0.7806 to 0.0731 and the RPD value increasing from 2.1387 to 4.1429. This demonstrates that using appropriate feature selection algorithms can effectively improve the model's prediction accuracy and generalization ability for the target variable. In summary, the method for detecting adulteration in Angelica sinensis based on terahertz spectroscopy and data fusion provided by this invention employs terahertz time-domain spectroscopy (THz-TDS) combined with chemometric methods. Using Angelica sinensis samples adulterated with Angelica pubescens as the research object, and through multimodal feature information fusion, the adulteration of Angelica sinensis is analyzed. The spectral data with time-series characteristics are mapped to a two-dimensional plane using Gram difference field. An image is created by calculating the angles between points in the sequence, and image processing algorithms are used to extract image features. These features are then fused with spectral features to construct a regression model, thus achieving the detection of adulteration in Angelica sinensis. The study found that the CARS-GLCM-GLDS-SVR model established using the fusion method of THz spectroscopy and Gram difference field images exhibits excellent performance, with a prediction set correlation coefficient reaching 0.9704. It also achieves low root mean square error (RMSE = 0.0731) and high residual prediction (RPD = 4.1429), and compared to the original spectral model, the accuracy is improved by 8.69%, providing an efficient and reliable solution for detecting adulteration in Angelica sinensis.

[0078] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for detecting the doping amount of Angelica sinensis based on terahertz spectroscopy and data fusion, characterized in that, Includes the following steps: (1) Using Angelica pubescens powder as an adulterant, different concentrations of Angelica pubescens powder were mixed into Angelica sinensis powder to prepare samples with different adulterant amounts. The amount of Angelica pubescens powder added ranged from 0% to 100%. (2) Place each sample into the terahertz system and measure each sample to obtain the terahertz absorption spectrum and time-domain spectrum information of each sample. (3) Feature extraction of terahertz absorption spectrum information of different samples; (4) The Gram angle difference field is used to convert the terahertz time-domain spectra of different samples into images and extract the image feature information. (5) A feature-level data fusion strategy is adopted to fuse the feature information of terahertz absorption spectrum and Gram angle difference field image to construct a quantitative adulteration detection model for Angelica sinensis.

2. The method for detecting the doping amount of Angelica sinensis based on terahertz spectroscopy and data fusion according to claim 1, characterized in that, Step (1) specifically includes: Before preparing the samples, Angelica sinensis and Angelica pubescens were dried and stored in a constant temperature oven at 50℃. Before tableting, the pulverized drugs were ground thoroughly in a mortar and the 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 thoroughly mixed by a vortex mixer before tableting. Each sample was pressed at a pressure of 12 MPa for one minute, and the sample thickness was about 1.2 mm. The experiment used Angelica pubescens powder and Angelica sinensis powder to prepare samples with different adulteration ratios. Using Angelica pubescens powder as the adulterant, a total of 21 adulteration concentration gradients were prepared, with the amount of Angelica pubescens powder ranging from 0% to 100% at 5% intervals. Three parallel samples were set for each concentration gradient, for a total of 63 mixed samples.

3. The method for detecting the doping amount of Angelica sinensis based on terahertz spectroscopy and data fusion according to claim 2, characterized in that, Step (2) specifically includes: Before the experiment, the terahertz time-domain spectrometer was preheated for 30 minutes. Each measurement was set to the average of 1028 scans, with a frequency range of 0.1-7 THz. After placing each sample into the optical cavity, a 3-minute interval was allowed before measurement. Two measurements were taken from each parallel sample of each concentration, with each measurement repeated 10 times. Sixty spectral lines were collected from each concentration group, resulting in a total of 1260 spectral data points. The absorbance of the samples was used as analytical data. The absorbance A(ω) of the sample was calculated using the following formula: Among them, A sam (ω) represents the amplitude of the sample frequency domain signal, A ref (ω) is the amplitude of the reference signal.

4. The method for detecting the doping amount of Angelica sinensis based on terahertz spectroscopy and data fusion according to claim 1, characterized in that, Step (3) specifically includes: First, standard normal variable transformation combined with baseline correction is used to preprocess the spectral data. Then, competitive adaptive reweighted sampling, iterative retention of information variables, and variable iterative space shrinkage method are used to extract features from the terahertz absorption spectrum.

5. The method for detecting the doping amount of Angelica sinensis based on terahertz spectroscopy and data fusion according to claim 1, characterized in that step (4) specifically includes: Image encoding of terahertz time-domain spectral data within the 15-20 ps range was performed using Gram angle difference fields. First, the one-dimensional data was normalized and then mapped to cosine angles in polar coordinates. The time series points are represented by radial distance r, as shown in the following expression: φ=arccos(x i ),-1≤x i ≤1,x i ∈X Where, x i Represents a time series, t i Representing the timestamp, N is a parameter used to adjust the span of the polar coordinates. The Gram angular difference field (GADF) is calculated using the following formula: Texture features of the GADF image are read sequentially from 0°, 45°, 90°, and 135° using the gray-level co-occurrence matrix. The image's energy, entropy, arcsecond moment, and correlation are calculated, and the mean and standard deviation of these parameters are output as feature subsets. Furthermore, the image's mean, contrast, arcsecond moment, and entropy are calculated sequentially using gray-level difference statistics as feature subsets.

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

Images