Method and system for identifying origin of authentic Chinese medicinal materials

Abstract

The present application provides a method and system for identifying the origin of authentic Chinese medicinal materials. The method includes: extracting the multidimensional features of each spectrum based on the multispectral response signal; inputting the multidimensional features of each spectrum into a graph convolutional network, extracting the spatial distribution characteristics of the multispectral response signal and the target topological relationship of the wavelength dimension through the extraction module in the graph convolutional network; dynamically updating the origin classification boundary based on the spatial distribution characteristics and the wavelength dimension through the recognition module in the graph convolutional network, and determining the origin identification result of the authentic Chinese medicinal material sample based on the updated origin classification boundary. The technical solution provided by the present application is based on the multidimensional features collected by the spectral sensor array, dynamically extracts the spatial distribution and wavelength topological relationship through the graph convolutional network, and adaptively optimizes the classification boundary, thereby realizing accurate and efficient identification of the origin of authentic Chinese medicinal materials.

Description

Technical Field

[0001] The present application relates to the technical field of origin traceability of authentic Chinese medicinal materials, and in particular to a method and system for identifying the origin of authentic Chinese medicinal materials. Background Art

[0002] Authentic Chinese medicinal materials (TCMs) are selected through long-term clinical application in Traditional Chinese Medicine (TCM). They are produced in specific regions (production areas) through specific production processes and are recognized by the TCM community as having superior quality, greater efficacy, and more stable quality compared to similar TCMs produced in other regions. In the field of origin traceability and quality control of authentic TCMs, there is an urgent need for a technology that can efficiently, objectively, and accurately identify the origin of medicinal materials.

[0003] Most existing solutions are methods for identifying the origin of Chinese medicinal materials based on a single spectrum and static classification model. This method uses a single type of spectral equipment to scan authentic Chinese medicinal material samples; then extracts features and inputs the extracted features into a static classifier, which is ultimately used for classification and identification of the origin.

[0004] However, the existing schemes have the following defects: (1) The features are local peaks or statistics, which make it difficult to effectively express the nonlinear coupling relationship between discrete feature peaks in the wavelength dimension, resulting in the loss of cross-scale spectral fingerprint information; (2) The classification boundary of the static classifier cannot be adjusted after training is completed, and it is difficult to adapt to the feature drift caused by natural variation of samples or environmental disturbances, resulting in a lack of adaptability of the classification boundary. Summary of the Invention

[0005] The present application provides a method and system for identifying the origin of authentic Chinese medicinal materials, which is used to solve the problem of low accuracy of origin identification caused by the loss of cross-scale spectral fingerprint information and lack of adaptability of classification boundaries in the existing technology.

[0006] In a first aspect, the present application provides a method for identifying the origin of authentic Chinese medicinal materials, comprising:

[0007] A spectral sensor array is used to synchronously excite authentic Chinese medicinal material samples in a pulse alternating mode to generate a multi-spectral response signal, which includes Raman and UV spectra.

[0008] extracting multidimensional features of each spectrum based on the multispectral response signal;

[0009] Inputting the multidimensional features of each spectrum into a graph convolutional network, and extracting the spatial distribution characteristics of the multispectral response signal and the target topological relationship in the wavelength dimension through an extraction module in the graph convolutional network;

[0010] Through the recognition module in the graph convolutional network, the origin classification boundary is dynamically updated based on the spatial distribution characteristics and the wavelength dimension, and the origin identification result of the authentic Chinese medicinal material sample is determined based on the updated origin classification boundary.

[0011] Optionally, dynamically updating the origin classification boundary based on the spatial distribution characteristics and the wavelength dimension through the recognition module in the graph convolutional network includes:

[0012] The recognition module in the graph convolutional network is used to integrate the regional density differences of the spatial distribution characteristics with the topological relationship evolution direction of the wavelength dimension to construct the evolution path of the origin characteristics;

[0013] Marking feature clusters that conflict with historical classification boundaries on the origin feature evolution path to generate a conflict feature identification set;

[0014] intercepting the offset segment in the origin feature evolution path according to the conflict feature identifier set, and reconstructing the origin decision tree branch;

[0015] The weight distribution of the branches of the origin decision tree is constrained by the timing window of the pulse alternating pattern to obtain an updated origin classification boundary.

[0016] Optionally, the method of fusing the regional density difference of the spatial distribution characteristics with the topological relationship evolution direction of the wavelength dimension through the recognition module in the graph convolutional network to construct the origin characteristic evolution path includes:

[0017] According to the spatial distribution characteristics, calculating the regional density difference of each spatial coordinate point of the authentic Chinese medicinal material sample on the two-dimensional surface to generate a spatial density difference set;

[0018] Track the evolution direction changes of topological nodes in the wavelength dimension and determine the evolution direction sequence of topological relationships;

[0019] Mapping the spatial density difference set to a topological node and assigning a spatial density weight value;

[0020] The node connection order of the topological relationship evolution direction sequence is adjusted according to the spatial density weight value to generate the origin characteristic evolution path.

[0021] Optionally, inputting the multidimensional features of each spectrum into a graph convolutional network, and extracting the spatial distribution characteristics of the multispectral response signal and the target topological relationship in the wavelength dimension through an extraction module in the graph convolutional network, includes:

[0022] Inputting the multidimensional features of each spectrum into a graph convolutional network, performing a spatial feature extraction operation through the extraction module, and generating a spatial distribution characteristic of the multispectral response signal;

[0023] Performing a wavelength dimension relationship extraction operation by the extraction module to generate an initial topological relationship of the wavelength dimension;

[0024] The spatial distribution characteristics are coupled with the initial topological relationship of the wavelength dimension to construct a joint feature topology, and the characteristic propagation path in the joint feature topology is analyzed to obtain the target topological relationship of the wavelength dimension.

[0025] Optionally, coupling the initial topological relationship between the spatial distribution characteristics and the wavelength dimension to construct a joint characteristic topology includes:

[0026] generating a spatial density distribution set according to the density distribution value of each region in the spatial distribution characteristic;

[0027] Screening the connection relationships whose node connection strengths in the initial topological relationships of the wavelength dimension are higher than a preset strength threshold, and generating a valid topological relationship sequence;

[0028] Extracting a density distribution value of each region from the spatial density distribution set and converting it into a spatial modulation weight value;

[0029] Adjusting the node connection strengths in the valid topological relationship sequence based on the spatial modulation weight value to reconstruct a node connection network;

[0030] The reconstructed node connection network is integrated with the spatial density distribution set to generate a joint feature topology.

[0031] Optionally, determining the origin identification result of the authentic Chinese medicinal material sample based on the updated origin classification boundary includes:

[0032] Extracting a descriptive attribute set from a preset knowledge base corresponding to the authentic Chinese medicinal material sample to generate a sample attribute set;

[0033] Generate available demarcation rules based on the updated origin classification boundaries and preset demarcation rules;

[0034] Comparing the sample attribute set with the available demarcation rule, and generating a position indication value by calculating an interaction value;

[0035] The origin identification result is determined based on matching the origin region label with the position indication value.

[0036] Optionally, extracting multidimensional features of each spectrum based on the multispectral response signal includes:

[0037] enhancing the intensity of the spectral response signal to generate an enhanced multi-spectral response signal;

[0038] The multidimensional features of each spectrum are extracted from the enhanced multispectral response signal, wherein the multidimensional features include peak value, intensity distribution and waveform feature.

[0039] In a second aspect, the present application provides a system for identifying the origin of authentic Chinese medicinal materials, comprising:

[0040] A generation module is used to synchronously excite authentic Chinese medicinal material samples using a spectral sensor array in a pulse alternating mode to generate a multi-spectral response signal, wherein the multi-spectrum includes a Raman spectrum and an ultraviolet spectrum;

[0041] An extraction module, configured to extract multidimensional features of each spectrum based on the multispectral response signal;

[0042] An input module is used to input the multidimensional features of each spectrum into a graph convolutional network, and extract the spatial distribution characteristics of the multispectral response signal and the target topological relationship in the wavelength dimension through an extraction module in the graph convolutional network;

[0043] An identification module is used to dynamically update the origin classification boundary based on the spatial distribution characteristics and wavelength dimension through the identification module in the graph convolutional network, and determine the origin identification result of the authentic Chinese medicinal material sample based on the updated origin classification boundary.

[0044] In a third aspect, the present application provides a computing device comprising a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a method for identifying the origin of authentic Chinese medicinal materials as described in any one of the first aspects.

[0045] In a fourth aspect, the present application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a method for identifying the origin of authentic Chinese medicinal materials as described in any one of the first aspects.

[0046] In the present application, a method for identifying the origin of authentic Chinese medicinal materials is provided, which includes: using a spectral sensor array to synchronously excite authentic Chinese medicinal material samples in a pulse alternating mode to generate a multi-spectral response signal, the multi-spectrum including a Raman spectrum and an ultraviolet spectrum; based on the multi-spectral response signal, extracting the multi-dimensional features of each spectrum; inputting the multi-dimensional features of each spectrum into a graph convolutional network, and extracting the spatial distribution characteristics and target topological relationship of the multi-spectral response signal in the wavelength dimension through an extraction module in the graph convolutional network; dynamically updating the origin classification boundary based on the spatial distribution characteristics and the wavelength dimension through a recognition module in the graph convolutional network, and determining the origin identification result of the authentic Chinese medicinal material sample based on the updated origin classification boundary.

[0047] This application uses a pulse-alternating spectral sensor array to synchronously collect Raman and ultraviolet multi-spectral response signals of authentic Chinese medicinal materials samples, and uses a graph convolutional network to deeply fuse spatial distribution characteristics and non-Euclidean correlations in wavelength dimensions to achieve adaptive extraction of topological features and dynamic optimization of classification boundaries, ultimately achieving high-precision and robust intelligent identification of origin.

[0048] Furthermore, the graph convolution recognition module is used to integrate the spatial regional density differences and the wavelength topology evolution direction to construct the origin feature evolution path; the path offset segments are intercepted based on the conflict feature identification set to reconstruct the decision tree branches, and the pulse timing window is used to dynamically constrain the branch weight distribution to achieve adaptive update of the classification boundary, further improving the accuracy of origin identification.

[0049] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0051] Figure 1 A flowchart of a method for identifying the origin of authentic Chinese medicinal materials provided in an embodiment of the present application;

[0052] Figure 2 A schematic diagram of a system for identifying the origin of authentic Chinese medicinal materials provided in an embodiment of the present application;

[0053] Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0055] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 11, 12, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0056] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0057] In order to solve the problem of low accuracy in origin identification caused by the loss of cross-scale spectral fingerprint information and lack of adaptability of classification boundaries in the existing technology, an embodiment of the present application provides a method for origin identification of authentic Chinese medicinal materials. The method adopts the following concept: to address the bottlenecks of difficulty in quantifying spatial heterogeneity and insufficient modeling of spectral cross-dimensional correlation in the identification of authentic Chinese medicinal materials, a three-layer progressive strategy of "data-driven-structure analysis-dynamic decision-making" is adopted. The spatial distribution of samples and Raman and ultraviolet multi-band responses are synchronously captured through a pulse alternating excitation mode to construct a three-dimensional matrix of original signals; the multi-dimensional features are mapped into a graph structure, and the graph convolution extraction module is used to explicitly deconstruct the spatial density field changes and wavelength topological chains; finally, the recognition module converts the spatial-spectral coupling relationship into a dynamic classification boundary update function, and optimizes the decision surface online to adapt to the drift of origin characteristics, forming a closed-loop feedback mechanism of "physical signals, topological relationships, and self-evolutionary decisions", realizing a subversive replacement for artificial feature engineering and static classification paradigms.

[0058] Figure 1 This is a flow chart of a method for identifying the origin of authentic Chinese medicinal materials provided in an embodiment of the present application, such as Figure 1 As shown, the method includes:

[0059] S11. Use a spectral sensor array to synchronously excite authentic Chinese medicinal material samples in a pulse alternating mode to generate a multi-spectral response signal, which includes a Raman spectrum and an ultraviolet spectrum.

[0060] Among them, authentic Chinese medicinal materials samples include but are not limited to: angelica samples, astragalus samples, Panax notoginseng samples, etc. The spectral sensor array may refer to a collection of multi-channel photodetectors arranged in a two-dimensional grid, which is used to synchronously collect spectral signals at different spatial locations on the sample surface. The pulse alternating mode may refer to an excitation method in which laser pulses and ultraviolet pulses are alternately triggered by a timing controller to avoid spectral crosstalk and improve the signal-to-noise ratio. The multi-spectral response signal may refer to a data cube containing three dimensions: spatial coordinates, wavelength, and intensity, which is generated by the fusion of the Raman scattering signal generated by laser excitation and the electron absorption signal generated by ultraviolet excitation. The Raman spectrum is a scattering spectrum that can reflect the characteristics of molecular vibration energy levels, and is obtained based on the frequency shift signal generated by the inelastic collision of the laser and the sample molecules. The ultraviolet spectrum may be an absorption spectrum used to reflect the characteristics of electronic transitions, and is obtained based on the absorbance change of the outer electron energy level transition of the molecule when irradiated with ultraviolet light.

[0061] In the present embodiment, a spectral sensor array arranged in a rectangular grid is first used to control the time-sharing triggering of a laser source and a UV light source using an alternating pulse pattern, synchronously stimulating different spatial locations on the surface of an authentic Chinese medicinal material sample. Secondly, molecular vibrational scattering signals are collected during the laser pulse period to generate a Raman spectrum, and electronic transition absorption signals are collected during the UV pulse period to generate a UV spectrum. Finally, the two spectra are integrated according to spatial coordinates into a multispectral response signal in the form of a three-dimensional data cube containing position, wavelength, and intensity information.

[0062] S12. Extracting multidimensional features of each spectrum based on the multispectral response signal.

[0063] The multidimensional features of each spectrum refer to a set of quantitative descriptors extracted from a single spectral curve, including the time-frequency domain wavelet coefficients constituting the frequency domain features, the first-order derivative sequence constituting the gradient features, and the peak parameters constituting the morphological features.

[0064] In the present embodiment, each independent spectral curve in the multispectral response signal is first preprocessed by filtering to smooth the noise and perform baseline correction. Secondly, a wavelet transform is used to extract the time-frequency domain energy coefficient as the frequency domain feature, and the spectral slope change characteristics are obtained as the gradient feature through first-order derivative calculations. Finally, a multidimensional feature vector is constructed by combining the peak wavelength position, full width at half maximum, and peak area integral. Each spatial location corresponds to a set of features including frequency domain features, gradient features, and morphological features.

[0065] S13. Input the multidimensional features of each spectrum into the graph convolutional network, and extract the spatial distribution characteristics of the multispectral response signal and the target topological relationship in the wavelength dimension through the extraction module in the graph convolutional network.

[0066] Graph convolutional networks (GCNs) refer to deep learning architectures for processing non-Euclidean data, modeling spatial correlations through node feature propagation and neighborhood aggregation. Spatial distribution characteristics are used to reflect the variation of spectral intensity between different sensor locations, extracted through neighborhood feature aggregation using graph convolution. The wavelength dimension characterizes the direction of electromagnetic wavelength variation and reflects the physical laws governing the evolution of characteristic peaks with wavelength. The target topology refers to the nonlinear connectivity between discrete characteristic peaks in the wavelength dimension. The coupling strength between peaks can be established using a learnable adjacency matrix.

[0067] In an embodiment of the present application, a graph structure is first constructed with spatial sites as nodes and adjacent sensor distances as edges, and the multidimensional feature vectors of each spectrum are input into the graph convolutional network as node attributes. Secondly, in the extraction module, the neighborhood node features are aggregated through multi-layer graph convolution operations, and the spatial distribution characteristics are fused using a message passing mechanism. At the same time, a dedicated topology learning layer is set in the wavelength dimension, and the nonlinear correlation between the feature peaks is modeled through a learnable adjacency matrix to form a target topological relationship that reflects the wavelength evolution law. Finally, a joint feature representation that fuses the spatial characteristics and the wavelength topology is output.

[0068] S14. Through the recognition module in the graph convolutional network, the origin classification boundary is dynamically updated based on the spatial distribution characteristics and wavelength dimension, and the origin identification result of the authentic Chinese medicinal material sample is determined based on the updated origin classification boundary.

[0069] Among them, the origin classification boundary refers to the decision hyperplane that divides different origin areas in the high-dimensional feature space, and its position is determined jointly by the spatial distribution characteristics and wavelength topology.

[0070] In an embodiment of the present application, an initial origin classification hyperplane is first constructed in the recognition module based on the spatial distribution characteristics. Secondly, the wavelength dimension target topological relationship is dynamically weighted through the topological attention mechanism, and the position and direction of the classification boundary are adjusted in real time according to the gradient backpropagation. Subsequently, the contribution weights of the spatial features and wavelength features are iteratively updated to make the classification boundary adaptive to the spatial density gradient changes. Finally, the authentic Chinese medicinal material samples to be identified are projected into the updated classification boundary space, and their origin identification results are determined according to the nearest neighbor principle.

[0071] The following is a specific example: First, a 32×32 spectral sensor array is used to alternately excite the surface of a Salvia miltiorrhiza sample with 100ms laser pulses and 50ms UV pulses. Raman and UV spectra are simultaneously collected at 256 spatial locations, generating a three-dimensional data cube containing spatial coordinates, wavelength, and intensity. Next, the spectrum of each location is subjected to wavelet denoising and baseline correction, and the frequency domain energy coefficient, first-order derivative series, and characteristic peak parameters are extracted to form a multidimensional feature vector. A spatial adjacency graph is then constructed, and spatial distribution patterns are captured by aggregating adjacent sensor features through three layers of graph convolution. Nonlinear topological associations between characteristic peaks are also learned in the wavelength dimension. Finally, the classification boundary is dynamically adjusted based on topological attention, and the direction of the decision hyperplane is updated according to changes in the spatial gradient to output the identification results of the Angelica sinensis sample.

[0072] By executing S11 to S14, the embodiment of the present application realizes the coordinated acquisition of Raman and ultraviolet spectra through a pulse alternating excitation mode, uses a graph convolutional network to simultaneously model the spatial distribution characteristics and wavelength dimension topological relationships, and dynamically optimizes the classification decision boundary, effectively overcoming the shortcomings of traditional three-dimensional convolution in cross-scale feature fusion and dynamic correlation modeling, and improving the accuracy and robustness of identification of the origin of authentic medicinal materials.

[0073] In a possible embodiment, S14, dynamically updating the origin classification boundary based on the spatial distribution characteristics and the wavelength dimension through the recognition module in the graph convolutional network, includes:

[0074] Step 141: The regional density difference of the spatial distribution characteristics and the topological relationship evolution direction of the wavelength dimension are integrated through the recognition module in the graph convolutional network to construct the evolution path of the origin characteristics.

[0075] Regional density difference refers to the degree of dispersion of the characteristic intensity distribution within each spatial subregion of the spectral sensor array, calculated based on the variance of the spectral intensity at all sites within the region. The direction of topological relationship evolution can refer to the changing trend of the correlation structure between characteristic peaks in the wavelength dimension, extracted through the eigenvalue decomposition of the topological relationship matrix. The origin characteristic evolution path can refer to the spatiotemporal trajectory that integrates spatial density gradients and wavelength topological changes, generated by the tensor product operation of the regional density difference vector and the topological evolution direction vector.

[0076] In this embodiment, the spatial density analysis unit of the recognition module first calculates the intensity distribution variance of different sensor regions to quantify the regional density differences in spatial distribution characteristics. Secondly, the eigenvectors of the topological relationship matrix are extracted in the wavelength dimension, and the trajectory of its principal component changes is determined as the topological relationship evolution direction. Subsequently, a tensor product operation is performed on the regional density difference vector and the topological relationship evolution direction vector to generate an origin characteristic evolution path that represents the spatiotemporal evolution of the origin characteristics.

[0077] Step 142: Mark the feature clusters that conflict with the historical classification boundaries on the origin feature evolution path to generate a conflict feature identification set.

[0078] The historical classification boundary refers to the origin division decision hyperplane determined by previous training data and stored in the weight parameters of the graph convolutional network recognition module. A conflicting feature cluster refers to a set of data points that are too close to the historical classification boundary in the feature evolution path, formed by determining the Euclidean distance threshold. A conflicting feature identifier set can refer to a structured dataset that records the spatial coordinates, cluster size, and boundary deviation of the conflicting feature cluster, and is used to mark the path correction area. The conflicting feature identifier set is a set consisting of the identifiers of each feature in the conflicting feature cluster, with each feature corresponding to a feature identifier.

[0079] In this embodiment, the evolution path of origin characteristics is first projected onto the decision space defined by the historical classification boundary. A density clustering algorithm is then used to identify clusters of feature points along the path whose distance from the historical classification boundary is less than a threshold. These clusters are labeled as conflicting feature clusters. The deviation between the centroid coordinates of each conflicting feature cluster and the boundary is then calculated to generate a conflicting feature identification set containing a cluster identifier, spatial location, and deviation value.

[0080] Step 143: intercept the offset segment in the origin feature evolution path according to the conflict feature identification set, and reconstruct the origin decision tree branch.

[0081] The "offset segment" refers to a continuous segment in the evolution path of origin characteristics that deviates from the historical boundary. It is generated based on the deviation index of the conflicting feature identification set. The origin decision tree branch refers to the decision subtree reconstructed for a specific origin characteristic offset, constructed using key wavelength nodes as split points and the Gini coefficient as the division criterion.

[0082] In an embodiment of the present application, first, based on the deviation value recorded in the conflicting feature identification set, the continuous segment in the origin feature evolution path whose deviation from the historical classification boundary exceeds a preset deviation threshold is located as an offset segment; then, the key wavelength node with the most drastic gradient change in the offset segment is used as the splitting point, and the recursive feature elimination method is used to screen the highly discriminative feature subset associated with the node; finally, the origin decision tree branch is reconstructed based on the feature subset, and the classification threshold of the branch node is dynamically optimized through the principle of minimizing the Gini coefficient, so that the reconstructed branch accurately adapts to the feature distribution law of the offset segment, thereby eliminating the conflict between the original classification boundary and the evolution path.

[0083] Step 144 : Constrain the weight distribution of the origin decision tree branches through the timing window of the pulse alternating pattern to obtain an updated origin classification boundary.

[0084] In this embodiment, a time window of laser and UV cycles in an alternating pulse pattern is first established, and the weight distribution of the origin decision tree branches is constrained to the activation period of the corresponding spectral type. Second, high-order weights are applied to the Raman feature branches during the laser window period, and high-order weights are applied to the UV feature branches during the UV window period. Finally, the decision tree branches are reconstructed through weighted fusion, and the dynamically adjusted origin classification boundaries are output.

[0085] The following is another specific example: First, the variance of the spectral intensity distribution of astragalus samples in the northeastern and southwestern regions is calculated through the recognition module, and the origin feature evolution path is generated by combining the principal component direction of the Raman feature peak correlation matrix. Secondly, the path is mapped to the historical boundary space, and three groups of feature clusters that are too close to the boundary are identified and the centroid coordinates are recorded to form a conflict feature identification set. Then, the continuous segments with excessive deviation values ​​in the path are intercepted, and the decision tree branches are reconstructed as splitting points using the Raman peaks. Finally, during the laser pulse window period, the Raman branch weight is increased to three times the ultraviolet branch weight, and the reverse weight configuration is performed during the ultraviolet pulse window period to generate the updated classification boundary between Sichuan and Longnan products.

[0086] By executing steps 141 to 144, the embodiment of the present application dynamically identifies the conflicting areas between the feature evolution path and the historical boundary, specifically reconstructs the decision tree branches and integrates the pulse timing constraints, so that the classification boundary adaptively responds to the dynamic changes of the spatial-spectral coupling relationship, and effectively improves the pattern resolution ability and identification stability of complex origin characteristics.

[0087] In one possible embodiment, step 141, fusing the regional density difference of spatial distribution characteristics with the topological relationship evolution direction of the wavelength dimension through the recognition module in the graph convolutional network to construct the origin characteristic evolution path, includes:

[0088] Step a1: Calculate the regional density difference of each spatial coordinate point of the authentic Chinese medicinal material sample on the two-dimensional surface according to the spatial distribution characteristics to generate a spatial density difference set.

[0089] The regional density difference refers to the absolute deviation between the average spectral intensity of a specific spatial coordinate point and all adjacent points within its neighborhood. It is used to quantify the uneven distribution of local features. The spatial density difference set is a data structure that stores the regional density difference of each spatial coordinate point on a two-dimensional surface, including a table mapping position coordinates to density differences.

[0090] In this embodiment, spectral intensity distribution data is first obtained for all spatial coordinate points on the two-dimensional surface of an authentic Chinese medicinal material sample. Next, with each coordinate point as the center, the average intensity difference between it and adjacent coordinate points within a preset neighborhood is calculated as the regional density difference. Finally, the regional density differences for all spatial coordinate points are encoded and stored according to their positional coordinates, generating a spatial density difference set containing spatial location identifiers and corresponding density differences.

[0091] Step a2: Track the evolution direction changes of topological nodes in the wavelength dimension and determine the evolution direction sequence of topological relationships.

[0092] A topological node is an abstract computational unit that represents the association between characteristic peak clusters in the wavelength dimension. It is a mathematical representation of the coupling relationship between spectral characteristic peaks extracted after dimensionality reduction through principal component analysis. A topological relationship evolution direction sequence can be an array that continuously records the angular changes of the main eigenvectors of topological nodes along the wavelength direction, reflecting the evolution trajectory of characteristic association patterns.

[0093] In this embodiment, we first extract a set of topological nodes representing characteristic peak relationships in the wavelength dimension. Next, we use principal component analysis to track the directional change angles of the eigenvectors of adjacent topological nodes along increasing wavelengths. We then record the offset angle sequence of the principal eigenvectors within each wavelength segment to construct a topological relationship evolution direction sequence that reflects the evolutionary trend of the characteristic correlation structure.

[0094] Step a3: Map the spatial density difference set to the topological node and assign a spatial density weight value.

[0095] The spatial density weight value refers to a normalized weight coefficient calculated based on the spatial density difference, which is used to measure the intensity of the spatial distribution influence on a specific topological node. The mapping and allocation logic is to establish a physical association between spatial coordinate points and topological nodes through the mapping step. The allocation step, based on this association, converts the normalized spatial density difference into a weight coefficient for the topological node. The two form a continuous calculation chain of "position binding, density transmission, and weight quantization," ensuring that spatial heterogeneity information is accurately injected into the wavelength-dimensional topological structure.

[0096] In this embodiment, the density differences in the spatial density difference set are first normalized into a probability distribution. The normalized density values ​​are then used as weight coefficients and assigned to corresponding topological nodes based on the mapping relationship between spatial coordinates and topological nodes. Finally, the spatial density weight of each topological node is calculated through weighted summation.

[0097] Step a4: Adjust the node connection order of the topological relationship evolution direction sequence according to the spatial density weight value to generate the origin characteristic evolution path.

[0098] Among them, the node connection order refers to the arrangement order of topological nodes in the feature evolution path, which is dynamically determined by the size relationship of the spatial density weight value.

[0099] In this embodiment, nodes in the topological relationship evolution direction sequence are first prioritized according to their spatial density weights. A weight threshold screening mechanism is then employed to remove topological node connections with spatial density weights below a threshold. The remaining nodes are then reconnected in descending order of weight to generate a characteristic evolution path for the origin that incorporates the spatial density distribution characteristics.

[0100] Here is another specific example: First, calculate the average intensity difference between 256 coordinate points on the surface of the Panax notoginseng leaf and its eight neighborhoods to form a spatial density difference set. Then, extract 12 topological nodes corresponding to the Raman characteristic peaks in the wavelength dimension, along the 1000-1600 cm - ¹ The directional changes of the principal components are tracked across the entire range to generate an evolutionary direction sequence. The density differences in the leaf tip region, where spatial density differences are concentrated, are then converted into weights and assigned to corresponding topological nodes. Finally, the node connections are reordered based on the weights to construct a characteristic evolutionary path that enhances leaf tip density characteristics for identification of producing regions.

[0101] By executing steps a1 to a4, the embodiment of the present application dynamically adjusts the construction logic of the wavelength dimension topological evolution path through spatial density differences, so that the feature evolution process adaptively integrates the distribution characteristics of the sample surface, thereby enhancing the ability of the origin identification model to express the heterogeneous characteristics of medicinal materials.

[0102] In one possible embodiment, S13, inputting the multidimensional features of each spectrum into a graph convolutional network, and extracting the spatial distribution characteristics of the multispectral response signal and the target topological relationship in the wavelength dimension through an extraction module in the graph convolutional network, including:

[0103] Step 131: Input the multidimensional features of each spectrum into the graph convolutional network, perform spatial feature extraction operations through the extraction module, and generate the spatial distribution characteristics of the multispectral response signal.

[0104] Among them, spatial feature extraction refers to the process of aggregating features of adjacent spatial sites using graph convolution operations, including steps such as neighborhood node feature aggregation and multi-scale feature fusion, which is used to capture the spectral distribution pattern of the sample surface.

[0105] In this embodiment, the multidimensional feature vectors of each spectrum are first loaded as node attributes into the extraction module of a graph convolutional network. Next, the connectivity between sensor nodes is defined using a spatial adjacency matrix, and multi-layer graph convolution operations are used to aggregate the feature information of each node and its neighbors. Skip connections are then used to fuse spatial features at different scales, ultimately outputting spatial distribution characteristics that characterize the spectral intensity distribution of the sample surface.

[0106] Step 132: Perform wavelength dimension relationship extraction operation through the extraction module to generate an initial topological relationship of the wavelength dimension.

[0107] Among them, the initial topological relationship refers to the initial correlation structure between characteristic peaks constructed through adaptive learning in the wavelength dimension, which includes node connection weights and direction information, reflecting the basic coupling strength of the characteristic peaks.

[0108] In this embodiment, a characteristic peak node network is first constructed in the wavelength dimension, with the characteristic peak center wavelength as the node position. Next, the topology learning layer calculates the correlation strength between any two characteristic peak nodes to form an initial adjacency matrix. A graph attention mechanism is then used to adaptively adjust the connection weights between nodes to generate an initial topological relationship that reflects the nonlinear coupling law of the characteristic peaks.

[0109] Step 133: Couple the initial topological relationship of the spatial distribution characteristics with the wavelength dimension, construct a joint feature topology, analyze the characteristic propagation path in the joint feature topology, and obtain the target topological relationship of the wavelength dimension.

[0110] Joint feature topology refers to a heterogeneous graph structure that integrates spatially distributed feature nodes with wavelength feature nodes, achieving the joint expression of spatial and spectral features through cross-dimensional edge connections. Feature propagation paths refer to the sequence of trajectories that transmit feature information between nodes in the joint feature topology. The effectiveness of information transmission is evaluated based on path weights and connection depth.

[0111] In this embodiment, spatial distribution characteristics are first converted into node feature enhancement vectors and injected into the initial topological relationship nodes in the wavelength dimension. Next, a joint feature topological graph containing nodes in both the spatial and wavelength dimensions is constructed, and cross-dimensional associations are integrated through edge joins. A random walk algorithm is then executed to analyze the feature propagation paths between nodes. Key connection relationships are selected based on path weights, ultimately generating an optimized target topological relationship in the wavelength dimension.

[0112] The following is a specific example: First, the multidimensional spectral features of the sample's 128 spatial coordinate points are input into a graph convolutional network. Three layers of graph convolution aggregate features of adjacent sites to generate spatial distribution characteristics. Next, a network of 10 Raman characteristic peak nodes is constructed in the wavelength dimension. The inter-peak correlation strength is calculated using an attention mechanism to form an initial topological relationship. The spatial distribution characteristics are then converted into node enhancement vectors and injected into the wavelength nodes to establish a joint feature topology connected across dimensions. Finally, a random walk is used to analyze feature propagation paths, screening for high-weighted paths and constructing the target topological relationship for producing area identification.

[0113] By executing steps 131 to 133, the embodiment of the present application achieves cross-dimensional fusion of spatial distribution and wavelength association through a joint feature topology, optimizes the topological structure by analyzing the characteristic propagation path, effectively improves the characterization capability of spectral fingerprint features, and enhances the accuracy of the origin identification model in capturing complex nonlinear relationships. In one possible embodiment, step 133, coupling the initial topological relationship of spatial distribution characteristics and wavelength dimensions to construct a joint feature topology, includes:

[0114] Step b1: Generate a spatial density distribution set according to the density distribution value of each area in the spatial distribution characteristics.

[0115] The spatial density distribution set refers to a data set storing the mean spectral intensity of spatial sub-regions, and includes a mapping relationship table between regional location identifiers and density distribution values.

[0116] In this embodiment, multiple spatial subregions are first obtained based on the spatial distribution characteristics. Next, the average spectral intensity of all sensor locations within each subregion is calculated as the density distribution value for that region. Finally, a mapping relationship is established between the spatial location identifier of each region and the corresponding density distribution value to form a spatial density distribution set.

[0117] Step b2: Filter the connection relationships in the initial topological relationship of the wavelength dimension whose node connection strength is higher than a preset strength threshold to generate a valid topological relationship sequence.

[0118] The preset strength threshold refers to the critical value used to screen topological connection relationships, which is set based on the statistical distribution of effective connection strengths in historical data. The effective topological relationship sequence refers to the ordered set of strong connection relationships retained in the wavelength dimension, including node pair identifiers and normalized connection strength values.

[0119] In this embodiment, the adjacency matrix data for the initial topological relationships in the wavelength dimension is first read. Next, the connection strength values ​​between all nodes in the matrix are traversed, and valid connections exceeding a preset strength threshold are selected. The valid connections are then sorted by wavelength to generate a valid topological relationship sequence containing connection strengths and node identifiers.

[0120] Step b3: extract the density distribution value of each region from the spatial density distribution set and convert it into a spatial modulation weight value.

[0121] The spatial modulation weight value refers to a normalization coefficient converted based on the spatial density distribution value, and is used to adjust the importance weight of the wavelength node in the network.

[0122] In the present embodiment, the density distribution value of each region is first extracted from the spatial density distribution set. Then, the density distribution value is linearly converted to the range of zero to one using the maximum and minimum value normalization method. Finally, the normalized value is assigned to the corresponding region as the spatial modulation weight value.

[0123] Step b4: adjusting the node connection strength in the effective topological relationship sequence based on the spatial modulation weight value to reconstruct the node connection network.

[0124] Node connection strength refers to the quantitative value of the correlation between two characteristic peak nodes in the topological relationship, which is calculated by attention weight. The node connection network refers to the graph structure composed of wavelength characteristic peak nodes and the connection relationships between nodes. The connection strength is represented by the adjacency matrix.

[0125] In this embodiment of the present application, the modulation coefficient of the wavelength node is first determined based on the spatial modulation weight value. Next, the node connection strength in the valid topological relationship sequence is multiplied by the modulation coefficient of the corresponding node. The adjacency matrix is ​​then reconstructed based on the adjusted connection strength to complete the reconstruction of the node connection network.

[0126] Step b5: Integrate the reconstructed node connection network and the spatial density distribution set to generate a joint feature topology.

[0127] In this embodiment of the present application, the reconstructed node connection network is first represented as a weighted adjacency matrix. Secondly, the spatial density distribution set is converted into node feature vectors. Finally, the adjacency matrix and feature vectors are fused through a feature concatenation operation to generate a joint feature topology containing both spatial and wavelength bimodal information.

[0128] The following is a specific example: First, the angelica sample is divided into eight spatial regions and the average intensity value of each region is calculated to generate a spatial density distribution set. Next, six groups of node pairs with connection strength exceeding the threshold in the initial topology of the Raman characteristic peak are selected to form a valid topological relationship sequence. The spatial density value is then converted into a weight coefficient, and the weight value of the petiole region is increased to twice that of the leaf region. The node connection strength of the valid sequence is then modulated to reconstruct a node connection network that strengthens the petiole characteristics. Finally, the reconstructed network is integrated with the spatial density data to generate a joint feature topology for identification of the Minxian production area.

[0129] By executing steps b1 to b5, the embodiment of the present application dynamically modulates the wavelength topological connection strength through spatial density weights, so that the joint feature topology adaptively fuses the surface distribution characteristics of the sample, enhances the model's ability to analyze non-uniform medicinal material characteristics, and improves the robustness of the origin identification results.

[0130] In a possible embodiment, S14, determining the origin identification result of the authentic Chinese medicinal material sample based on the updated origin classification boundary, includes:

[0131] Step c1: extracting a descriptive attribute set from a preset knowledge base corresponding to authentic Chinese medicinal material samples to generate a sample attribute set.

[0132] The pre-set knowledge base refers to a structured database storing standard information on authentic Chinese medicinal materials, including authoritative data sources such as plant morphological characteristics, ecological and environmental parameters, and historical distribution of origins. The descriptive attribute set refers to a set of sample characteristic description entries extracted from the knowledge base, including qualitative and quantitative attributes such as leaf morphology, texture characteristics, and active ingredient content. The sample attribute set refers to the integrated sample feature dataset, structured and stored according to three attributes: morphological characteristics, chemical composition, and biomarkers.

[0133] In this embodiment, a pre-set knowledge base storing prior knowledge about authentic Chinese medicinal materials is first accessed to extract descriptive attribute entries relevant to the current sample. Next, key attribute entries, including morphological characteristics, growth environment, and harvest season, are screened and categorized by attribute type. Finally, a sample attribute set containing multidimensional descriptive information is generated.

[0134] Step c2: Generate available demarcation rules based on the updated origin classification boundaries and preset demarcation rules.

[0135] Preset demarcation rules refer to the constraints for origin demarcation developed based on experience in medicinal materials science, including geographical feature rules such as altitude gradient thresholds and soil type matching. Available demarcation rules refer to a subset of valid classification boundaries verified by preset rules, retaining the decision hyperplane equation that conforms to the laws of medicinal materials science.

[0136] In this embodiment, the updated parameterized expression for the origin classification boundary is first read. Next, the validity of the classification boundary conforming to the rules is verified based on the logical constraints in the pre-set boundary rule library. Boundary segments that violate the constraints are then removed, generating a usable boundary rule set that is compatible with the rule library.

[0137] Step c3: compare the sample attribute set with the available demarcation rules, and generate a position indication value by calculating the interaction value.

[0138] Among them, the interaction value refers to the dot product calculation result of the sample attribute vector and the classification hyperplane normal vector, reflecting the algebraic distance from the sample to the classification boundary. The position indicator value refers to the normalized relative position coordinate, which is obtained by dividing the interaction value by the maximum classification interval, and the value range is zero to one. Calculating the interaction value in this application means the algebraic distance of the sample to the classification decision hyperplane in the feature space. The core output of the comparison processing is the interaction value, which realizes the quantitative comparison of sample attributes and classification rules through vector dot product.

[0139] In step c3, the comparison process involves matching the sample attribute set against the applicable demarcation rules. The result is a quantitative match value, indicating the degree to which the sample attributes conform to the demarcation rules. Specifically, the comparison result can be used to generate a location indicator value by calculating an interaction value to convert the match value into a decision output. The location indicator value ultimately represents the sample's origin classification, thereby enabling automated determination of the origin of authentic Chinese medicinal materials.

[0140] In the present embodiment, the sample attribute set is first vectorized into a feature matrix. The available demarcation rules are then converted into a decision hyperplane equation. The dot product between the feature matrix and the decision hyperplane is then calculated as the interaction value, and a position indicator value representing the relative position of the sample in the classification space is generated through normalization.

[0141] Step c4: Match the origin region label based on the location indication value to determine the origin identification result.

[0142] In this embodiment of the present application, a mapping relationship table between location indicator values ​​and origin region labels is first established. Next, a corresponding geographic region code is matched based on the numerical range of the location indicator value. Finally, the geographic region code is output as the origin identification result of the authentic Chinese medicinal material sample.

[0143] The following is a specific example: First, the leaf serration characteristics and taproot diameter data of the Angelica sinensis sample are extracted from the knowledge base to form a sample attribute set. Next, the updated classification boundary is verified to comply with the pre-set demarcation rule for altitudes above 2,000 meters, generating a usable demarcation rule. Next, the dot product between the sample attribute and the classification hyperplane is calculated to obtain an interaction value, which is converted into a location indicator value of 0.73. Finally, the regional label of the origin of the western Qinling Mountains is matched to output the identification result of the Angelica sinensis sample.

[0144] By executing steps c1 to c4, the embodiment of the present application establishes an explainable origin decision-making mechanism through the regular integration of knowledge base attributes and dynamic classification boundaries, thereby enhancing the traceability of the results while ensuring the accuracy of identification, and effectively overcoming the decision-making black box defect of the pure data-driven model.

[0145] In a possible embodiment, S12, extracting multidimensional features of each spectrum based on the multispectral response signal, includes:

[0146] Step 121: Enhance the intensity of the spectral response signal to generate an enhanced multi-spectral response signal.

[0147] Among them, the enhanced multispectral response signal refers to the spectral data cube that has undergone noise suppression and detail enhancement. It is obtained by processing the original signal based on adaptive gain control and nonlinear transformation, retaining the correlation between the space-wavelength dimension while improving the effective information intensity.

[0148] In this embodiment, adaptive gain control is first performed on the multispectral response signal, dynamically adjusting the gain coefficient by calculating the mean and variance of the signal intensity within a local window. Next, a nonlinear transformation function is used to exponentially enhance the low-intensity segments while compressing the dynamic range of the high-intensity segments. Finally, a multi-scale fusion technique is used to integrate the enhancement results at different resolutions, generating an enhanced multispectral response signal with improved signal-to-noise ratio and preserved detail.

[0149] Step 122: extracting multidimensional features of each spectrum from the enhanced multispectral response signal, where the multidimensional features include peak value, intensity distribution, and waveform features.

[0150] Among them, the peak value can refer to the quantitative description of the local intensity maximum point in the spectral curve, including the precise wavelength position and absolute intensity value corresponding to the point, reflecting the core identification information of the characteristic peak. The intensity distribution refers to the set of statistical characteristics of the spectral signal in the spatial or wavelength dimension, including the distribution parameters such as the intensity mean, variance, and skewness of each sampling point, which characterize the diffusion pattern of the spectral energy. The waveform feature refers to a comprehensive indicator that describes the morphological changes of the spectral curve. It is constructed based on the frequency band energy distribution of the wavelet transform and the slope change statistics of the first-order derivative curve, reflecting the overall profile characteristics of the characteristic peak.

[0151] In this embodiment, adaptive gain control is first performed on the multispectral response signal, dynamically adjusting the gain coefficient by calculating the mean and variance of the signal intensity within a local window. Next, a nonlinear transformation function is used to exponentially enhance the low-intensity segments while compressing the dynamic range of the high-intensity segments. Finally, a multi-scale fusion technique is used to integrate the enhancement results at different resolutions, generating an enhanced multispectral response signal with improved signal-to-noise ratio and preserved detail.

[0152] The following is a specific example: First, dynamic gain adjustment is performed on the multi-spectral response signal of the Panax notoginseng sample, and a triple gain coefficient is used to enhance the weak signal in the vein area to generate an enhanced multi-spectral response signal. - Identify the Raman characteristic peak at ¹ and record the peak intensity, calculate the intensity mean distribution along the long axis of the blade. Finally, extract the 3400-3600cm - The wavelet energy coefficient and first-order derivative variation coefficient of the segment ¹ constitute a multidimensional feature vector for subsequent graph convolution processing.

[0153] By executing steps 121 to 122, the embodiment of the present application optimizes the spectral quality through adaptive signal enhancement, combines multi-dimensional feature extraction to comprehensively capture spectral fingerprint information, effectively improves the robustness and discrimination of feature representation, and provides a high-fidelity data foundation for subsequent topological relationship construction.

[0154] Figure 2This is a schematic diagram of a system for identifying the origin of authentic Chinese medicinal materials provided in an embodiment of the present application, as shown in FIG. Figure 2 As shown, the system includes:

[0155] The generating module 21 is used to use a spectral sensor array to synchronously excite the authentic Chinese medicinal material sample in a pulse alternating mode to generate a multi-spectral response signal, where the multi-spectrum includes a Raman spectrum and an ultraviolet spectrum.

[0156] The extraction module 22 is used to extract the multi-dimensional features of each spectrum based on the multi-spectral response signal.

[0157] The input module 23 is used to input the multidimensional features of each spectrum into the graph convolutional network, and extract the spatial distribution characteristics of the multispectral response signal and the target topological relationship in the wavelength dimension through the extraction module in the graph convolutional network.

[0158] The updating module 24 is used to dynamically update the origin classification boundary based on the spatial distribution characteristics and wavelength dimension through the recognition module in the graph convolutional network, and determine the origin identification result of the authentic Chinese medicinal material sample based on the updated origin classification boundary.

[0159] Figure 2 The system for identifying the origin of authentic Chinese medicinal materials can be used to Figure 1 The implementation principle and technical effects of the method for identifying the origin of authentic Chinese medicinal materials described in the illustrated embodiment will not be elaborated on here. The specific manner in which each module and unit performs operations in the system for identifying the origin of authentic Chinese medicinal materials in the above embodiment has been described in detail in the embodiment of the method and will not be elaborated on here.

[0160] In one possible design, Figure 2 The system for identifying the origin of authentic Chinese medicinal materials in the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32 .

[0161] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0162] The processing component 32 is used to: use a spectral sensor array to synchronously excite authentic Chinese medicinal material samples in a pulse alternating mode to generate a multi-spectral response signal, where the multi-spectrum includes a Raman spectrum and an ultraviolet spectrum; based on the multi-spectral response signal, extract the multi-dimensional features of each spectrum; input the multi-dimensional features of each spectrum into a graph convolutional network, and extract the spatial distribution characteristics and target topological relationship of the multi-spectral response signal in the wavelength dimension through the extraction module in the graph convolutional network; dynamically update the origin classification boundary based on the spatial distribution characteristics and the wavelength dimension through the recognition module in the graph convolutional network, and determine the origin identification result of the authentic Chinese medicinal material sample based on the updated origin classification boundary.

[0163] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0164] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as random access memory (RAM), static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0165] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0166] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0167] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0168] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0169] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The embodiment shown is a method for identifying the origin of authentic Chinese medicinal materials.

[0170] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0171] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0172] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0173] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for identifying the origin of authentic Chinese medicinal materials, characterized in that: include: A spectral sensor array is used to synchronously excite authentic Chinese medicinal material samples in a pulse alternating mode to generate a multi-spectral response signal, which includes Raman and UV spectra. extracting multidimensional features of each spectrum based on the multispectral response signal; Inputting the multidimensional features of each spectrum into a graph convolutional network, and extracting the spatial distribution characteristics of the multispectral response signal and the target topological relationship in the wavelength dimension through an extraction module in the graph convolutional network; Dynamically updating the origin classification boundary based on the spatial distribution characteristics and wavelength dimension through the recognition module in the graph convolutional network, and determining the origin identification result of the authentic Chinese medicinal material sample based on the updated origin classification boundary; The method of dynamically updating the origin classification boundary based on the spatial distribution characteristics and the wavelength dimension through the recognition module in the graph convolutional network includes: The recognition module in the graph convolutional network is used to integrate the regional density differences of the spatial distribution characteristics with the topological relationship evolution direction of the wavelength dimension to construct the evolution path of the origin characteristics; Marking feature clusters that conflict with historical classification boundaries on the origin feature evolution path to generate a conflict feature identification set; intercepting the offset segment in the origin feature evolution path according to the conflict feature identifier set, and reconstructing the origin decision tree branch; The weight distribution of the branches of the origin decision tree is constrained by the timing window of the pulse alternating pattern to obtain an updated origin classification boundary.

2. The method according to claim 1, characterized in that The recognition module in the graph convolutional network integrates the regional density difference of the spatial distribution characteristics with the topological relationship evolution direction of the wavelength dimension to construct the origin characteristic evolution path, including: According to the spatial distribution characteristics, calculating the regional density difference of each spatial coordinate point of the authentic Chinese medicinal material sample on the two-dimensional surface to generate a spatial density difference set; Track the evolution direction changes of topological nodes in the wavelength dimension and determine the evolution direction sequence of topological relationships; Mapping the spatial density difference set to a topological node and assigning a spatial density weight value; The node connection order of the topological relationship evolution direction sequence is adjusted according to the spatial density weight value to generate the origin characteristic evolution path.

3. The method according to claim 1, characterized in that Inputting the multidimensional features of each spectrum into a graph convolutional network, and extracting the spatial distribution characteristics of the multispectral response signal and the target topological relationship in the wavelength dimension through an extraction module in the graph convolutional network, includes: Inputting the multidimensional features of each spectrum into a graph convolutional network, performing a spatial feature extraction operation through the extraction module, and generating a spatial distribution characteristic of the multispectral response signal; Performing a wavelength dimension relationship extraction operation by the extraction module to generate an initial topological relationship of the wavelength dimension; The spatial distribution characteristics are coupled with the initial topological relationship of the wavelength dimension to construct a joint feature topology, and the characteristic propagation path in the joint feature topology is analyzed to obtain the target topological relationship of the wavelength dimension.

4. The method according to claim 3, characterized in that The coupling of the spatial distribution characteristics with the initial topological relationship of the wavelength dimension to construct a joint characteristic topology includes: generating a spatial density distribution set according to the density distribution value of each region in the spatial distribution characteristic; Screening the connection relationships whose node connection strengths in the initial topological relationships of the wavelength dimension are higher than a preset strength threshold, and generating a valid topological relationship sequence; Extracting a density distribution value of each region from the spatial density distribution set and converting it into a spatial modulation weight value; Adjusting the node connection strengths in the valid topological relationship sequence based on the spatial modulation weight value to reconstruct a node connection network; The reconstructed node connection network is integrated with the spatial density distribution set to generate a joint feature topology.

5. The method according to claim 1, characterized in that Determining the origin identification result of the authentic Chinese medicinal material sample based on the updated origin classification boundary includes: Extracting a descriptive attribute set from a preset knowledge base corresponding to the authentic Chinese medicinal material sample to generate a sample attribute set; Generate available demarcation rules based on the updated origin classification boundaries and preset demarcation rules; Comparing the sample attribute set with the available demarcation rule, and generating a position indication value by calculating an interaction value; The origin identification result is determined based on matching the origin region label with the position indication value.

6. The method according to claim 1, wherein The extracting of multidimensional features of each spectrum based on the multispectral response signal includes: enhancing the intensity of the spectral response signal to generate an enhanced multi-spectral response signal; The multidimensional features of each spectrum are extracted from the enhanced multispectral response signal, wherein the multidimensional features include peak value, intensity distribution and waveform feature.

7. A system for identifying the origin of authentic Chinese medicinal materials, characterized in that: include: A generation module is used to synchronously excite authentic Chinese medicinal material samples using a spectral sensor array in a pulse alternating mode to generate a multi-spectral response signal, wherein the multi-spectrum includes a Raman spectrum and an ultraviolet spectrum; An extraction module, configured to extract multidimensional features of each spectrum based on the multispectral response signal; An input module is used to input the multidimensional features of each spectrum into a graph convolutional network, and extract the spatial distribution characteristics of the multispectral response signal and the target topological relationship in the wavelength dimension through an extraction module in the graph convolutional network; An updating module, configured to dynamically update the origin classification boundary based on the spatial distribution characteristics and the wavelength dimension through the recognition module in the graph convolutional network, and determine the origin identification result of the authentic Chinese medicinal material sample based on the updated origin classification boundary; The method of dynamically updating the origin classification boundary based on the spatial distribution characteristics and the wavelength dimension through the recognition module in the graph convolutional network includes: The recognition module in the graph convolutional network is used to integrate the regional density differences of the spatial distribution characteristics with the topological relationship evolution direction of the wavelength dimension to construct the evolution path of the origin characteristics; Marking feature clusters that conflict with historical classification boundaries on the origin feature evolution path to generate a conflict feature identification set; intercepting the offset segment in the origin feature evolution path according to the conflict feature identifier set, and reconstructing the origin decision tree branch; The weight distribution of the branches of the origin decision tree is constrained by the timing window of the pulse alternating pattern to obtain an updated origin classification boundary.

8. A computing device, characterized in that It comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement the method for identifying the origin of authentic Chinese medicinal materials as described in any one of claims 1 to 6.

9. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, the method for identifying the origin of an authentic Chinese medicinal material according to any one of claims 1 to 6 is implemented.

Images