Fingerprint-based patrinia effective component quality detection device and detection method

Abstract

The invention relates to the technical field of patrinia effective component quality detection, and discloses a patrinia effective component quality detection device and method based on a fingerprint spectrum, and the patrinia effective component quality detection method based on the fingerprint spectrum comprises the following steps: sample pretreatment and multi-dimensional data acquisition; map feature extraction and multi-source data fusion; constructing a sample characteristic spectrum database and establishing a reference standard; development of a hybrid expert system and intelligent quality diagnosis; automatic quality evaluation and detection of the sample are realized; according to the method, the patrinia effective component characteristics are represented in multiple dimensions, environmental interference is reduced, intelligent quality diagnosis is realized, dependence on expert experience is reduced, data information is deeply mined, and the comprehensiveness, accuracy and efficiency of patrinia effective component quality detection are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of quality detection of the active ingredients of Patrinia scabiosaefolia Fisch., and more specifically, it relates to a quality detection device and method for the active ingredients of Patrinia scabiosaefolia Fisch. based on fingerprint Background Art

[0002] Patrinia scabiosaefolia Fisch. is an important traditional Chinese medicine, containing a variety of bioactive ingredients and having extensive clinical application value. With the continuous deepening of its application, the requirements for the quality detection of the active ingredients of Patrinia scabiosaefolia Fisch. are also increasing day by day. At present, the quality detection of the active ingredients of Patrinia scabiosaefolia Fisch. mainly relies on methods such as traditional pharmacopoeia detection methods, single high-performance liquid chromatography methods, fingerprint technology, and the combination of physical and chemical methods and spectroscopy

[0003] Although the above methods have been applied to a certain extent in the quality detection of Patrinia scabiosaefolia Fisch., there are still technical problems such as the single detection method, large interference from environmental factors, difficulty in tracing the source of quality anomalies, excessive dependence on expert experience, and insufficient depth of data analysis. These problems lead to dilemmas such as one-sided quality evaluation results, unstable detection results, difficult quality control, low efficiency, and limited detection accuracy

[0004] Therefore, it is necessary to develop a detection method that can multi-dimensionally characterize the characteristics of the active ingredients of Patrinia scabiosaefolia Fisch., reduce environmental interference, realize intelligent quality diagnosis, reduce dependence on expert experience, and deeply mine data information, so as to improve the comprehensiveness, accuracy, and efficiency of the quality detection of the active ingredients of Patrinia scabiosaefolia Fisch Summary of the Invention

[0005] The present invention provides a quality detection device and method for the active ingredients of Patrinia scabiosaefolia Fisch. based on fingerprint, which solves the technical problems of the single detection method, large interference from environmental factors, difficulty in tracing the source of quality anomalies, excessive dependence on expert experience, and insufficient depth of data analysis in the related technology

[0006] The present invention provides a quality detection method for the active ingredients of Patrinia scabiosaefolia Fisch. based on fingerprint, including the following steps:

[0007] Sample pretreatment and multi-dimensional data acquisition, including sample pretreatment, HPLC fingerprint data acquisition, electronic nose volatile component data acquisition, and optical image data acquisition;

[0008] Spectrum feature extraction and multi-source data fusion, including fingerprint spectrum feature extraction, electronic nose data feature extraction, optical image feature extraction, and multi-source data feature fusion;

[0009] Construction of sample characteristic spectrum library and establishment of reference standards, including collection and analysis of standard samples, establishment of quality evaluation index system, construction of characteristic spectrum library, and establishment of quality grading standards;

[0010] Development of a hybrid expert system and intelligent quality diagnosis, including construction of machine learning models, construction of knowledge bases and expert rules, integration of hybrid diagnostic systems, and implementation of diagnostic result interpretation systems;

[0011] Implementation of automatic quality evaluation and detection of samples, including pretreatment and data acquisition of samples to be tested, feature extraction and similarity evaluation, quality grade determination, anomaly detection and intelligent diagnosis, and output and storage of detection results.

[0012] In a preferred embodiment, the sample pretreatment includes: cleaning, drying, pulverizing, and sieving the Patrinia scabiosaefolia samples to prepare standard sample powders with a moisture content of less than 5% and a particle size of 80-120 mesh.

[0013] In a preferred embodiment, the HPLC fingerprint data acquisition uses a high-performance liquid chromatograph, equipped with a C18 chromatographic column, a column temperature of 30 °C, a mobile phase of acetonitrile-0.1% phosphoric acid aqueous solution for gradient elution, a flow rate of 1.0 mL / min, a detection wavelength of 254 nm, and an injection volume of 10 μL.

[0014] In a preferred embodiment, the fingerprint feature extraction includes: extracting the peak position feature RT i , the peak area feature A i , the peak width W i , the peak height H i , the peak asymmetry coefficient AS i , and the peak ratio feature AR ij .

[0015] In a preferred embodiment, the multi-source data feature fusion uses a weighted fusion method, expressed as:

[0016] F = α·F chrom + β·F enose + γ·F image ;

[0017] where F represents the fused comprehensive feature vector, and F chrom , F enose , F image represent the high-performance liquid chromatography fingerprint features, electronic nose volatile component features, and optical image features respectively. α, β, and γ are the corresponding weight coefficients, respectively reflecting the contribution degrees of the high-performance liquid chromatography fingerprint features, electronic nose volatile component features, and optical image features to the comprehensive quality evaluation, and satisfying α + β + γ = 1.

[0018] In a preferred embodiment, the quality evaluation index system includes the component integrity index CI, the active ingredient content index EC, and the impurity control index IP.

[0019] In a preferred embodiment, the quality grading standard divides the quality of Patrinia villosa samples into four grades: excellent, first-class, qualified, and unqualified, and conducts grading by calculating the comprehensive quality index:

[0020] Q = w CI ·CI + w EC ·EC + w IP ·IP;

[0021] Wherein, Q represents the comprehensive quality index, CI represents the component integrity index, EC represents the active ingredient content index, IP represents the impurity control index, and w CI 、w EC 、w IP respectively represent the weight coefficients of the component integrity index, the active ingredient content index, and the impurity control index, satisfying w CI + w EC + w IP = 1.

[0022] In a preferred embodiment, the hybrid diagnostic system uses Dempster-Shafer evidence theory to fuse the diagnostic results of different models and calculate the comprehensive diagnostic result and its credibility.

[0023] In a preferred embodiment, the feature extraction and similarity evaluation include calculating the similarity S corr between the fingerprint spectrum of the sample to be tested and the standard spectrum, as well as the characteristic peak matching degree and the comprehensive feature similarity.

[0024] In a preferred embodiment, a quality detection device for the active ingredients of Patrinia villosa based on fingerprint spectrum includes a memory and one or more processors. Executable code is stored in the memory. When the one or more processors execute the executable code, it is used to implement the steps in a quality detection method for the active ingredients of Patrinia villosa based on fingerprint spectrum.

[0025] The beneficial effects of the present invention are as follows:

[0026] The present invention can characterize the characteristics of the active ingredients of Patrinia villosa in multiple dimensions, reduce environmental interference, realize intelligent quality diagnosis, reduce the dependence on expert experience, deeply mine data information, and improve the comprehensiveness, accuracy, and efficiency of the quality detection of the active ingredients of Patrinia villosa. Brief Description of the Drawings

[0027] Figure 1 is a flowchart of a quality detection method for the active ingredients of Patrinia villosa based on fingerprint spectrum according to the present invention;

[0028] Figure 2 is a detailed flowchart of sample pretreatment and multi-dimensional data acquisition according to the present invention;

[0029] Figure 3 is the detailed flowchart of the feature extraction and multi-source data fusion of the present invention's fingerprint spectrum;

[0030] Figure 4 is the detailed flowchart of the construction of the sample fingerprint spectrum library and the establishment of reference standards of the present invention;

[0031] Figure 5 is the detailed flowchart of the development of the expert system and the intelligent quality diagnosis of the present invention;

[0032] Figure 6 is the detailed flowchart of the automatic quality evaluation and detection implementation of the samples of the present invention. Specific embodiments

[0033] Now, the subject matter described herein will be discussed with reference to example embodiments. It should be understood that discussing these embodiments is only to enable those skilled in the art to better understand and thus implement the subject matter described herein. Without departing from the scope of protection of the content of this specification, changes can be made to the functions and arrangements of the elements discussed. Each example can omit, substitute, or add various processes or components as needed. Additionally, the features described in some examples can also be combined in other examples.

[0034] In at least one embodiment of the present invention, a method for detecting the quality of the effective components of Patrinia scabiosaefolia based on fingerprint spectrum is disclosed. As Figures 1-6 shown, it includes the following steps:

[0035] Step 1, sample pretreatment and multi-dimensional data acquisition, including sample pretreatment, HPLC fingerprint spectrum data acquisition, electronic nose volatile component data acquisition, and optical image data acquisition;

[0036] Specifically, it includes the following steps:

[0037] Step 1.1, sample pretreatment;

[0038] Perform pre-treatment on the Patrinia scabiosaefolia samples, including cleaning, drying, pulverizing, and sieving, to obtain sample powders with uniform particle size and standard moisture content. The pre-treatment is carried out in an environment with a temperature controlled at 25 ± 2 °C and a relative humidity controlled at 40% ± 5% to ensure the consistency and stability of the sample treatment process.

[0039] Step 1.2, obtaining high-performance liquid chromatography fingerprint spectrum data;

[0040] The pretreated Herba Patriniae samples were analyzed by high performance liquid chromatography (HPLC) to obtain fingerprint data. The specific operation is as follows: 0.5 g of the pretreated Herba Patriniae sample powder was taken, 10 mL of extraction solvent (70% methanol aqueous solution) was added, ultrasonic extraction was performed for 30 minutes, centrifugation was performed (12000 rpm, 10 minutes), the supernatant was filtered through a 0.22 μm filter membrane and injected into the HPLC system for analysis to generate Herba Patriniae fingerprint data. Chromatographic conditions: C18 chromatographic column (250 mm × 4.6 mm, 5 μm), the mobile phase was acetonitrile-0.1% phosphoric acid aqueous solution gradient elution, the flow rate was 1.0 mL / min, the detection wavelength was 254 nm, the column temperature was 30 ° C, and the injection volume was 10 μL.

[0041] Step 1.3, electronic nose volatile component data collection;

[0042] The volatile components of the Herba Patriniae were detected using an electronic nose system. 2.0 g of the pretreated Herba Patriniae sample was placed in a sealed headspace bottle. After equilibration at 60°C for 15 minutes, the sample was detected using an electronic nose system with a metal oxide sensor array to obtain the electrical signal response curve of the volatile components of Herba Patriniae. The sensor array contains 12 sensors with different chemical sensitivity characteristics, which have different response sensitivities to different volatile compounds.

[0043] Step 1.4, optical image data acquisition;

[0044] A high-resolution digital camera (24 million pixels) was used to photograph the Patrinia scabra samples to obtain RGB high-definition images under standard lighting conditions (5500K color temperature, uniform lighting). At the same time, a multispectral imaging system was used to collect multispectral images of 8 different bands in the wavelength range of 400-1000nm to obtain more comprehensive optical feature information.

[0045] Step 2, graph feature extraction and multi-source data fusion, including fingerprint graph feature extraction, electronic nose data feature extraction, optical image feature extraction and multi-source data feature fusion;

[0046] The specific steps include:

[0047] Step 2.1, fingerprint feature extraction;

[0048] For the HPLC fingerprint, a multi-peak feature extraction algorithm is used to analyze the spectrum and extract the following key features:

[0049] Peak position feature: Identify the characteristic peak position in the fingerprint spectrum, expressed as retention time RT i , i=1,2,...,n, where n is the number of characteristic peaks;

[0050] Peak area feature: Calculate the peak area A of each characteristic peak i , where i = 1, 2, ..., n, and n is the number of characteristic peaks, representing the relative content of the corresponding component;

[0051] Peak shape feature: Extract the peak width, peak height, and peak asymmetry coefficient of each characteristic peak to form the peak shape feature vector S i = W i , H i , AS i , where W i represents the peak width, H i represents the peak height, and AS i represents the peak asymmetry coefficient, i = 1, 2, ..., n, and n is the number of characteristic peaks;

[0052] Peak ratio feature: Calculate the peak area ratio AR of important components ij = A i / A j , and construct the relative ratio matrix of characteristic peaks, where AR ij represents the ratio of the i-th peak area to the j-th peak area, A i represents the i-th peak area, and A j represents the j-th peak area.

[0053] The identification of characteristic peaks adopts an adaptive peak detection algorithm. The peak position is determined by calculating the zero-crossing point position of the second derivative curve, and the peak detection threshold is automatically adjusted according to the signal-to-noise ratio to ensure the accuracy and robustness of the detection.

[0054] Step 2.2, Feature extraction of electronic nose data;

[0055] Extract features from the sensor response curve collected by the electronic nose, including:

[0056] Static feature: Calculate the maximum value E max,j , mean value E mean,j , and variance E var,j of each sensor response, where j = 1, 2, ..., m, and m is the number of sensors;

[0057] Dynamic feature: Extract the rise time, peak time, and recovery time of each sensor response curve to construct the dynamic feature matrix:

[0058] D j = [T rise,j , T peak,j , T recovery,j ;

[0059] where D j represents the dynamic feature matrix, and T rise,j represents the rise time of each sensor response curve, Tpeak,j Denotes the peak time, T, of each sensor response curve recovery,j Denotes the recovery time of each sensor response curve

[0060] Summarize the electronic nose features as a feature vector:

[0061] E = [E max , E mean , E var , D];

[0062] Where E represents the electronic nose feature vector, E max Denotes the vector composed of the maximum values of all sensor responses, E mean Denotes the vector composed of the means of all sensor responses, E var Denotes the vector composed of the variances of all sensor responses, and D represents the matrix composed of all sensor dynamic response features

[0063] Step 2.3, Optical image feature extraction

[0064] Extract features from the collected optical images, including:

[0065] Color features: Calculate the color statistics of the RGB image, including:

[0066] Mean values of the RGB three channels:

[0067] C mean = [R mean , G mean , B mean ;

[0068] Where C mean Denotes the mean values of the RGB three channels, and R mean , G mean , B mean Denote the pixel means of the image in the red, green, and blue channels respectively

[0069] Standard deviation:

[0070] C std = [R std , G std , B std ;

[0071] Where C std Denotes the standard deviation, and R std , G std , B std Denote the standard deviations of the pixel values of the image in the red, green, and blue channels respectively

[0072] Color histogram feature: H color ;

[0073] Texture features: The texture features are extracted using the Gray Level Co - occurrence Matrix (GLCM), including contrast Con, correlation Cor, energy Ene, and homogeneity Hom, to form a texture feature vector:

[0074] T texture = [Con, Cor, Ene, Hom];

[0075] Among them, T texture represents the texture feature vector, Con represents the contrast of the gray - level difference in the image, Cor represents the correlation between the gray - level values of different pixels in the image, Ene represents the degree of uniformity of the gray - level distribution in the image, and Hom represents the consistency of the gray - level distribution in the local area of the image;

[0076] Multispectral features: Calculate the average reflectance M k of each spectral band, k = 1, 2,..., 8, to form a multispectral feature vector M = [M1, M2,..., M8]. Among them, M represents the multispectral feature vector, and M1, M2, M8 represent the average reflectances of the 1st, 2nd, and 8th spectral bands respectively.

[0077] Step 2.4, Multi - source data feature fusion;

[0078] Adopt a hierarchical feature fusion algorithm to effectively fuse the extracted multi - source data features:

[0079] First, normalize each feature vector so that all feature values are mapped to the interval [0, 1];

[0080] Use Principal Component Analysis (PCA) to reduce the dimension of each feature set and remove redundant information;

[0081] Use a weighted fusion method to combine feature vectors from different sources to form a comprehensive feature vector F:

[0082] F = α·F chrom + β·F enose + γ·F image ;

[0083] Among them, F represents the fused comprehensive feature vector, F chrom 、F enose 、F image represent the features of high - performance liquid chromatography fingerprint, the features of volatile components of electronic nose, and the features of optical images respectively. α, β, and γ are the corresponding weight coefficients, which respectively reflect the contribution degrees of the features of high - performance liquid chromatography fingerprint, the features of volatile components of electronic nose, and the features of optical images to the comprehensive quality evaluation, and satisfy α + β + γ = 1. The weight coefficients are optimized and determined through cross - validation of the training set samples to maximize the classification and evaluation performance.

[0084] Step 3, construction of the sample characteristic spectrum library and establishment of reference standards, including collection and analysis of standard samples, establishment of a quality evaluation index system, construction of the characteristic spectrum library, and establishment of quality grading standards;

[0085] Specifically, it includes the following steps:

[0086] Step 3.1, collection and analysis of standard samples;

[0087] Select 100 batches of Patrinia villosa samples that meet the pharmacopoeia standards as the standard sample set. These samples come from different origins and different harvesting times, but all have been professionally identified as meeting the Chinese Pharmacopoeia standards, including:

[0088] 80 batches of qualified samples: samples identified as superior and first-class quality according to the pharmacopoeia standards;

[0089] 20 batches of control samples: samples including artificially added impurities, artificial aging, and different degrees of extraction, used to simulate different degrees of quality problems.

[0090] Perform data collection and feature extraction on all standard samples according to the methods in Step 1 and Step 2 to form a standard sample characteristic data set.

[0091] Step 3.2, establishment of a quality evaluation index system;

[0092] Based on the multi-dimensional characteristic data of the standard samples, establish a quality evaluation index system for Patrinia villosa:

[0093] Primary indicators: include the component integrity index CI, the active ingredient content index EC, and the impurity control index IP;

[0094] Secondary indicators: specific evaluation parameters corresponding to each primary indicator, such as the number of characteristic peaks, the peak area of key components, the peak area ratio, etc.;

[0095] Calculation method:

[0096] Component integrity index:

[0097]

[0098] Among them, CI represents the component integrity index, n is the total number of characteristic peaks, w i is the weight of the i-th characteristic peak, and δ i is the indicator function for the presence of the characteristic peak (1 for presence, 0 for absence);

[0099] Active ingredient content index:

[0100]

[0101] Among them, EC represents the effective ingredient content index, m is the number of key effective ingredients, and v j is the weight of the j-th key effective ingredient, A j is the peak area of this ingredient in the sample, and A ref,j is the average peak area of the corresponding ingredient in the reference sample;

[0102] Impurity control index:

[0103]

[0104] Among them, IP represents the impurity control index, p is the number of impurity types, and u k is the weight of the k-th type of impurity, I k is the detected value of this type of impurity in the sample, and I max,k is the allowable maximum value.

[0105] The weight coefficient of each index is determined by the analytic hierarchy process combined with expert scoring to ensure the scientificity and rationality of the evaluation system.

[0106] Step 3.3, construction of the characteristic spectrum library;

[0107] Based on the analysis results of the standard sample set, construct the characteristic spectrum library of Patrinia scabiosaefolia Fisch.:

[0108] Calculate the distribution characteristics of each characteristic parameter in qualified samples, including the mean μ i , standard deviation σ i and coefficient of variation CV i ;

[0109] Determine the normal range of each characteristic parameter:

[0110] [L i ,U i = [μ i -kσ i ,μ i +kσ i ;

[0111] Among them, [L i ,U i represents the normal range of each characteristic parameter, μ i represents the average value of the i-th characteristic parameter in all qualified samples, σ i represents the standard deviation of the i-th characteristic parameter in all qualified samples, L i represents the normal lower limit of the i-th characteristic parameter, U i represents the normal upper limit, and k is the range coefficient, which is set to 2 - 3 according to the importance of the characteristics;

[0112] Establish a characteristic spectrum reference template, including reference values such as the standard fingerprint spectrum profile, relative retention times of key peaks, and relative peak area ratios;

[0113] Construct a characteristic correlation network and record the correlation coefficient matrix CR = [cr ij n×n , where CR represents the correlation coefficient matrix between features, and cr ij represents the Pearson correlation coefficient between the i-th feature and the j-th feature for subsequent anomaly detection.

[0114] Step 3.4, Establish the quality grading standard;

[0115] Establish a quality grading standard for Patrinia scabiosaefolia samples, and divide the sample quality into four grades: excellent, first-class, qualified, and unqualified:

[0116] Calculate the comprehensive quality index:

[0117] Q = w CI ·CI + w EC ·EC + w IP ·IP;

[0118] Among them, Q represents the comprehensive quality index, CI represents the component integrity index, EC represents the active ingredient content index, IP represents the impurity control index, and w CI , w EC , w IP represent the weight coefficients of the component integrity index, active ingredient content index, and impurity control index respectively, satisfying w CI + w EC + w IP = 1.

[0119] Formulate the grading standard:

[0120] Excellent: Q ≥ 0.9 and all first-level indicators are not less than 0.85;

[0121] First-class: 0.8 ≤ Q < 0.9 and all first-level indicators are not less than 0.75;

[0122] Qualified: 0.7 ≤ Q < 0.8 and all first-level indicators are not less than 0.65;

[0123] Unqualified: Q < 0.7 or any first-level indicator is less than 0.65.

[0124] Step 4, Development of a hybrid expert system and intelligent quality diagnosis, including construction of a machine learning model, construction of a knowledge base and expert rules, integration of a hybrid diagnosis system, and implementation of a diagnosis result interpretation system;

[0125] Specifically, it includes the following steps: ​

[0126] Step 4.1, Machine learning model construction;

[0127] Based on the data of the standard sample set, develop multiple machine learning models for the quality evaluation and anomaly detection of Patrinia scabiosaefolia:

[0128] Support Vector Machine (SVM) model: used for the classification of samples and the determination of quality grades, with the fused feature vector F as the input and the sample quality grade as the output;

[0129] Random forest model: used for feature importance analysis and the detection of abnormal samples, improving the detection stability through a voting mechanism;

[0130] Deep convolutional neural network model: directly takes the original spectral data and optical images as the input, automatically learns and extracts deep features, and realizes end-to-end quality evaluation;

[0131] Anomaly detection model: combines the Isolation Forest algorithm and one-class SVM to realize the detection of quality problems of unknown types.

[0132] Each model is trained using 80% of the standard sample data and 20% for validation to ensure the generalization ability of the model.

[0133] Step 4.2, Knowledge base and expert rule construction;

[0134] Establish a knowledge base for the diagnosis of quality problems of Patrinia scabiosaefolia and an expert rule system:

[0135] Collect and summarize the performance characteristics of common quality problems, including the correlation relationships between the absence of specific peaks, abnormal peak areas, peak shape changes, etc. and quality problems;

[0136] Design inference rules and construct a set of rules in the form of "IF-THEN":

[0137] Rule 1: IF characteristic peak P i is missing AND characteristic peak P j decreases by more than 50% THEN it may be that the active ingredient C k degrades;

[0138] Rule 2: IF characteristic ratio R 12 exceeds the upper limit of the normal range AND the response of electronic nose sensor S3 is abnormal THEN there may be mildew;

[0139] Rule 3: IF optical feature T contrast significantly decreases AND the ratio of multi-spectral feature M5 / M2 is abnormal THEN there may be doping problems;

[0140] Assign a credibility coefficient λ i to each rule for the credibility evaluation of the final diagnosis result.

[0141] Step 4.3, Hybrid Diagnostic System Integration;

[0142] Integrate the machine learning model with the expert rule system to build a hybrid expert diagnostic system:

[0143] Adopt a hierarchical integration architecture. First, use the machine learning model for quality level assessment and anomaly detection;

[0144] For the detected abnormal samples, call the expert rule system for problem diagnosis and cause analysis;

[0145] Use Dempster-Shafer evidence theory to fuse the diagnostic results of different models, and calculate the comprehensive diagnostic result and its credibility:

[0146] Assign a basic belief m i (A) to represent the degree of support for the diagnostic result A;

[0147] Use Dempster's combination rule to calculate the fusion result of multiple evidence sources:

[0148]

[0149] Among them, m(A) represents the degree of belief in proposition A after fusion, m1(B) and m2(C) respectively represent the degrees of belief of evidence source 1 and evidence source 2 in propositions B and C, and B∩C = A means that the intersection of proposition B supported by evidence source 1 and proposition C supported by evidence source 2 is A, indicating that proposition B supported by evidence source 1 and proposition C supported by evidence source 2 are incompatible;

[0150] Output the final diagnostic result and its credibility.

[0151] Step 4.4, Implementation of Diagnostic Result Explanation System;

[0152] Develop a diagnostic result explanation system to make the detection results transparent and understandable to users:

[0153] Design a visual interface to intuitively display the sample quality evaluation results and key indicators;

[0154] Implement feature contribution degree analysis to identify and display the key features that play a decisive role in quality determination;

[0155] For abnormal samples, provide a problem diagnosis report, including:

[0156] Problem type and credibility;

[0157] Abnormal features and their deviation degrees;

[0158] Suggested improvement measures or treatment methods;

[0159] Historical case references for similar problems (if any).

[0160] Through the interpretation system, the detection results not only inform the quality status of the samples, but also provide analysis of the problem causes and improvement suggestions, realizing intelligent quality management.

[0161] Step 5, realization of automatic quality evaluation and detection of samples, including pretreatment and data collection of samples to be tested, feature extraction and similarity evaluation, quality grade determination, anomaly detection and intelligent diagnosis, output and storage of detection results;

[0162] Specifically, it includes the following steps:

[0163] Step 5.1, pretreatment and data collection of samples to be tested;

[0164] Perform standardized pretreatment on the samples of Patrinia scabiosaefolia to be tested, and conduct multi-source data collection according to the method described in Step 1:

[0165] Adopt the same sample pretreatment method to ensure consistency with the reference standard;

[0166] Collect fingerprint data using the same liquid chromatography conditions;

[0167] Collect electronic nose data and optical image data according to the unified parameter settings;

[0168] Input the collected multi-source data into the system for subsequent processing.

[0169] Step 5.2, feature extraction and similarity evaluation;

[0170] Extract features from the multi-source data of the samples to be tested and evaluate the similarity with the standard:

[0171] Extract multi-dimensional features of the samples to be tested according to the method described in Step 2;

[0172] Calculate the similarity between the sample features and the reference standard:

[0173] Fingerprint similarity: Calculate the similarity between the fingerprint of the sample to be tested and the standard fingerprint using the correlation coefficient method:

[0174]

[0175] Among them, S corr represents the correlation coefficient between the fingerprint of the sample to be tested and the standard fingerprint, X i represents the signal intensity of the i-th data point of the fingerprint of the sample to be tested, X represents the average signal intensity of all data points of the fingerprint of the sample to be tested, Y i$S_i$ represents the signal intensity of the $i$-th data point of the standard fingerprint spectrum, $Y$ represents the average signal intensity of all data points of the standard fingerprint spectrum, and $\sum$ represents the summation over all data points.

[0176] Characteristic peak similarity: Calculate the matching degree between the characteristic peaks of the sample to be tested and the standard characteristic peaks, including the consistency of the number of peaks, peak positions, and peak area ratios.

[0177] Comprehensive feature similarity: Calculate the Euclidean distance or cosine similarity between the fused feature vector and the standard feature vector.

[0178] Step 5.3, Quality grade determination.

[0179] Based on the feature similarity and quality evaluation indicators, determine the quality grade of the sample:

[0180] Calculate the quality evaluation index values of the sample to be tested, including the component integrity index CI, the active ingredient content index EC, and the impurity control index IP.

[0181] Using the quality grading standard established in Step 3, combined with the classification results of the machine learning model, comprehensively determine the quality grade of the sample.

[0182] Calculate the credibility of the quality determination result, based on the consistency of multiple models and the evidence theory framework.

[0183] Step 5.4, Anomaly detection and intelligent diagnosis.

[0184] Perform anomaly detection and intelligent diagnosis on samples that do not meet the quality standards:

[0185] Adopt an anomaly detection algorithm to identify the abnormal points in the sample characteristics.

[0186] Call a hybrid expert system for problem diagnosis to determine the possible causes of quality problems.

[0187] Generate a diagnostic report, including:

[0188] Type of quality problem;

[0189] Key abnormal characteristics and their deviation degrees;

[0190] Analysis of problem causes and diagnostic credibility;

[0191] Suggestions for improvement or handling.

[0192] Step 5.5, Output and storage of detection results.

[0193] The system generates a complete detection report and stores the results:

[0194] Generate a standardized detection report, including sample information, detection data, quality grade, anomaly diagnosis results, etc.

[0195] Provide visual quality evaluation results, including characteristic spectrum comparison, evaluation index radar chart, highlighting of abnormal features, etc.;

[0196] Store the detection data and results in the database for historical traceability and continuous system learning;

[0197] Update the knowledge base, incorporate new sample characteristics and diagnostic experiences into the system, and achieve self-optimization and iterative upgrade of the system.

[0198] Technical effects:

[0199] The quality detection method of the active ingredients of Patrinia villosa based on fingerprint spectrum proposed by the present invention is realized through multi-source data acquisition, feature extraction and fusion, construction of a characteristic spectrum library, development of a hybrid expert system, and automatic quality evaluation, and has achieved the following remarkable technical effects:

[0200] Improved detection comprehensiveness and accuracy:

[0201] Multi-dimensional feature fusion effect: By fusing HPLC fingerprint spectrum, electronic nose data and optical image data, this method realizes multi-dimensional and all-round characterization of Patrinia villosa samples. Compared with traditional single detection methods, the amount of detection information has increased by more than 200%.

[0202] Significantly improved detection accuracy: Verified, the accuracy rate of this method for determining the quality grade of Patrinia villosa samples reaches 95.8%, which is better than 82.3% of traditional methods; the diagnostic accuracy rate for common quality problems reaches 91.2%, which is 1.5 times that of traditional methods.

[0203] Enhanced sensitivity to minor quality differences: This method shows extremely high sensitivity to minor quality changes in Patrinia villosa samples, and can detect 5% content changes and 10% component changes that are difficult to discover by traditional methods, providing the possibility for early warning of quality problems.

[0204] Realization of intelligent diagnosis ability:

[0205] Automated quality evaluation: This method realizes the automatic evaluation and grading of the quality of Patrinia villosa samples through a hybrid expert system, shortening the time required for traditional manual evaluation from an average of 4 hours to within 10 minutes, and the efficiency has increased by 24 times.

[0206] Diagnosis of the causes of quality problems: The system can automatically analyze the causes of quality abnormalities, diagnose 15 common types of quality problems, and the diagnostic credibility reaches 87.5%, filling the gap that traditional methods can only determine whether it is qualified or not and cannot give reasons.

[0207] Result interpretability: Through the diagnostic result interpretation system, this method provides transparent and understandable quality evaluation results, enabling users to clearly understand the basis and key influencing factors of quality determination, and enhancing the credibility and usability of the detection results.

[0208] Advantages of generality and adaptability:

[0209] Method standardization and normalization: This method has established standardized sample processing, data collection, and analysis processes, reducing the interference of human factors and enabling the detection results to have good repeatability (relative standard deviation < 3%) and consistency among laboratories.

[0210] System self-learning ability: By continuously accumulating sample data and diagnostic experience, the system has the ability of self-learning and optimization. The detection performance improves with the increase in the sample size, showing obvious technological progressiveness.

[0211] Expansion of application scope: Although this method takes Patrinia scabiosaefolia as the research object, its technical framework and method system can be extended and applied to the quality detection of other traditional Chinese medicines and natural products, showing broad application prospects.

[0212] In summary, the quality detection method of the active ingredients of Patrinia scabiosaefolia based on fingerprint proposed by the present invention, through the innovative application of multi-source data fusion and hybrid expert system, solves the technical problems existing in traditional detection methods, such as large environmental interference, slow quality anomaly diagnosis, dependence on expert experience, and insufficient data analysis, and realizes the comprehensiveness, accuracy, intelligence, and standardization of the quality detection of the active ingredients of Patrinia scabiosaefolia, showing significant technological innovation and practical value.

[0213] Application examples:

[0214] To verify the effectiveness and practicality of the method of the present invention, Patrinia scabiosaefolia samples with different quality grades and different quality problems were selected for testing. The specific implementation process and results are as follows:

[0215] Application example 1: Detection and grading of Patrinia scabiosaefolia samples with different quality grades;

[0216] Sample source: A total of 30 batches of Patrinia scabiosaefolia samples were collected from 6 different production areas (Anhui, Zhejiang, Hubei, Jiangxi, Sichuan, Henan), including 8 batches of superior samples, 10 batches of first-class samples, 7 batches of qualified samples, and 5 batches of unqualified samples identified by national drug inspection agencies.

[0217] Experimental equipment:

[0218] High-performance liquid chromatograph: Agilent 1260 series, equipped with a diode array detector;

[0219] Electronic nose system: AlphaMOS Heracles NEO electronic nose, equipped with 10 metal oxide sensors with different characteristics;

[0220] Image acquisition system: Canon EOS 90D high-definition camera, equipped with a standard light source box and a DigiEye multispectral imaging system;

[0221] Experimental procedure:

[0222] Perform sample pretreatment and multi-dimensional data acquisition according to Step 1 of the method of the present invention;

[0223] Perform data processing, feature extraction, quality evaluation and diagnosis according to Steps 2-5;

[0224] Test results and analysis:

[0225] Quality grade determination results:

[0226] Perform quality grade determination on 30 batches of samples by the method of the present invention. The consistency of the results with the identification results of the national drug inspection agency is 93.3% (28 / 30). Among the two inconsistent samples, one was determined to be of a lower grade (from first grade to qualified) by this method, and one was determined to be of a higher grade (from qualified to first grade). Further analysis shows that the quality of these two samples is in a critical state. Through the analysis of the key indicators given by this method, it is found that some indicators of these two samples are indeed close to the critical values, proving the sensitivity and rationality of this method.

[0227] Quality index analysis:

[0228] Perform statistical analysis on the quality indicators of samples of different grades. The results show that:

[0229] Superior grade samples: average comprehensive quality index Q = 0.936 ± 0.023, component integrity index CI = 0.957 ± 0.018;

[0230] First grade samples: average comprehensive quality index Q = 0.845 ± 0.031, component integrity index CI = 0.863 ± 0.039;

[0231] Qualified samples: average comprehensive quality index Q = 0.732 ± 0.025, component integrity index CI = 0.745 ± 0.042;

[0232] Unqualified samples: average comprehensive quality index Q = 0.583 ± 0.078, component integrity index CI = 0.612 ± 0.093;

[0233] Analysis of variance showed that there were significant differences in the quality indexes of samples at different grades (p<0.001), indicating that the quality evaluation index system established by this method had good discrimination ability.

[0234] Repeatability test:

[0235] Three batches of samples at different grades (superior grade, first grade, qualified) were selected, and each batch of samples was tested 6 times. The relative standard deviation (RSD) of the quality indexes was calculated to be less than 3%, indicating that this method had good repeatability and reliability.

[0236] Application example 2: Intelligent diagnosis of the quality problem of Patrinia scabiosaefolia Fisch.

[0237] Sample preparation:

[0238] Five batches of Patrinia scabiosaefolia Fisch. samples meeting the pharmacopoeia standards were selected, and samples with different types of quality problems were prepared through artificial intervention:

[0239] Sample A: Normal control sample;

[0240] Sample B: Simulating the degradation of active ingredients (heating at 60°C for 30 days);

[0241] Sample C: Simulating microbial contamination (inoculating molds and culturing for 72 hours);

[0242] Sample D: Simulating adulteration (adding 30% of other plant materials);

[0243] Sample E: Simulating heavy metal contamination (adding lead salt solution);

[0244] Detection and diagnosis process:

[0245] The samples were detected and intelligently diagnosed according to the complete steps of the method of the present invention, and the diagnosis results and diagnosis bases were recorded.

[0246] Diagnosis results and analysis:

[0247] Diagnosis results of sample B (degradation of active ingredients):

[0248] Quality grade determination: Unqualified;

[0249] Problem diagnosis: Degradation of active ingredients, diagnosis credibility 92.7%;

[0250] Abnormal characteristics: The peak areas of 3 key active ingredient peaks in the liquid chromatography fingerprint decreased by more than 70%, and the peak shapes changed; significant changes occurred in the volatile component spectra in the electronic nose data;

[0251] Suggestions given by the system: Check the sample storage conditions, avoid high temperature and exposure to the sun, and pay attention to the expiration date;

[0252] Diagnosis result of Sample C (microbial contamination):

[0253] Quality grade determination: unqualified;

[0254] Problem diagnosis: microbial contamination / mildew, diagnosis credibility 89.5%;

[0255] Abnormal characteristics: significant increase in the response values of electronic nose sensors S3 and S7; detection of abnormal color areas in the optical image; appearance of non-characteristic peaks in the fingerprint spectrum;

[0256] Suggestions given by the system: check the humidity of the storage environment, strengthen moisture-proof and sterilization treatments;

[0257] Diagnosis result of Sample D (adulteration):

[0258] Quality grade determination: unqualified;

[0259] Problem diagnosis: adulteration with foreign plants, diagnosis credibility 94.3%;

[0260] Abnormal characteristics: appearance of non-Patrinia scabiosaefolia characteristic peaks in the fingerprint spectrum; abnormal texture and color parameters in the optical characteristics; significant reduction in the comprehensive feature similarity;

[0261] Suggestions given by the system: purchase from regular channels, strengthen raw material acceptance and variety identification;

[0262] Diagnosis result of Sample E (heavy metal contamination):

[0263] Quality grade determination: unqualified;

[0264] Problem diagnosis: heavy metal contamination, diagnosis credibility 85.2%;

[0265] Abnormal characteristics: abnormal reflectance in specific bands in the multi-spectral characteristics; change in the response pattern of specific electronic nose sensors; baseline drift in the fingerprint spectrum;

[0266] Suggestions given by the system: check the planting environment, strengthen heavy metal detection, use products from planting bases meeting GAP requirements for comparative analysis with traditional methods;

[0267] Compare the method of the present invention with the traditional quality detection method for Patrinia scabiosaefolia, and the results show:

[0268] Comparison of detection time:

[0269] Traditional method: sample treatment, HPLC analysis, and data interpretation altogether take 240 ± 30 minutes;

[0270] Method of the present invention: the whole process of detection and diagnosis only takes 35 ± 5 minutes, with an efficiency improvement of about 85%;

[0271] Cost comparison:

[0272] Traditional method: The detection cost for each batch of samples is approximately 300 yuan (excluding labor).

[0273] Method of the present invention: The initial equipment investment is relatively large, but the detection cost for a single batch of samples is reduced to 180 yuan, which is more economical for long-term use.

[0274] Comparison of diagnostic capabilities:

[0275] Traditional method: Mainly rely on experienced experts for judgment, and the diagnosis of the causes of quality problems highly depends on personal experience.

[0276] Method of the present invention: Automated intelligent diagnosis, which can identify 15 common quality problems and give credibility and suggestions, with a diagnostic accuracy rate of 91.2%.

[0277] Application example 3: System continuous learning and optimization effect;

[0278] To verify the system's ability of continuous learning and self-optimization in the method of the present invention, a 6-month system operation test was conducted. During this period, a total of 200 batches of samples were detected, and the verified sample data was fed back into the system:

[0279] Change in detection accuracy rate:

[0280] Initial stage (the first 50 batches): The accuracy rate of quality grade determination is 89.3%, and the accuracy rate of problem diagnosis is 82.6%.

[0281] Mid-term stage (51 - 150 batches): The accuracy rate of quality grade determination is 93.7%, and the accuracy rate of problem diagnosis is 88.4%.

[0282] Late stage (151 - 200 batches): The accuracy rate of quality grade determination is 96.8%, and the accuracy rate of problem diagnosis is 93.5%.

[0283] The results show that with the increase in the sample size and the continuous learning of the system, the detection and diagnosis performance are significantly improved.

[0284] New problem identification ability:

[0285] During the test process, the system successfully identified two new types of quality problems (excessive residual pesticides and abnormal components caused by climate change in specific regions) that were not covered previously, and through continuous learning, incorporated these new problems into the diagnostic knowledge base, enriching the diagnostic capabilities.

[0286] Parameter self-optimization effect:

[0287] Through the analysis of the cumulative data, the system automatically optimized the feature fusion weights (α, β, γ) and the quality index weight coefficients, making the determination results more in line with the actual situation. The accuracy rate of quality index determination for feature fusion increased from the initial 89.3% to 96.8%.

[0288] Through the detailed testing and analysis of the above application examples, the significant technical effects and practical value of the method of the present invention in the quality detection of the active ingredients of Patrinia scabiosaefolia Fisch. ex Link are fully verified. It is proved that this method not only has the ability of high-precision and high-efficiency quality detection, but also has the characteristics of intelligent diagnosis and continuous optimization, providing an advanced technical means for the quality control of Patrinia scabiosaefolia Fisch. ex Link and other Chinese medicinal materials.

[0289] The above embodiments of the present invention have been described, but the present embodiments are not limited to the above specific implementation manners. The above specific implementation manners are only illustrative and not restrictive. Under the inspiration of the present embodiments, those of ordinary skill in the art can also make more equivalent embodiments in various forms, all of which fall within the protection scope of the present embodiments.

Claims

1. A method for detecting the quality of active ingredients of Patrinia scabra based on fingerprint, characterized in that: The following steps are involved: Sample pretreatment and multi-dimensional data collection, including sample pretreatment, HPLC fingerprint data collection, electronic nose volatile component data collection and optical image data collection; Spectrum feature extraction and multi-source data fusion, including fingerprint spectrum feature extraction, electronic nose data feature extraction, optical image feature extraction and multi-source data feature fusion; Construction of sample characteristic spectrum library and establishment of reference standards, including standard sample collection and analysis, establishment of quality evaluation index system, construction of characteristic spectrum library and establishment of quality grading standards; Hybrid expert system development and quality intelligent diagnosis, including machine learning model construction, knowledge base and expert rule construction, hybrid diagnosis system integration and diagnosis result interpretation system implementation; Automatic sample quality evaluation and detection are realized, including sample preprocessing and data collection, feature extraction and similarity evaluation, quality grade determination, anomaly detection and intelligent diagnosis, and test result output and storage.

2. A fingerprint-based quality detection method for effective components of Patrinia scabra according to claim 1, characterized in that: The sample pretreatment comprises: washing, drying, crushing and sieving the Patrinia scabra sample to prepare a standard sample powder with a moisture content of less than 5% and a particle size of 80-120 mesh.

3. The method for detecting the quality of active ingredients of Patrinia scabra based on fingerprint according to claim 1, characterized in that: The HPLC fingerprint data was collected using a high performance liquid chromatograph equipped with a C18 chromatographic column, a column temperature of 30° C., a mobile phase of acetonitrile-0.1% phosphoric acid aqueous solution for gradient elution, a flow rate of 1.0 mL / min, a detection wavelength of 254 nm, and an injection volume of 10 μL.

4. The method for detecting the quality of active ingredients of Patrinia scabra based on fingerprint spectrum according to claim 1, characterized in that: The fingerprint spectrum feature extraction includes: extracting the peak position feature RT i , Peak area characteristic A i , peak width W i , peak height H i , peak asymmetry coefficient AS i Sum peak ratio feature AR ij .

5. The method for detecting the quality of active ingredients of Patrinia scabra based on fingerprint spectrum according to claim 1, characterized in that: The multi-source data feature fusion adopts a weighted fusion method, which is expressed as: F=α·F chrom +β·F enose +γ·F image ; Among them, F represents the integrated feature vector after fusion, F chrom 、F enose 、F image They represent the HPLC fingerprint characteristics, the volatile component characteristics of the electronic nose and the optical image characteristics respectively. α, β and γ are the corresponding weight coefficients, which respectively reflect the contribution of the HPLC fingerprint characteristics, the volatile component characteristics of the electronic nose and the optical image characteristics to the comprehensive quality evaluation, satisfying α+β+γ=1.

6. The method for detecting the quality of active ingredients of Patrinia scabra based on fingerprint spectrum according to claim 1, characterized in that: The quality evaluation index system includes a component integrity index CI, an effective component content index EC and an impurity control index IP.

7. The method for detecting the quality of active ingredients of Patrinia scabra based on fingerprint spectrum according to claim 1, characterized in that: The quality grading standard divides the quality of Patrinia scabra samples into four grades: excellent, first-class, qualified and unqualified, and the grading is performed by calculating the comprehensive quality index: Q=w CI ·CI+in EC ·EC+in IP ·IP; Among them, Q represents the comprehensive quality index, CI represents the ingredient integrity index, EC represents the active ingredient content index, IP represents the impurity control index, and w CI 、w EC 、w IP Respectively represent the weight coefficients of ingredient integrity index, effective ingredient content index, and impurity control index, satisfying w CI +w EC +w IP =1.

8. The method for detecting the quality of active ingredients of Patrinia scabra based on fingerprint spectrum according to claim 1, characterized in that: The hybrid diagnosis system adopts the Dempster-Shafer evidence theory to fuse the diagnosis results of different models and calculate the comprehensive diagnosis result and its credibility.

9. The method for detecting the quality of active ingredients of Patrinia scabra based on fingerprint spectrum according to claim 1, characterized in that: The feature extraction and similarity evaluation includes calculating the similarity S between the fingerprint spectrum of the sample to be tested and the standard spectrum. corr , as well as feature peak matching and comprehensive feature similarity.

10. A fingerprint-based quality detection device for active ingredients of Patrinia scabra, characterized in that: It comprises a memory and one or more processors, wherein the memory stores executable codes, and when the one or more processors execute the executable codes, they are used to implement the steps in the method for detecting the quality of active ingredients of Patrinia scabra based on fingerprint spectrum as described in any one of claims 1 to 9.

Images