A method and system for detecting and analyzing the quality of a traditional Chinese medicine decoction piece finished product

Abstract

This invention discloses a method and system for quality testing and analysis of finished traditional Chinese medicine (TCM) decoction pieces, belonging to the pharmaceutical field. The method includes the following steps: S1. Constructing a compound component database, which integrates pharmacopoeia standards, clinical prescription structures, medicinal material source attributes, and efficacy-related parameters, and establishes a standard model in a multi-level structured manner; S2. Collecting multimodal spectral data of TCM decoction piece samples; S3. Encoding the collected spectral data using a heterogeneous feature extraction algorithm to generate a fused structural feature vector; S4. Inputting the fused structural feature vector into a multi-channel evaluation model for weight comparison, classification judgment, and threshold assessment; S5. Generating a quality testing report. Beneficial effects: This method achieves high-precision quality testing combining multimodal fusion, intelligent discrimination, and traceability visualization, effectively improving the quality consistency assessment and risk identification capabilities of TCM decoction pieces.

Description

Technical Field

[0001] This invention relates to the field of medicine, and more specifically, to a method and system for quality testing and analysis of finished Chinese herbal medicine products. Background Technology

[0002] With the rapid development of the traditional Chinese medicine (TCM) industry, the quality control of traditional Chinese medicinal herbs has received increasing attention. In the past, quality testing mainly relied on sensory evaluation and chemical composition analysis, but these methods have significant limitations. In recent years, with the advancement of modern technology, the introduction of instrumental analysis techniques such as liquid chromatography, gas chromatography, and mass spectrometry has made the quality testing of TCM herbs more accurate and repeatable. These techniques can quantitatively analyze active ingredients and compare different batches of herbs using fingerprinting, thereby improving the scientific rigor and reliability of the testing. Furthermore, with the continuous development of data analysis technology, more and more intelligent methods are being applied to the quality control of TCM. The combination of big data and machine learning algorithms makes it possible to analyze the quality of TCM herbs from multiple angles and dimensions, providing a more comprehensive and accurate quality assessment method. Modern testing technologies have made significant progress in improving testing efficiency, reducing errors, and broadening testing dimensions.

[0003] However, despite the significant support that modern instrumental analysis techniques have provided for the quality testing of traditional Chinese medicine (TCM) decoction pieces, existing technologies still have some shortcomings. Traditional quality testing methods tend to focus on chemical analysis, often neglecting the complexity and overall characteristics of TCM compound formulas, resulting in a lack of comprehensive evaluation of multi-component and multi-characteristic decoction pieces. Furthermore, existing methods still rely heavily on the quantification of single components, making it difficult to effectively reflect the quality changes of TCM decoction pieces under different environmental and storage conditions. Although fingerprinting technology provides multi-component analysis, its efficiency and accuracy in large-scale sample applications still face challenges. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for quality testing and analysis of finished Chinese herbal medicine (TCM) decoction pieces, addressing the problems mentioned in the background: traditional quality testing methods tend to focus on chemical analysis, often neglecting the complexity and overall characteristics of TCM compound formulas, resulting in a lack of comprehensive evaluation of multi-component and multi-characteristic decoction pieces. Furthermore, existing methods still rely heavily on the quantification of single components, making it difficult to effectively reflect the quality changes of TCM decoction pieces under different environmental and storage conditions. Although fingerprinting technology provides multi-component analysis, its efficiency and accuracy in large-scale sample applications still face challenges.

[0005] Technical solution. The method for quality testing and analysis of finished Chinese herbal medicine slices includes the following steps:

[0006] S1. Construct a compound component database, which integrates pharmacopoeia standards, clinical prescription structures, medicinal material source attributes and efficacy-related parameters, and establishes a standard model in a multi-level structured manner;

[0007] S2. Collect multimodal spectral data of Chinese herbal medicine decoction pieces, including infrared spectrum, ultraviolet absorption spectrum, mass spectrometry peak diagram, tissue structure image and colorimetric data;

[0008] S3. The collected spectral data is encoded using a heterogeneous feature extraction algorithm to generate a fused structural feature vector;

[0009] S4. Input the fusion structure feature vector into the multi-channel evaluation model to perform weight comparison, classification judgment and threshold evaluation;

[0010] S5. Generate a quality inspection report, which includes sample-level data, key quality attribute indicators, similarity analysis results with database reference values, and risk identification recommendations.

[0011] Preferably, the step S1 of constructing the compound component database further includes the following steps:

[0012] S1-1. Based on data from previous editions of the Pharmacopoeia and information from traditional classic prescriptions, extract the generic names, properties and meridians of the constituent medicinal materials, dosage ranges and compatibility relationships, and form a basic compatibility network;

[0013] S1-2. Model the information on the origin distribution, harvesting time, and processing method of the medicinal materials into a regional source and processing technology structure library;

[0014] S1-3. Integrate modern pharmacological and target research data, and incorporate known active ingredients, biological pathway correlations, and dose-response relationships of each medicinal material into the model.

[0015] Preferably, the compound component database further integrates a genotyping identification system and environmental variable influencing factors. By performing gene sequencing and climate factor modeling on medicinal material samples from different producing areas, a full factor structure mapping relationship between the origin of medicinal materials and their intrinsic quality is established, and this relationship is used as one of the traceability dimensions of the database to participate in the judgment of sample quality consistency.

[0016] Preferably, the step S2 of acquiring multimodal spectral data further includes the following steps:

[0017] S2-1. Use a multi-band near-infrared spectrometer to perform spectral scanning on the sample, extract absorption characteristic peaks and perform preliminary characteristic spectrum localization;

[0018] S2-2. Use a laser confocal microscopy imaging device to photograph the tissue structure of the sample cross section and extract the morphological features of vascular bundles, cortical cells and epidermis.

[0019] S2-3. The fingerprint ion peaks of the samples were determined by time-of-flight mass spectrometry, and the ion peaks were calibrated and the spectra were normalized.

[0020] S2-4. Acquire sample chromaticity images and quantify chromaticity features using a standardized color difference model.

[0021] Preferably, before fusion, the multimodal spectral data is coupled with features aligned by an embedded multi-channel structure registration module, and the coupling nodes between different modal features are identified by a graph matching algorithm. The main peak shift, image structure density distribution and mass spectrometry ion co-occurrence mode of the spectral data are expressed in a unified vector space.

[0022] Preferably, step S2-2, which involves using a laser confocal microscopy device to capture tissue structure images, further includes the following steps:

[0023] S2-2-1. Perform automated isothermal staining on sample sections to enhance image contrast between cell walls and vascular bundles;

[0024] S2-2-2. Multi-channel fluorescence illumination technology was used to perform layered scanning of the sample cross-section to construct a three-dimensional cell structure model;

[0025] S2-2-3. Using a deep learning image segmentation network, identify the characteristic structural regions of medicinal materials and extract morphological parameters.

[0026] Preferably, the multi-channel evaluation model further introduces an interpretable mechanism, which generates a visual explanation path of the model's decision source through heatmap analysis of the characteristic response channels within the model, weight contribution inversion, and structural source mapping, and forms multimodal source tracing prompts in the final quality report to assist in clinical review and re-examination.

[0027] Preferably, the step S5 of generating a quality inspection report further includes the following steps:

[0028] S5-1. Output sample-level labels based on the classification results of the multi-channel evaluation model and associate them with sample batch number information;

[0029] S5-2. Generate a fluctuation trend graph based on the sample quality parameters and the standard deviation of the database, and mark the quality stability curve;

[0030] S5-3. Output a quality risk map by identifying the main regions of difference between the sample and the standard spectrum;

[0031] S5-4. Link the source of medicinal materials, processing technology and batch variation causes at the database index level to form a closed-loop feedback chain.

[0032] Preferably, the step S5-2 of generating the fluctuation trend chart further includes the following steps:

[0033] S5-2-1. Implement sliding window regression processing on key parameters of five or more consecutive batches of samples to construct a short-term variation model;

[0034] S5-2-2. An anomaly detection mechanism is adopted to mark batches with drastic parameter changes and differentiate them by color.

[0035] S5-2-3. Perform correlation analysis between sample variation and time dimension, origin and batch, and generate a structure diagram of the causes of fluctuation.

[0036] Preferably, the quality testing and analysis system for finished Chinese herbal medicine pieces includes the quality testing and analysis method for finished Chinese herbal medicine pieces as described in claims 1-9.

[0037] Compared with the prior art, the advantages of this invention are:

[0038] (1) This invention precisely controls the moisture distribution and pulverization process of Chinese herbal medicine samples and uses a low-speed grinder for segmented pulverization, avoiding the loss of effective components caused by traditional high-temperature pulverization, thereby ensuring the stability of the sample components.

[0039] (2) In the quantitative analysis of chemical components, a combination of high performance liquid chromatography (HPLC) and gas chromatography (GC) is used, and quantitative analysis is carried out in conjunction with standard reference curves. At the same time, system adaptability verification is performed, thereby improving the accuracy and repeatability of the analysis results.

[0040] (3) The present invention establishes a more comprehensive sensory detection model. In addition to traditional color, odor and morphology analysis, it also adds the detection of subtle changes in the sample, such as changes in odor intensity, thereby improving the accuracy of sensory analysis and comprehensive evaluation capabilities.

[0041] (4) In fingerprint spectrum comparison analysis, in addition to constructing a standard fingerprint spectrum, the present invention adopts multi-dimensional data fusion technology, which makes the spectrum comparison more accurate and can effectively identify the small differences between the sample and the standard spectrum, thereby improving the sensitivity of quality detection.

[0042] (5) The present invention adopts a multimodal data fusion algorithm to comprehensively analyze various detection results, effectively combining data such as chemical composition, sensory characteristics and fingerprint spectrum, to achieve more comprehensive quality assessment and intelligent judgment, and greatly improve detection efficiency and accuracy. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the overall process of a method for quality testing and analysis of finished Chinese herbal medicine slices according to the present invention; Detailed Implementation

[0044] For examples, please refer to Figure 1A method for quality testing and analysis of finished Chinese herbal medicine slices includes the following steps:

[0045] S1. Construct a compound component database, which integrates pharmacopoeia standards, clinical prescription structures, medicinal material source attributes and efficacy-related parameters, and establishes a standard model in a multi-level structured manner;

[0046] Specifically, the data structure was constructed by first collecting medicinal material data from various editions of the Pharmacopoeia and information on classic Chinese medicine prescriptions, and then establishing a multi-level structured database of medicinal materials. Python and R language tools were used, combined with a medicinal material database API, for data extraction and formatting. The model was built based on the properties, flavors, meridian tropism, dosage ranges, and compatibility relationships of the medicinal materials. Cluster analysis techniques were used to obtain the typical combination characteristics of each medicinal material in different prescriptions.

[0047] Pharmacodynamic correlation modeling: Utilizing bioinformatics tools (such as the STRING database) and pharmacological research findings, known active ingredients, targets, and metabolic pathways of medicinal materials are integrated to form bioactivity correlations for each material. Based on network analysis, synergistic effects and incompatibilities among medicinal materials are established.

[0048] S2. Collect multimodal spectral data of Chinese herbal medicine decoction pieces, including infrared spectrum, ultraviolet absorption spectrum, mass spectrometry peak diagram, tissue structure image and colorimetric data;

[0049] Specifically: Infrared spectrum acquisition was performed using a near-infrared spectrometer (NIR) to scan the sample, covering a wavelength range of 1000 nm to 2500 nm. Key peaks in the spectrum were selected based on absorption characteristics, and then normalization and spectral smoothing were applied.

[0050] Ultraviolet absorption spectrum acquisition: The absorption spectrum of the sample was measured using a UV-Vis spectrophotometer. Characteristic absorption peaks in the ultraviolet region were extracted by wavelength scanning.

[0051] Mass spectrometry peak acquisition: The sample was ionized using a time-of-flight mass spectrometer (TOF-MS) to extract fingerprint ion peaks. The mass spectrometry data was calibrated using time / mass analysis to quantify and normalize the ion peaks.

[0052] Tissue structure image acquisition: Images of tissue sections were captured using a laser confocal microscope, and image features of structural regions such as vascular bundles, epidermis, and cortical cells were extracted. Cell morphology features, such as cell wall thickness and vascular structure, were extracted using image processing algorithms.

[0053] Colorimetric data acquisition: The colorimetric values ​​of the samples are measured using a colorimeter, and their RGB values ​​are recorded. The colorimetric values ​​are then quantified using a color difference standardization model.

[0054] S3. The collected spectral data is encoded using a heterogeneous feature extraction algorithm to generate a fused structural feature vector;

[0055] Specifically: Feature extraction, using PCA (Principal Component Analysis) and LDA (Linear Discriminant Analysis) to perform dimensionality reduction for each data mode. Principal components are extracted from infrared spectroscopy and mass spectrometry data, and structural features are extracted from tissue images and colorimetric data.

[0056] Encoding and Fusion: Deep learning feature extraction is performed on tissue images using convolutional neural networks (CNNs), and data denoising is achieved using an autoencoder algorithm. Finally, the multimodal data is mapped to a unified vector space, and the final feature vector is generated through weighted fusion.

[0057] S4. Input the fused structural feature vector into the multi-channel evaluation model to perform weight comparison, classification judgment and threshold evaluation;

[0058] Specifically, the multi-channel evaluation model employs a multi-channel network architecture based on convolutional neural networks (CNNs) and recurrent neural networks (RNNs). The feature vectors of each modality are input into multiple parallel convolutional layers to extract their respective local features, which are then aggregated through fully connected layers.

[0059] Weight alignment and classification: Classification algorithms (such as Support Vector Machine (SVM) and KNN) are used to classify the final feature vectors and output the quality level of the samples. Cross-validation is combined to optimize the classification performance of the model, ensuring accuracy and stability.

[0060] Threshold assessment: A threshold determination mechanism is added to the model to define different quality standards (such as excellent, good, qualified, etc.) and compare them with the standards in the reference database to conduct the final quality assessment.

[0061] S5. Generate a quality inspection report, which includes sample-level data, key quality attribute indicators, similarity analysis results with database reference values, and risk identification recommendations.

[0062] Specifically: Quality assessment report generation. Based on the output of the classification model, a detailed report on sample quality is generated. The report includes the quality category, the measured values ​​of relevant quality attributes (such as water content, component concentration, etc.), and is compared with the standard values ​​in the database.

[0063] Risk assessment: Through similarity analysis, a quality risk map is generated to identify potential quality problems in the samples. Based on historical data, relevant suggestions are output, such as retesting or product improvement.

[0064] S1 The construction of a compound component database further includes the following steps:

[0065] S1-1. Based on data from previous editions of the Pharmacopoeia and information from traditional classic prescriptions, extract the generic names, properties and meridians of the constituent medicinal materials, dosage ranges and compatibility relationships, and form a basic compatibility network;

[0066] Specifically, data mining techniques (such as text mining and natural language processing) are used to extract medicinal material information from pharmacopoeias and traditional literature. Clustering algorithms and graph network modeling are employed to transform the compatibility relationships of medicinal materials into a network structure.

[0067] S1-2. Model the information on the origin, harvesting time, and processing method of medicinal materials into a database of regional origin and processing technology structure;

[0068] Specifically, climate data models (such as those using remote sensing data and climate databases) are used to assess environmental factors in medicinal herb producing areas. Environmental impact models of medicinal herbs are developed by combining factors such as harvesting cycles and processing methods, and then input into the database.

[0069] S1-3. Integrate modern pharmacological and target research data, and incorporate known active ingredients, biological pathway correlations, and dose-response relationships of each medicinal material into the model.

[0070] Specifically: Pharmacological information of medicinal materials is extracted from databases (such as PubMed and ChEMBL). Biological network analysis methods (such as gene enrichment analysis and pathway analysis) are applied to combine pharmacological data with experimental data to create a relationship model between active ingredients and targets.

[0071] The compound component database further integrates the genotyping identification system and environmental variable influencing factors. By performing gene sequencing and climate factor modeling on medicinal material samples from different producing areas, it establishes a full factor structure mapping relationship between the origin of medicinal materials and their intrinsic quality, and uses it as one of the traceability dimensions of the database to participate in the judgment of sample quality consistency.

[0072] Specifically, the genotypic differences of medicinal herbs from different authentic producing areas are analyzed using gene sequencing technologies (such as NGS), and a full-factor quality model of the herbs is established by combining environmental variables (such as temperature, humidity, and precipitation). Through big data analysis, a mapping relationship between origin and quality is established and integrated into a compound prescription database as one of the traceability dimensions.

[0073] S2 acquisition of multimodal spectral data further includes the following steps:

[0074] S2-1. Use a multi-band near-infrared spectrometer to perform spectral scanning on the sample, extract absorption characteristic peaks and perform preliminary characteristic spectrum localization;

[0075] Specifically: a near-infrared spectrometer was used for spectral scanning, with the wavelength range set from 1000 nm to 2500 nm. Key absorption peaks were extracted based on known chemical spectral data. Regularization and smoothing algorithms were then used to optimize the spectral signal.

[0076] S2-2. Use a laser confocal microscopy imaging device to photograph the tissue structure of the sample cross section and extract the morphological features of vascular bundles, cortical cells and epidermis.

[0077] Specifically: Tissue images of cross-sections of samples are captured using a confocal microscope, and image processing algorithms (such as filtering and thresholding) are employed to extract cell morphology features. Fractal dimension algorithms are then used to extract the geometric features of cell arrangement.

[0078] S2-3. The fingerprint ion peaks of the samples were determined by time-of-flight mass spectrometry, and the ion peaks were calibrated and the spectra were normalized.

[0079] Specifically: Time-of-flight mass spectrometry (TOF-MS) was used to determine the fingerprint ion peaks of the samples. Combined with a standard mass spectrometry database, the ion peaks were calibrated and normalized to eliminate baseline drift and noise interference.

[0080] S2-4. Acquire sample chromaticity images and quantify chromaticity features using a standardized color difference model.

[0081] Specifically: Colorimeters or cameras are used to capture chromaticity images of the samples, and standardized color difference models (such as the CIELab model) are employed for quantitative analysis of the chromaticity data. Algorithms are used to automatically assess the color stability of the samples.

[0082] Before fusion, the multimodal spectral data are aligned by an embedded multi-channel structure registration module, and the coupling nodes between different modal features are identified by a graph matching algorithm. The main peak shift, image structure density distribution and mass spectrometry ion co-occurrence mode of the spectral data are expressed in a unified vector space.

[0083] Specifically, the spectral data fusion process in this embodiment employs a multi-channel registration module. By aligning features of data from different modalities, a graph matching algorithm is used to identify coupling nodes between data from different modalities. In this way, it ensures that all spectral data are uniformly represented and analyzed within a unified vector space, thereby achieving effective fusion of the main peak shift, image structure distribution, and mass spectrometry ion co-occurrence modes of the spectral data.

[0084] S2-2, using a laser confocal microscopy imaging device to capture tissue structure images, further includes the following steps:

[0085] S2-2-1. Perform automated isothermal staining on sample sections to enhance image contrast between cell walls and vascular bundles;

[0086] S2-2-2. Multi-channel fluorescence illumination technology was used to perform layered scanning of the sample cross-section to construct a three-dimensional cell structure model;

[0087] S2-2-3. Using a deep learning image segmentation network, identify the characteristic structural regions of medicinal materials and extract morphological parameters.

[0088] Specifically, in this embodiment, an automated isothermal staining technique is used to process the sample sections, enhancing the contrast between the cell wall and the vascular bundles to ensure image quality. During the image acquisition stage, multi-channel fluorescence illumination technology is used to perform layered scanning of the sample, and a deep learning network is used to segment the image to extract the characteristic structural regions and morphological parameters of the medicinal materials.

[0089] The multi-channel evaluation model further introduces an interpretable mechanism. Through heatmap analysis of characteristic response channels within the model, weight contribution inversion, and structural source mapping, a visual explanation path of the model's decision-making sources is generated, and multimodal source attribution prompts are formed in the final quality report to assist in clinical review and re-examination.

[0090] Specifically, in this embodiment, to improve the interpretability of the model, heatmap analysis technology is used to analyze the contribution of each feature response channel in the model, and a weighted contribution inversion mechanism is used to identify key factors affecting decision-making. Furthermore, a visual explanatory path is generated through structural source mapping to ensure that detailed multimodal causal hints are provided in the quality report, offering decision support for clinical review and re-examination.

[0091] According to claim 1, a method for quality testing and analysis of finished Chinese herbal medicine slices is characterized in that, step S5, generating a quality testing report, further includes the following steps:

[0092] S5-1. Output sample-level labels based on the classification results of the multi-channel evaluation model and associate them with sample batch number information;

[0093] S5-2. Generate a fluctuation trend graph based on the sample quality parameters and the standard deviation of the database, and mark the quality stability curve;

[0094] S5-3. Output a quality risk map by identifying the main regions of difference between the sample and the standard spectrum;

[0095] S5-4. Link the source of medicinal materials, processing technology and batch variation causes at the database index level to form a closed-loop feedback chain.

[0096] Specifically, in the quality inspection report generation process, the sample level label is first output through the classification results of the multi-channel evaluation model and linked to the sample batch number information to ensure data traceability. Subsequently, based on the quality parameters and the standard deviation of the database, a fluctuation trend graph and quality stability curve for the sample are generated to aid in the analysis of quality stability. By identifying the difference regions between the sample and the standard spectrum, a quality risk map is generated, and a feedback loop is formed by linking the source of medicinal materials, processing technology, and causes of batch variation through the database.

[0097] S5-2 generates a fluctuation trend chart, which further includes the following steps:

[0098] S5-2-1. Implement sliding window regression processing on key parameters of five or more consecutive batches of samples to construct a short-term variation model;

[0099] S5-2-2. An anomaly detection mechanism is adopted to mark batches with drastic parameter changes and differentiate them by color.

[0100] S5-2-3. Perform correlation analysis between sample variation and time dimension, origin and batch, and generate a structure diagram of the causes of fluctuation.

[0101] Specifically, for generating the fluctuation trend chart, a short-term variation model is constructed by using a sliding window regression method combined with key parameters of continuous samples. Simultaneously, an outlier identification mechanism is employed to mark batches with drastically changing parameters, and these are visualized using color differentiation. Finally, by combining the time dimension and the origin of the batches, a correlation analysis of the variation is performed to generate a structure diagram of the causes of fluctuations, further revealing the root causes of sample quality fluctuations.

[0102] A quality testing and analysis system for finished Chinese herbal medicine pieces includes a quality testing and analysis method for finished Chinese herbal medicine pieces as claimed in claims 1-9.

[0103] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and modifications can be made to the present invention without departing from the spirit and scope thereof, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting and analyzing the quality of Chinese herbal medicine decoction pieces, characterized in that, The method for quality testing and analysis of finished Chinese herbal medicine pieces includes the following steps: S1. Construct a compound component database, which integrates pharmacopoeia standards, clinical prescription structures, medicinal material source attributes and efficacy-related parameters, and establishes a standard model in a multi-level structured manner; S2. Collect multimodal spectral data of Chinese herbal medicine decoction pieces, including infrared spectrum, ultraviolet absorption spectrum, mass spectrometry peak diagram, tissue structure image and colorimetric data; S3. The collected spectral data is encoded using a heterogeneous feature extraction algorithm to generate a fused structural feature vector; S4. Input the fusion structure feature vector into the multi-channel evaluation model to perform weight comparison, classification judgment and threshold evaluation; S5. Generate a quality inspection report, which includes sample-level data, key quality attribute indicators, similarity analysis results with database reference values, and risk identification recommendations; The step S1, constructing the compound component database, further includes the following steps: S1-1. Based on data from previous editions of the Pharmacopoeia and information from traditional classic prescriptions, extract the generic names, properties and meridians of the constituent medicinal materials, dosage ranges and compatibility relationships, and form a basic compatibility network; S1-2. Model the information on the origin distribution, harvesting time, and processing method of the medicinal materials into a regional source and processing technology structure library; S1-3. Integrate modern pharmacological and target research data, and incorporate known active ingredients, biological pathway correlations, and dose-response relationships of each medicinal material into the model.

2. The method for detecting and analyzing the quality of finished traditional Chinese medicine decoction pieces according to claim 1, characterized in that, The compound component database further integrates the genotyping identification system and environmental variable influencing factors. By performing gene sequencing and climate factor modeling on medicinal material samples from different producing areas, a full factor structure mapping relationship between the origin of medicinal materials and their intrinsic quality is established, and this relationship is used as one of the traceability dimensions of the database to participate in the judgment of sample quality consistency.

3. The method for detecting and analyzing the quality of Chinese medicine decoction pieces according to claim 1, characterized in that, The S2 acquisition of multimodal spectral data further includes the following steps: S2-1. Use a multi-band near-infrared spectrometer to perform spectral scanning on the sample, extract absorption characteristic peaks and perform preliminary characteristic spectrum localization; S2-2. Use a laser confocal microscopy imaging device to photograph the tissue structure of the sample cross section and extract the morphological features of vascular bundles, cortical cells and epidermis. S2-3. The fingerprint ion peaks of the samples were determined by time-of-flight mass spectrometry, and the ion peaks were calibrated and the spectra were normalized. S2-4. Acquire sample chromaticity images and quantify chromaticity features using a standardized color difference model.

4. The method for quality testing and analysis of finished traditional Chinese medicine decoction pieces according to claim 1, characterized in that, Before fusion, the multimodal spectral data is aligned by an embedded multi-channel structure registration module, and the coupling nodes between different modal features are identified by a graph matching algorithm. The main peak shift, image structure density distribution and mass spectrometry ion co-occurrence mode of the spectral data are expressed in a unified vector space.

5. The method for detecting and analyzing the quality of Chinese medicine decoction pieces according to claim 3, characterized in that, S2-2, which uses a laser confocal microscopy imaging device to capture tissue structure images, further includes the following steps: S2-2-1. Perform automated isothermal staining on sample sections to enhance image contrast between cell walls and vascular bundles; S2-2-2. Multi-channel fluorescence illumination technology was used to perform layered scanning of the sample cross-section to construct a three-dimensional cell structure model; S2-2-3. Using a deep learning image segmentation network, identify the characteristic structural regions of medicinal materials and extract morphological parameters.

6. The method for detecting and analyzing the quality of Chinese medicine decoction pieces according to claim 1, characterized in that, The multi-channel evaluation model further introduces an interpretable mechanism. Through heatmap analysis of the characteristic response channels within the model, weight contribution inversion, and structural source mapping, a visual explanation path of the model's decision-making sources is generated, and multimodal source attribution prompts are formed in the final quality report to assist in clinical review and re-examination.

7. The method for detecting and analyzing the quality of Chinese medicine decoction pieces according to claim 1, characterized in that, The S5 process for generating a quality inspection report further includes the following steps: S5-1. Output sample-level labels based on the classification results of the multi-channel evaluation model and associate them with sample batch number information; S5-2. Generate a fluctuation trend graph based on the sample quality parameters and the standard deviation of the database, and mark the quality stability curve; S5-3. Output a quality risk map by identifying the main regions of difference between the sample and the standard spectrum; S5-4. Link the source of medicinal materials, processing technology and batch variation causes at the database index level to form a closed-loop feedback link.

8. The method for detecting and analyzing the quality of Chinese medicine decoction pieces according to claim 7, characterized in that, The S5-2 method for generating a fluctuation trend chart further includes the following steps: S5-2-1. Implement sliding window regression processing on key parameters of five or more consecutive batches of samples to construct a short-term variation model; S5-2-2. An anomaly detection mechanism is adopted to mark batches with drastic parameter changes and differentiate them by color. S5-2-3. Perform correlation analysis between sample variation and time dimension, origin and batch, and generate a structure diagram of the causes of fluctuation.

9. A Chinese herbal piece finished product quality detection and analysis system, characterized in that, The aforementioned system for quality testing and analysis of finished Chinese herbal medicine pieces includes the method for quality testing and analysis of finished Chinese herbal medicine pieces as described in any one of claims 1-8.

Images