A multi-view-based traditional Chinese medicinal material identification method and system

Abstract

This invention discloses a method and system for identifying Chinese medicinal materials based on multiple views, belonging to the field of Chinese medicinal material identification and intelligent information processing technology. First, this invention constructs a low-rank tensor-constrained multi-view subspace clustering model, mining higher-order correlations in multi-view Chinese medicinal material data through tensor kernel norms. Based on this, a hypergraph regularization term is introduced to mine the local manifold structure of the data, and simultaneously, sparse constraint terms and Frobenius norm constraint terms are constructed to strengthen the block diagonal structure of the subspace representation matrix, enhancing the separability of data points between classes and the consistency of data points within classes. Finally, the augmented Lagrange multiplier method is used iteratively to solve the problem, and spectral clustering is used to output the identification results. This invention overcomes the shortcomings of existing technologies that neglect intra-class consistency due to excessive focus on view diversity. By significantly improving the clustering compactness and accuracy in Chinese medicinal material identification scenarios through Frobenius norm constraints, it achieves the best performance in multiple evaluation metrics compared to existing methods.

Description

Technical Field

[0001] This invention relates to the field of Chinese medicinal herb identification technology, and in particular to a method and system for identifying Chinese medicinal herbs based on multiple views. Background Technology

[0002] With the rapid development of information technology, the traditional Chinese medicine (TCM) industry is generating massive amounts of data daily. In the context of big data, the sources of TCM data are often diverse, data quality varies greatly, and dimensionality is high, making traditional TCM identification methods ineffective. Accurate and efficient TCM identification technology is a crucial foundation for ensuring the safety and efficacy of TCM. Several technical challenges remain in effectively extracting valuable information from TCM data: First, existing methods cannot effectively aggregate similar TCMs and separate different types simultaneously; second, the extraction of discriminative information from TCM data is insufficient, as existing methods typically do not consider the essential identification information within the data.

[0003] Existing multi-view recognition methods primarily focus on mining complementary information between views. These methods build upon the low-rank tensor constraint (LT-MSC) framework by introducing manifold regularization, sparsity constraints, and diversity regularization (HSIC). The diversity regularization term aims to capture complementary information between different views, enhancing the model's generalization ability by maximizing the non-independence of view representations.

[0004] However, when applying this method to the identification of Chinese medicinal materials, the inventors discovered the following technical shortcomings in the existing technology:

[0005] 1. Insufficient consideration of both intra-class consistency and inter-class sparsity: The task of identifying Chinese medicinal materials requires highly clustered feature representations for samples of the same class (such as the same medicinal material under different lighting conditions), while the feature representations of samples of different classes (such as two medicinal materials with similar morphology) are significantly separated. In existing methods, the diversity regularization term (HSIC) focuses on preserving the differences in the global view, but neglects to impose a direct constraint of "intra-class compactness" on the subspace representation matrix. This results in relatively loose clusters of samples of the same class in fine-grained identification tasks, which is prone to misclassification.

[0006] 2. Limited robustness to data noise and redundant information: Images of Chinese medicinal materials often suffer from noise interference such as cluttered backgrounds and similar textures. While enhancing view diversity, the HSIC term may amplify redundant noise between views, interfering with the extraction of essential low-rank structures.

[0007] 3. The optimization solution has high complexity: The HSIC term involves the calculation of the kernel matrix, which increases the computational overhead of the objective function in the iterative optimization process.

[0008] Therefore, there is an urgent need for a multi-view recognition method that can enhance intra-class consistency and optimize solution efficiency, taking into account the identification characteristics of Chinese medicinal materials. Summary of the Invention

[0009] To address existing problems, this invention provides a multi-view-based method and system for identifying traditional Chinese medicinal materials. Within the LT-MSC framework, this invention creatively replaces the traditional diversity regularization term with a Frobenius norm constraint. The Frobenius norm constraint directly acts on the subspace representation matrix, forcing the representation coefficients of similar samples in the subspace to converge, thereby significantly reducing intra-class differences. Simultaneously, combined with a sparsity constraint term, it increases the inter-class distance, forming a subspace representation matrix with an ideal block diagonal structure, greatly improving the accuracy of traditional Chinese medicinal material identification. The specific scheme is as follows:

[0010] A method for identifying Chinese medicinal materials based on multiple views includes the following steps:

[0011] S1. Obtain a set of multi-view Chinese medicinal materials data matrices, and for each view, decompose the data matrix into a low-rank part, a subspace representation matrix, and a reconstruction error part;

[0012] S2. Merge the subspace representation matrices corresponding to each view to construct a tensor, and calculate the nuclear norm of the tensor to construct a low-rank tensor-constrained multi-view subspace clustering model;

[0013] S3. Calculate the hypergraph Laplacian matrix of each view, and based on the hypergraph Laplacian matrix and the multi-view subspace clustering model, construct a manifold regularization term to mine the local geometric structure of the data, construct a sparse constraint term to weaken the correlation between data points of different categories, and construct a Frobenius norm constraint term to enhance the correlation consistency between data points of the same category.

[0014] S4. Integrate the multi-view subspace clustering model, the manifold regularization term, the sparse constraint term, and the Frobenius norm constraint term to construct the target function for Chinese medicinal material identification;

[0015] S5. Introducing penalty parameters The augmented Lagrange multiplier method is used to perform alternating direction multiplier iteration on the target function for identifying Chinese medicinal materials. During the iteration process, the original variable, auxiliary variable, and dual variable are updated alternately, and the penalty parameter is updated according to a preset rule. Continue until the preset convergence condition is met to obtain the updated subspace representation matrix;

[0016] S6. Calculate the similarity matrix based on the updated subspace representation matrix, and output the Chinese medicinal material identification result based on the similarity matrix using the spectral clustering algorithm.

[0017] Preferably, in step S1, for a given first... A matrix of Chinese medicinal materials in multiple views Decompose according to the following formula:

[0018] ;

[0019] in, Indicates the first The low-rank portion corresponding to each view; Indicates the first The subspace representation matrix corresponding to each view; Indicates the first The reconstruction error portion corresponding to each view; , This indicates the dimension of the samples in the Chinese medicinal materials data. This indicates the number of samples in the data on Chinese medicinal materials. Indicates the total number of views.

[0020] Preferably, step S2 specifically includes:

[0021] S21. Represent the matrix of each single-view subspace. Construct a order tensor ,

[0022]

[0023] in, This represents a matrix represented by subspaces of different views. After merging, construct order tensor ;

[0024] Will order tensor In the Expanding on each pattern yields a matrix. The expression is:

[0025]

[0026] in, Representation matrix dimensionality;

[0027] S22, Calculation order tensor The nuclear norm number is given by the formula:

[0028]

[0029] in, The nuclear norm of a tensor or matrix. It is a constant greater than 0, and satisfies ;

[0030] S23. Constructing a low-rank tensor-constrained multi-view subspace clustering LT-MSC model based on tensor nuclear norm:

[0031]

[0032] in, This represents the matrix formed by concatenating the reconstruction error matrices of each view. Representing a matrix Norm.

[0033] Preferably, step S3 specifically includes:

[0034] S31. Calculate the hypergraph Laplacian matrix for each view. ;

[0035] S32. Constructing manifold regularization terms ;in, Represents the trace of a matrix;

[0036] S33. Constructing sparse constraint terms ,in, Representing a matrix Norms are used to enhance the block diagonal property of subspace representations of matrices;

[0037] S34. Constructing Frobenius norm constraint terms ,in, The Frobenius norm of a matrix is ​​used to reduce the differences between similar data and enhance intra-class consistency.

[0038] Preferably, the expression for the multi-view-based target function for identifying Chinese medicinal materials constructed in step S4 is:

[0039]

[0040] in, , , and These are the balancing parameters.

[0041] Preferably, the iterative optimization solution using the augmented Lagrange multiplier method in step S5 is as follows:

[0042] S51, Update Variables The updated formula is:

[0043]

[0044] in, This represents a given positive penalty parameter. This represents a penalty parameter that is greater than 0. This is the preset upper limit for the penalty parameter;

[0045] S52. In each iteration, update the original variable, auxiliary variable, and dual variable sequentially. The original variable includes the subspace representation matrix. and reconstruction error matrix The auxiliary variables include auxiliary matrices. , order tensor Vectorization and auxiliary matrix Vectorization The dual variables include Lagrange multipliers. and Lagrange multipliers ; until satisfied ,and This yields the updated subspace representation matrix; where, It represents a very small positive number that satisfies the convergence condition.

[0046] Preferably, step S6 specifically includes:

[0047] S61. Calculate the similarity matrix based on the subspace representation matrix obtained from the final iteration. ;

[0048] S62: The similarity matrix The input spectrum clustering algorithm performs feature decomposition and K-means clustering to output the division results of each Chinese medicinal material sample, which serve as the identification results of the Chinese medicinal materials.

[0049] This invention also discloses a multi-view-based system for identifying Chinese medicinal materials, the system comprising:

[0050] The data acquisition module is used to acquire multi-view data on Chinese medicinal materials.

[0051] The data recognition module is used to execute a multi-view-based Chinese medicinal material recognition method as described above based on the multi-view Chinese medicinal material data to obtain the recognition result of the Chinese medicinal materials.

[0052] The output display module is used to output the identification results of the Chinese medicinal materials.

[0053] This invention discloses a computer-readable storage medium and a computer system, wherein the computer-readable storage medium stores a computer program, and after the computer program is executed, it performs the method described in any of the preceding claims. A computer system includes a processor and a storage medium, wherein the storage medium stores a computer program, and the processor reads from the storage medium and runs the computer program to perform the method described in any of the preceding claims.

[0054] The beneficial effects of this invention are as follows:

[0055] 1. Significantly Improved Recognition Accuracy: By directly optimizing the intra-class compactness of the subspace representation through Frobenius norm constraints, this invention solves the problem of loose clustering caused by existing technologies' focus on view diversity. Experiments show that this invention outperforms existing methods in accuracy (AC), normalized mutual information (NMI) precision, F-score, adjusted Rand index (AR), and purity on multiple Chinese medicinal herb datasets.

[0056] 2. Enhanced model robustness: The Frobenius norm constraint term avoids the problem of redundant noise amplification that may be introduced by the diversity regularization term, making the model more robust to background interference in Chinese medicinal material images.

[0057] 3. Optimize computational efficiency: Compared to calculating the HSIC term of the kernel matrix, the gradient calculation of the Frobenius norm constraint term is simpler, reducing the computational complexity of each iteration and accelerating model convergence. Attached Figure Description

[0058] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0059] Figure 1 This is an overall flowchart of the multi-view-based Chinese medicinal herb identification method of the present invention.

[0060] Figure 2 These are partial images of two categories of traditional Chinese medicinal materials from eight animal databases.

[0061] Figure 3 These are partial images of two categories of Chinese medicinal materials from an eight-category database of plant roots, stems, and barks.

[0062] Figure 4 These are partial images of two categories of Chinese medicinal materials from a database of four categories of plant fruits. Detailed Implementation

[0063] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] The basic theoretical knowledge involved in this invention is introduced as follows:

[0065] 1. K-means clustering method

[0066] The basic idea of ​​the K-means method is to cluster objects based on k points in space, grouping them into categories that are closest to these k points. The values ​​of each cluster center are updated iteratively until the best clustering result is obtained.

[0067] 2. Spectral Clustering Method

[0068] Spectral clustering is an unsupervised learning method based on graph theory. This method divides a weighted undirected graph into two or more optimal subgraphs, making the subgraphs as similar as possible internally and as far apart as possible between them, in order to achieve the purpose of clustering.

[0069] 3. Evaluation Criteria for the Model

[0070] The evaluation criteria employ six methods: accuracy (AC), normalized mutual information (NMI), precision, F-score, adjusted Rand index (AR), and purity.

[0071] 4. The Augmented Lagrange Multiplier (ALM) method is based on the Lagrange multiplier method, incorporating a penalty function to construct a more suitable objective function. This method obtains an approximate optimal solution to the objective function by continuously adjusting the Lagrange multipliers and penalty parameters during the search for the optimal solution.

[0072] Example 1:

[0073] This embodiment provides a method for identifying Chinese medicinal materials based on multiple views. See [link to relevant documentation]. Figure 1The method includes:

[0074] S1: Multi-view Chinese medicinal materials data matrix , , ..., Decompose into matrices sum matrix ,Right now:

[0075]

[0076] in, , and They represent the first The low-rank part corresponding to each view, the subspace representation part, and the reconstruction error part. , This indicates the dimension of the samples in the Chinese medicinal materials data. This indicates the number of data points for Chinese medicinal materials. Indicates the number of views;

[0077] S2: Merge the subspace representation matrices corresponding to each view to construct a tensor, and calculate the nuclear norm of the tensor to construct a low-rank tensor-constrained multi-view subspace clustering model;

[0078] Specifically:

[0079] S21: Based on the subspace representation of each single view Construct a order tensor :

[0080]

[0081] in, This represents a matrix represented by subspaces of different views. After merging, construct order tensor ;

[0082] Will order tensor In the Expanding on each pattern yields a matrix. The expression is:

[0083]

[0084] in, Representation matrix dimensionality;

[0085] S22: Calculation order tensor The nuclear norm number is given by the formula:

[0086]

[0087] in, The nuclear norm of a tensor or matrix. It is a constant greater than 0, and satisfies ;

[0088] S23: Constructing a low-rank tensor-constrained multi-view subspace clustering LT-MSC model:

[0089]

[0090] in, This represents the matrix formed by concatenating the reconstruction error matrices of each view. Representing a matrix Norm.

[0091] S3. Calculate the hypergraph Laplacian matrix of each view, and based on the hypergraph Laplacian matrix and the multi-view subspace clustering model, construct a manifold regularization term to mine the local geometric structure of the data, construct a sparse constraint term to weaken the correlation between data points of different categories, and construct a Frobenius norm constraint term to enhance the correlation consistency between data points of the same category.

[0092] Specifically:

[0093] S31: Calculate the hypergraph Laplacian matrix for each view. ;

[0094] S32: Construct the manifold regularization term ;in, Represents the trace of a matrix;

[0095] S33: Constructing Sparse Constraints ;in, Representing a matrix Norm;

[0096] S34: Constructing Frobenius norm constraint terms ;in, Denotes the Frobenius norm of a matrix;

[0097] S4. Integrate the multi-view subspace clustering model, the manifold regularization term, the sparse constraint term, and the Frobenius norm constraint term to construct the target function for Chinese medicinal material identification;

[0098] Specifically:

[0099] Step 9: Based on the multi-view subspace clustering LT-MSC model and the manifold regularization term sparse constraint terms Frobenius norm constraint terms Construct a multi-view-based target function for identifying Chinese medicinal herbs. The expression of the target function is as follows:

[0100]

[0101] in, , , and These are the balancing parameters.

[0102] S5. Introducing penalty parameters The augmented Lagrange multiplier method is used to perform alternating direction multiplier iteration on the target function for identifying Chinese medicinal materials. During the iteration process, the original variable, auxiliary variable, and dual variable are updated alternately, and the penalty parameter is updated according to a preset rule. The process continues until the preset convergence condition is met, resulting in the updated subspace representation matrix.

[0103] Specifically:

[0104] S51: Update Variables The updated formula is:

[0105]

[0106] in, This represents a given positive penalty parameter. This represents a penalty parameter that is greater than 0.

[0107] S52: In each iteration, update the original variable, auxiliary variable, and dual variable sequentially. The original variable includes the subspace representation matrix. and reconstruction error matrix The auxiliary variables include auxiliary matrices. , order tensor Vectorization and auxiliary matrix Vectorization The dual variables include Lagrange multipliers. and Lagrange multipliers ; until satisfied ,and This yields the updated subspace representation matrix; where, It represents a very small positive number that satisfies the convergence condition.

[0108] Wherein, the updated subspace representation matrix The update process includes:

[0109] According to the formula The multi-view-based Chinese medicinal herb recognition model is further represented in the following form:

[0110]

[0111] in, , , , ;

[0112] Optimizing the above representation of the present invention can be accomplished using the Augmented Lagrange Multiplier Method (ALM). When solving the problem using the ALM method, the objective function of the present invention needs to be separable; therefore, auxiliary variables are introduced. to replace Then we have:

[0113]

[0114] in, and They are respectively the order tensor and the auxiliary variables Vectorization, It is a permutation matrix used for calibration. and the auxiliary variables The corresponding elements between them;

[0115] By using the augmented Lagrange multiplier method, the multi-view-based Chinese medicinal herb identification model is further represented as follows:

[0116]

[0117] in, and It is a Lagrange multiplier. For penalty parameters, This represents the augmented Lagrangian function of the model in this embodiment;

[0118] When the reconstruction error of each view and auxiliary variables When fixed, the subspace represents a matrix. The update formula is:

[0119]

[0120] in, Indicates selection and the first Each view corresponds to an element, and then these elements are reconstructed into a matrix. Represents a matrix consisting entirely of 1s;

[0121] The updated subspace representation matrix is ​​then obtained. as follows:

[0122]

[0123] in, , , Represents the identity matrix.

[0124] S6. Calculate the similarity matrix based on the updated subspace representation matrix, and output the Chinese medicinal material identification result based on the similarity matrix using the spectral clustering algorithm.

[0125] Specifically:

[0126] S61: Calculate the similarity matrix ;

[0127] S62: The similarity matrix The input spectrum clustering algorithm performs feature decomposition and K-means clustering to output the division results of each Chinese medicinal material sample, which serve as the identification results of the Chinese medicinal materials.

[0128] Example 2:

[0129] This embodiment provides a method for identifying Chinese medicinal materials based on multiple views. See [link to relevant documentation]. Figure 1 The method includes:

[0130] (1) Given a multi-view matrix of Chinese medicinal materials data , , ..., ,here ,in This indicates the dimension of the samples in the Chinese medicinal materials data. This indicates the number of data points for Chinese medicinal materials. Indicates the number of views;

[0131] (2) The first in the multi-view medicinal herb data The matrix corresponding to each view Decompose into matrices sum matrix ,Right now:

[0132]

[0133] in, , and They represent the first The low-rank part corresponding to each view, the subspace representation part, and the reconstruction error part (or sparse noise).

[0134] (3) Based on the subspace representation of each single view Construct a order tensor :

[0135]

[0136] in, This represents a matrix represented by subspaces of different views. After merging, construct order tensor .

[0137] Will order tensor In the Expanding on each pattern yields a matrix. The expression is:

[0138]

[0139] in, Representation matrix The dimension of.

[0140] (4) According to the formula calculate order tensor nuclear norm number;

[0141] in, The nuclear norm of a tensor or matrix. It is a constant greater than 0, and satisfies ;

[0142] (5) Construct a low-rank tensor-constrained multi-view subspace clustering model (LT-MSC):

[0143]

[0144] in, The matrix formed by concatenating the reconstruction error matrices of each view together. Representing a matrix Norm;

[0145] (6) Calculate the Laplacian matrix of the hypergraph for a single view. :

[0146] A hypergraph A set of vertices The set of superedges and the set of weights of each hyperedge Composition. Each hyperedge They were all given a positive weight. Weight matrix The elements in and each superedge related, and satisfy , This represents the operation of finding the union of sets;

[0147] SuperMap correlation matrix The elements of the incidence matrix, used to represent the relationships between vertices and hyperedges, are defined as follows:

[0148]

[0149] in, This represents the degree of each hyperedge.

[0150] SuperMap Laplace matrix for:

[0151]

[0152] in, It is a diagonal matrix whose diagonal elements are the degree of each vertex. A diagonal matrix representing the degree of a hyperedge.

[0153] (7) Construct the manifold regularization term ;in, Represents the trace of a matrix.

[0154] In this embodiment, manifold regularization refers to hypergraph regularization, and the manifold regularization term is: .

[0155] (8) Construct sparse constraint terms ;in, Representing a matrix Norm.

[0156] (9) Construct Frobenius norm constraint terms ;in, Denotes the Frobenius norm of a matrix;

[0157] (10) Construct a target function for identifying Chinese medicinal materials based on multiple views;

[0158] Based on the LT-MSC framework, the aforementioned manifold regularization term and sparse constraint term are introduced, and the Frobenius norm constraint term is also incorporated. The formula for the multi-view-based Chinese medicinal herb identification objective function in this embodiment is expressed as follows:

[0159]

[0160] in, , , and For balancing parameters, Indicates the first A matrix representing the subspace of each view. It is based on The Laplacian matrix of the hypergraph of the concrete view on the graph, with the manifold regularization term being: The sparse constraint term is The Frobenius norm constraint term is .

[0161] (11) Update variables ;

[0162]

[0163] in, This represents a given positive penalty parameter. This represents a penalty parameter that is greater than 0.

[0164] (12) Update the subspace representation matrix ;

[0165] According to the formula The model of the multi-view-based Chinese medicinal herb identification method in this embodiment can be further represented in the following form:

[0166]

[0167] in, , , , Indicates the balance parameters;

[0168] Optimizing the above representation of the present invention can be accomplished using the Augmented Lagrange Multiplier (ALM) method. When solving the problem using the ALM method in this embodiment of the invention, it is necessary to make the objective function of this embodiment separable. Therefore, auxiliary variables are introduced into the model of this embodiment. to replace Then we have:

[0169]

[0170] in, and They are tensors sum matrix Vectorization, It is a permutation matrix used for calibration. and The corresponding elements between them;

[0171] By using the Augmented Lagrange Multiplier (ALM) method, the objective function of this embodiment can be further expressed as follows:

[0172]

[0173] in, and It is a Lagrange multiplier. For penalty parameters, This represents the augmented Lagrangian function of the model in this invention;

[0174] :when and When fixed, This can be obtained by solving the following subproblem:

[0175]

[0176] in, Indicates selection and the first Each view corresponds to an element, and then these elements are reconstructed into a matrix. Represents a matrix consisting entirely of 1s;

[0177] It can be obtained as follows:

[0178]

[0179] in, , , Represents the identity matrix.

[0180] (13) Update the reconstruction error matrix ;

[0181] :when and When fixed, we can obtain as follows:

[0182]

[0183] in, It is a matrix The solution to the above equation can be obtained by connecting the columns vertically together to form a matrix, as shown below:

[0184]

[0185] in, Representation matrix The List.

[0186] (14) Update auxiliary variables ;

[0187] Auxiliary variables The iterative formula is as follows:

[0188]

[0189] in, , It is based on the nth pattern of the expansion Reshape into the corresponding matrix. This represents the spectral soft thresholding operation of a matrix, where the spectrum of a matrix refers to the set of all eigenvalues ​​of the matrix.

[0190] (15) Update the Lagrange multipliers ;

[0191] Lagrange multipliers The update rules are as follows:

[0192]

[0193] (16) Update tensor Vectorization;

[0194] Regarding the above updates It can be achieved by direct replacement Update the corresponding elements ,Right now .

[0195] (17) Update auxiliary variables Vectorization ;

[0196] : Updates and Similar to the update, you can get as follows:

[0197]

[0198] (18) Update the Lagrange multipliers ;

[0199] Lagrange multipliers Updates and similar, The update rules are as follows:

[0200]

[0201] Until satisfied ,and ,in, It represents a very small positive number that satisfies the convergence condition;

[0202] (19) Calculate the similarity matrix ;

[0203] (20) Using the spectral clustering method, based on the above similarity matrix Identification of Chinese medicinal herbs;

[0204] (21) Output the identification results of Chinese medicinal materials.

[0205] The performance of this invention can be further illustrated by the following simulation experiments.

[0206] 1. Simulation conditions and parameters

[0207] To verify the effectiveness of the multi-view-based Chinese medicinal herb identification method of the present invention, simulation experiments were conducted on three multi-view Chinese medicinal herb datasets.

[0208] like Figure 2 These are partial images of two categories of traditional Chinese medicinal materials from eight animal databases.

[0209] like Figure 3 These are partial images of two categories of Chinese medicinal materials from an eight-category database of plant roots, stems, and barks.

[0210] like Figure 4 These are partial images of two categories of Chinese medicinal materials from a database of four categories of plant fruits.

[0211] This dataset contains 120 images of 8 different animal-derived medicinal materials, including centipedes, seahorses, toads, spotted cheetahs, leeches, stink bugs, sea dragons, and silkworms. Each animal has 15 images.

[0212] This dataset contains 120 images of 8 types of plant root and stem bark, including agarwood, mimosa bark, astragalus, sappanwood, trichosanthes root, euphorbia bark, asarum, and turmeric. Each type has 15 images.

[0213] This dataset contains 120 images of 4 types of plant fruits, including white hyacinth bean, wolfberry, white cardamom, and cowpea. Each type has 30 images.

[0214] When performing simulation experiments, the parameters of the method of this invention... The value range is {0.01, 0.1, 1, 10, 100}, parameter The value range is {0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9}, parameter The value range is [0.00001, 1], parameter The value range is [0.00001, 1], and the nearest neighbor parameter is set to 5.

[0215] All experiments were repeated 20 times, and the final experimental result was the average of the 20 experimental results.

[0216] The method of the present invention compares LT-MSC (Zhang C, Fu H, Liu S, et al. Low-rank tensor constrained multiview subspace clustering[C]. Proceedings of the IEEE international conference on computer vision. 2015: 1582-1590.), LMSC (Zhang C, Hu Q, Fu H, et al. Latent multi-view subspace clustering[C]. Proceedings of the IEEE conference on computer vision and pattern recognition. 2017: 4279-4287.), GMC (Wang H, Yang Y, Liu B. GMC: Graph-based multi-view clustering[J]. IEEE Transactions on Knowledge and Data Engineering, 2019, 32(6): 1116-1129.), MCLES (Chen M S, Huang L, Wang C D, et al. Multi-view clustering in latent embedding space[C]. Proceedings of the AAAI conference on artificial intelligence. 2020, 34(04): 3513-3520.), RSE (Li Z, Hu Z, Nie F, et al. Multi-view clustering based on generalized low rank approximation[J]. Neurocomputing, 2022, 471: 251-259.), MCDCF (Chang S, Hu J, Li T, et al. Multi-view clustering via deep concept factorization[J]. Knowledge-Based Systems, 2021, 217: 106807.), MCDHSIC (Liu G, Ge H, Su S, et al.Seven multi-view learning methods, including Multi-view clustering via dual-norm and hsic[J]. Multimedia Tools and Applications, 2024, 83(12):36399-36418.

[0217] 2. Simulation Content and Result Analysis

[0218] Clustering experiments conducted on a multi-view dataset of Chinese medicinal materials.

[0219] The clustering results of each method are given in the table below, where AC represents the clustering accuracy, NMI represents the normalized information, F-score represents the F-value, AR represents the adjusted Land coefficient, Precision represents the clustering precision, and Purity represents the purity. The higher the value of the above six evaluation indicators, the better the clustering effect of the method.

[0220] Experiment 1: A comparative experiment was conducted on a dataset of 8 types of animal medicinal materials. The experimental results of different methods on the dataset of 8 types of animal medicinal materials are shown in Table 1.

[0221] Table 1. Recognition results of each method on the dataset of 8 types of animal-derived medicinal materials (%)

[0222]

[0223] As can be seen from Table 1, the present invention achieves the highest scores across all six evaluation indicators compared to other methods, demonstrating that the method of the present invention exhibits the best recognition performance.

[0224] Experiment 2: A comparative experiment was conducted on 8 types of plant root, stem, and bark datasets. The experimental results of different methods on the 8 types of plant root, stem, and bark datasets are shown in Table 2.

[0225] Table 2. Recognition results of each method on 8 types of plant root, stem, and bark datasets (%)

[0226]

[0227] The experimental results in Table 2 demonstrate that the method of the present invention has the highest recognition accuracy compared to other methods.

[0228] Experiment 3: Comparative experiments were conducted on four types of plant fruit datasets. The experimental results of different methods on the four types of plant fruit datasets are shown in Table 3.

[0229] Table 3. Recognition results (%) of each method on four types of plant fruit datasets.

[0230]

[0231] As can be seen from Table 3, compared with other multi-view clustering methods, the method of the present invention still shows the best recognition performance on the six evaluation indicators.

[0232] This invention superimposes the subspace representation matrices of different views of Chinese medicinal materials into a tensor to obtain high-order correlations between multi-view data. Low-rank constraints are applied to the tensor to obtain a correlation objective function. Based on the objective function, Frobenius norm constraints are used to enhance the connections between similar data points, resulting in a subspace representation matrix with intra-class consistency. Furthermore, sparsity constraints are applied to the subspace representation matrix to obtain a subspace representation matrix with inter-class sparsity. In addition, hypergraph regularization is introduced to explore the geometric structure in the Chinese medicinal material data to mine manifold information. The update rule of the subspace representation matrix is ​​obtained by optimizing the designed objective function, leading to a similarity matrix. A spectral clustering algorithm is applied to obtain the final clustering result, which is then used as the identification result for Chinese medicinal materials. This invention obtains a subspace representation matrix with intra-class consistency and inter-class sparsity by applying Frobenius norm and sparsity constraints to the subspace representation matrix, thereby obtaining a more complete data representation. This can improve the identification performance of the constructed method for Chinese medicinal materials to a certain extent. Compared with existing multi-view recognition schemes, the method of the present invention can achieve the best clustering effect and recognition performance in the scenario of Chinese medicinal material recognition, thereby effectively improving the accuracy of Chinese medicinal material recognition.

[0233] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or a combination of both. To clearly illustrate this interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps are described above in a generalized manner in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. Those skilled in the art may implement the described functionality in different ways for each specific application, but such implementation decisions should not be construed as departing from the scope of the invention.

[0234] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-view-based traditional Chinese medicinal material recognition method, characterized in that, Includes the following steps: S1. Obtain a set of multi-view Chinese medicinal materials data matrices, and for each view, decompose the data matrix into a low-rank part, a subspace representation matrix, and a reconstruction error part; S2. Merge the subspace representation matrices corresponding to each view to construct a tensor, and calculate the nuclear norm of the tensor to construct a low-rank tensor-constrained multi-view subspace clustering model; S3. Calculate the hypergraph Laplacian matrix of each view, and based on the hypergraph Laplacian matrix and the multi-view subspace clustering model, construct a manifold regularization term to mine the local geometric structure of the data, construct a sparse constraint term to weaken the correlation between data points of different categories, and construct a Frobenius norm constraint term to enhance the correlation consistency between data points of the same category. S4. Integrate the multi-view subspace clustering model, the manifold regularization term, the sparse constraint term, and the Frobenius norm constraint term to construct the target function for Chinese medicinal material identification; S5, introducing a penalty parameter , using the augmented Lagrange multiplier method to perform alternating direction multiplier iteration on the traditional Chinese medicinal material identification target function, alternately updating the original variable, the auxiliary variable and the dual variable in the iteration process, and updating the penalty parameter according to a preset rule , until a preset convergence condition is met, to obtain an updated subspace representation matrix; S6. Calculate the similarity matrix based on the updated subspace representation matrix, and output the Chinese medicinal material identification result based on the similarity matrix using the spectral clustering algorithm.

2. The method according to claim 1, characterized in that, In step S1, the Chinese medicinal material data matrix for a given first view is decomposed according to the following formula: X = X1 + X2 ； wherein, represents a low-rank part corresponding to the th view; represents a low-rank part corresponding to the th view; represents a low-rank part corresponding to the th view; , represents the dimension of the sample in the traditional Chinese medicinal material data, represents the number of samples of the traditional Chinese medicinal material data, represents the total number of views.

3. The method of claim 2, wherein, Step S2 specifically includes: S21, the single view subspace representation matrix constructing one rank tensor , wherein represents the construction of a subspace representation matrix tensor of rank ; will be described below. rank tensor In the first mode, we get a matrix with the expression wherein denotes the dimension of the matrix denotes the dimension of the matrix S22, compute rank tensor norm of the kernel, which is given by wherein, represents the nuclear norm of a tensor or matrix, is a constant greater than 0 and satisfying ; S23. Constructing a low-rank tensor-constrained multi-view subspace clustering LT-MSC model based on tensor nuclear norm: wherein, denotes the matrix formed by concatenating the reconstruction error matrices of the individual views, denotes the Frobenius norm of the matrix. denotes the Frobenius norm of the matrix.

4. The method of claim 3, wherein, Step S3 specifically includes: S31, calculating the hypergraph Laplacian matrix of each view ; S32, constructing a manifold regularization term ; wherein, denotes the trace of a matrix; S33. Constructing sparse constraint terms ,in, Representing a matrix Norms are used to enhance the block diagonal property of subspace representations of matrices; S34. Constructing Frobenius norm constraint terms ,in, The Frobenius norm of a matrix is ​​used to reduce the differences between similar data and enhance intra-class consistency.

5. The method according to claim 4, characterized in that, The expression for the target function for Chinese medicinal herb identification constructed in step S4 is as follows: in, , , and These are the balancing parameters.

6. The method according to claim 5, characterized in that, The process of iterative optimization using the augmented Lagrange multiplier method in step S5 is as follows: S51, Update Variables The updated formula is: in, This represents a given positive penalty parameter. This represents a penalty parameter that is greater than 0. This is the preset upper limit for the penalty parameter; S52. In each iteration, update the original variable, auxiliary variable, and dual variable sequentially. The original variable includes the subspace representation matrix. and reconstruction error matrix The auxiliary variables include auxiliary matrices. , order tensor Vectorization and auxiliary matrix Vectorization The dual variables include Lagrange multipliers. and Lagrange multipliers ; until satisfied ,and This yields the updated subspace representation matrix; where, It represents a very small positive number that satisfies the convergence condition.

7. The method according to claim 6, characterized in that, Step S6 specifically includes: S61. Calculate the similarity matrix based on the subspace representation matrix obtained from the final iteration. ; S62: The similarity matrix The input spectrum clustering algorithm performs feature decomposition and K-means clustering to output the division results of each Chinese medicinal material sample, which serve as the identification results of the Chinese medicinal materials.

8. A multi-view-based system for identifying Chinese medicinal herbs, characterized in that, The system includes: The data acquisition module is used to acquire multi-view data on Chinese medicinal materials. The data recognition module is used to execute a multi-view-based Chinese medicinal material recognition method as described in any one of claims 1-7 based on the multi-view Chinese medicinal material data, so as to obtain the recognition result of the Chinese medicinal material; The output display module is used to output the identification results of the Chinese medicinal materials.

9. A computer-readable storage medium, characterized in that: The medium contains a computer program, which, when executed, performs the method as described in any one of claims 1 to 7.

10. A computer system, characterized in that: It includes a processor and a storage medium, on which a computer program is stored, and the processor reads from the storage medium and runs the computer program to perform the method as described in any one of claims 1 to 7.

Images