A modeling-free mixture component intelligent identification method and system

Abstract

The application provides a modeling-free mixture component intelligent identification method and system, which comprises collecting a plurality of mixture spectra of a to-be-tested sample, and pre-processing the plurality of mixture spectra; arranging the pre-processed plurality of mixture spectra into a matrix, and separating and outputting a plurality of source spectra; constructing a standard spectrum library, and correcting the standard spectra in the standard library and the separated spectra; detecting local extreme values of each source spectrum to obtain a peak position set; matching the peak position set with each pure component standard spectrum in the standard spectrum library, and judging an intersection number; if the intersection number is greater than a preset value, it is determined that the component exists, and an existing component list and a corresponding matching degree or similarity score are output. The application realizes modeling-free output of existence conclusion, solves the pain points of time-consuming verification of traditional methods depending on modeling, and realizes non-destructive, rapid and accurate qualitative verification of specific components in a textile mixture.

Description

Technical Field

[0001] This invention relates to the field of material composition analysis technology, and in particular to a model-free intelligent identification method and system for mixture components. Background Technology

[0002] The spectra of mixtures (such as textile fibers, agricultural products, pharmaceuticals, petrochemical products, etc.) are often formed by the linear superposition of the spectra of several pure components.

[0003] Traditional identification processes require: pre-collecting a large number of standard spectra of pure components and establishing a multivariate calibration model (PLS, PCR, SVM, etc.); acquiring spectra of unknown mixtures; and inputting the spectra into the model for prediction. Traditional identification processes have the following limitations: long modeling cycles, requiring a large number of representative standard samples; the model is sensitive to instrument drift, baseline changes, and temperature variations, requiring frequent maintenance; it is only applicable to already modeled components, and must be remodeled for "new components"; and quantitative models may introduce unnecessary errors for qualitative tasks.

[0004] In recent years, blind source separation techniques (such as Independent Component Analysis (ICA), Non-negative Matrix Factorization (NMF), and Sparse Component Analysis (SCA) have been widely used in speech, image, and biomedical signal processing. Their advantages include: requiring only the mixed signal without standard samples; not relying on prior concentration / content information; and being able to separate the mixed spectrum into several independent / sparse / non-negative "source spectra" even under unknown mixing ratios.

[0005] In summary, existing technologies require the prior establishment of calibration models, which is complex; they are sensitive to changes in instruments and the environment, resulting in high maintenance costs; they cannot respond in a timely manner to substances outside the component library; and quantitative models are used for qualitative tasks, leading to over-design. Summary of the Invention

[0006] Therefore, the purpose of this invention is to provide a model-free intelligent identification method and system for mixture components, in order to overcome the shortcomings of the prior art.

[0007] In a first aspect, the present invention provides a model-free intelligent identification method for mixture components, the method comprising:

[0008] Several mixture spectra of the sample to be tested are collected, and the mixture spectra are preprocessed.

[0009] The preprocessed spectra of the mixture are arranged into a matrix, and several source spectra are separated and output.

[0010] Construct a standard spectral library and calibrate the standard spectra in the library and the separated spectra;

[0011] Detect the local extrema of each of the source spectra to obtain a set of peak positions;

[0012] The peak position set is matched with the standard peak positions of each pure component in the standard spectral library, and the number of intersections is determined.

[0013] If the number of intersections is greater than a preset value, the component is determined to exist, and a list of existing components and the corresponding matching degree or similarity score are output.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: by arranging several pre-processed mixture spectra into a matrix and outputting the source spectrum, and by detecting the local extrema, the peak set can be obtained. By judging the number of intersections between the peak set and each component in the standard spectrum, the existence of the component can be determined. Thus, the existence conclusion is output without modeling, which solves the pain point of traditional methods relying on modeling and time-consuming verification, and realizes non-destructive, rapid and accurate qualitative verification of specific components in textile mixtures.

[0015] Furthermore, the step of acquiring several mixture spectra of the sample to be tested and preprocessing the several mixture spectra includes:

[0016] The original spectral vector of the sample to be tested is obtained based on the spectrometer to obtain several spectra of the mixture;

[0017] The baseline correction, normalization, smoothing, and noise reduction processes were sequentially applied to several of the mixture spectra.

[0018] Furthermore, the step of arranging the preprocessed mixture spectra into a matrix and separating and outputting several source spectra includes:

[0019] Arrange several of the mixture spectra that match the number of separated components into a matrix;

[0020] Based on the spectral characteristics, independent vector analysis, nonnegative matrix factorization, constrained nonnegative spectroscopy, and sparse component analysis are employed. The matrix is ​​then analyzed using spectral sparsity or based on time-frequency masking to output several source spectra.

[0021] Furthermore, the step of correcting the standard spectra in the standard library and the separated spectra includes:

[0022] The standard spectra in the standard library and the separated spectra are baseline corrected using the downsampling baseline correction method.

[0023] Furthermore, after the step of detecting the local extrema of each of the source spectra to obtain the peak set, the method further includes:

[0024] Calculate the cosine similarity, Euclidean distance, or correlation coefficient between each pure component in the standard spectral library and the peak set, and determine whether the component exists based on the cosine similarity, Euclidean distance, or correlation coefficient.

[0025] Secondly, the present invention also provides a model-free intelligent identification system for mixture components, the system comprising:

[0026] The acquisition and preprocessing module is used to acquire several mixture spectra of the sample to be tested and to preprocess the several mixture spectra.

[0027] The output module is used to arrange the preprocessed spectra of the mixture into a matrix and separate and output several source spectra.

[0028] A calibration module is constructed to build a standard spectral library and to calibrate the standard spectra in the standard library and the separated spectra.

[0029] The detection module is used to detect the local extrema of each of the source spectra to obtain the peak set;

[0030] The matching and judgment module is used to match the peak position set with the standard peak positions of each pure component in the standard spectral library and determine the number of intersections.

[0031] The determination output module is used to determine the existence of a component if the number of intersections is greater than a preset value, and output a list of existing components and the corresponding matching degree or similarity score.

[0032] Furthermore, the acquisition preprocessing module includes:

[0033] The acquisition unit is used to acquire the original spectral vector of the sample to be tested based on the spectrometer, so as to obtain several spectra of the mixture;

[0034] The processing unit is used to sequentially perform baseline correction, normalization, smoothing, and noise reduction on several of the mixture spectra.

[0035] Furthermore, the arrangement output module includes:

[0036] Arrangement unit, used to arrange several spectra of the mixture, matching the number of separated components, into a matrix;

[0037] The analysis output unit is used to perform independent vector analysis, nonnegative matrix decomposition, constrained nonnegative spectroscopy, and sparse component analysis based on spectral characteristics, and to analyze the matrix using spectral sparsity or based on time-frequency masking to output several source spectra.

[0038] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-described model-free intelligent identification method for mixture components.

[0039] Fourthly, the present invention also provides a storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described model-free intelligent identification method for mixture components. Attached Figure Description

[0040] Figure 1 This is a flowchart of the model-free intelligent identification method for mixture components in the first embodiment of the present invention;

[0041] Figure 2 This is a separation diagram of the IVA full-spectrum retrieval method in the first embodiment of the present invention. Figure 2 (a) in the image represents the original cotton spectrum. Figure 2 (b) in the image shows the separated cotton spectrum. Figure 2 (c) in the image represents the original pure polyester spectrum. Figure 2 (d) in the figure represents the separation spectrum;

[0042] Figure 3 This is a separation diagram of the NMF full-spectrum retrieval method in the first embodiment of the present invention. Figure 3 (a) in the image is the reference pure cotton spectrum. Figure 3 (b) in the image shows the separated cotton component spectrum. Figure 3 (c) in the image is the reference pure polyester spectrum. Figure 3 (d) in the figure represents the separation spectrum;

[0043] Figure 4 This is a matching diagram of the NMF algorithm peak position matching method in the first embodiment of the present invention. Figure 4 (a) in the image represents the original spectrum. Figure 4 (b) in the table shows the peak position matching details;

[0044] Figure 5 This is a separation diagram of the SCA full-spectrum retrieval method in the first embodiment of the present invention. Figure 5 (a) in the image represents the original pure cotton spectrum. Figure 5 (b) in the image represents the separated cotton spectrum. Figure 5 (c) in the image represents the original pure polyester spectrum. Figure 5 (d) in the figure represents the separation spectrum;

[0045] Figure 6 This is a separation diagram of the TFMBSS full-spectrum retrieval method in the first embodiment of the present invention. Figure 6 (a) in the image represents the original pure cotton spectrum. Figure 6 (b) in the image shows the separated cotton spectrum. Figure 6(c) in the image represents the original pure polyester spectrum. Figure 6 (d) in the figure represents the separation spectrum;

[0046] Figure 7 Spectral diagram of cotton-polyester textiles;

[0047] Figure 8 This is a structural block diagram of the model-free intelligent identification system for mixture components according to the second embodiment of the present invention;

[0048] Figure 9 This is a structural block diagram of the electronic device in the third embodiment of the present invention.

[0049] Explanation of key component symbols:

[0050] 10. Acquisition and Preprocessing Module; 20. Arrangement and Output Module; 30. Construction and Correction Module; 40. Detection Module; 50. Matching and Judgment Module; 60. Judgment and Output Module;

[0051] 70. Bus; 71. Processor; 72. Memory; 73. Communication interface.

[0052] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0053] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0054] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0056] Example 1

[0057] Please see Figure 1The figure shows a model-free intelligent identification method for mixture components according to the first embodiment of the present invention, the method comprising steps S1 to S6:

[0058] S1, Collect several mixture spectra of the sample to be tested, and preprocess the several mixture spectra;

[0059] Specifically, step S1 includes steps S11 to S12:

[0060] S11, Based on the spectrometer, obtain the original spectral vector of the sample to be tested to obtain several spectra of the mixture;

[0061] S12, perform baseline correction, normalization, smoothing and noise reduction on several of the mixture spectra in sequence;

[0062] It should be noted that a portable spectrometer is used to acquire the original spectral vector of the sample under test in near-infrared, mid-infrared or Raman spectra. In this embodiment, the original spectral vector is represented as x(λ). Conventional preprocessing such as baseline correction, normalization, smoothing and denoising is performed to reduce instrument and environmental interference.

[0063] S2, arrange the pre-processed spectra of the mixture into a matrix, and separate and output several source spectra;

[0064] Specifically, step S2 includes steps S21 to S22:

[0065] S21, arrange several of the mixture spectra that match the number of separated components into a matrix;

[0066] S22, based on the spectral characteristics, independent vector analysis, non-negative matrix decomposition, constrained non-negative spectroscopy, and sparse component analysis are used to analyze the matrix using spectral sparsity or based on time-frequency masking to output several source spectra.

[0067] It is understandable that, in this embodiment, the mixing matrix is ​​constructed by arranging m mixture spectra that match the number of separated components into a mixing spectral matrix. ,in Given the number of wavelength points, and based on the spectral characteristics, independent vector analysis (IVA), nonnegative matrix factorization (NMF), constrained nonnegative spectroscopy (SCA), and sparse component analysis (TFMBSS) can be used for analysis to output the source spectrum and source spectral matrix. k is the estimated number of components (which can be set by information criteria such as AIC / BIC or experience).

[0068] In practical implementation, Independent Vector Analysis (IVA) is used:

[0069] Robust IVA optimization with multiple random initializations is implemented. To improve the probability of global convergence, a maximum number of run cycles is set; max_runs=10 (expandable to 30, 50, or 100). The execution flow for each cycle is as follows:

[0070] Random orthogonal initialization: Generate a random matrix wnit(:,:,k)∈ for each segment k=1…K. ^{N×N}, and immediately orthogonalize wnit(:,:,k)=orth(wnit(:,:,k)) to ensure that the initial estimate is non-singular and has a good condition number;

[0071] IVA core decomposition: Calling the function [W1,~]=iva_a_ggd_decp_FS_P(X,'initW',wnit), where: iva_a_ggd_decp_FS_P is a fast and stable implementation of independent vector analysis based on the Generalized Gaussian Distribution (GGD); the input parameter Z is the aforementioned three-dimensional observation tensor; the keyword initW specifies the initial separation matrix for this round; the output W1∈ ^{N×N×K} is the set of separation matrices corresponding to each data segment; when the algorithm fails to converge or numerical anomalies occur, the round is skipped using try-catch to avoid abnormal results polluting the statistics;

[0072] Signal reconstruction: Each segment of local data is independently separated using the latest W1, and then spliced ​​together in the original order to obtain a global estimate; Where, estimated_sources: the estimated source signal matrix, and the final reconstructed global source signal result;

[0073] k: Index of the data segment (from 1 to K, corresponding to the K segments mentioned earlier);

[0074] T_half: The number of sample points contained in each data segment (half time / sample length);

[0075] W1: The set of separation matrices is a three-dimensional matrix (with dimensions N×N×K), where W1(:,:,k) represents the separation matrix corresponding to the k-th data segment;

[0076] Z: The original mixed signal matrix, which is also a three-dimensional matrix (with dimensions N×T_half×K), where Z(:,:,k) represents the k-th segment of the mixed signal data;

[0077] In matrix indexing, : represents "all elements". For example, W1(:,:,k) means taking all rows and columns of the k-th segment in the three-dimensional matrix W1.

[0078] This represents matrix multiplication.

[0079] Similarity assessment: The estimated source vectors are normalized to the [0,1] interval, and Pearson correlation coefficients are calculated with known standard spectra of pure cotton and pure polyester. The correlation coefficient of pure cotton is corr_cotton = corr(pure_cotton_norm,source1), where corr refers to the Pearson correlation coefficient. The correlation coefficient of pure polyester is corr_polyester = corr(pure_polyester_norm,source2). The total similarity is total_sim = corr_cotton + corr_polyester. The cached variables are updated only when total_sim is higher than the historical best value. The best similarity is best_similarity, the best weight is best_W1, and the best source vector is best_sources. Corr_cotton is the correlation coefficient of pure cotton (a combination of cotton correlations), which refers to the Pearson correlation coefficient between the normalized standard spectrum of pure cotton and the first estimated source vector.

[0080] corr: Pearson correlation coefficient (abbreviation for correlation coefficient), used to measure the degree of linear correlation between two variables, with a value range of [-1, 1];

[0081] pure_cotton_norm: Normalized pure cotton standard spectrum (combination of pure cotton normalized), refers to pure cotton standard spectrum data that has been normalized in the [0,1] interval;

[0082] source1: The first estimated source vector, referring to the first source signal vector obtained through the algorithm; source2: The second estimated source vector, referring to the first source signal vector obtained through the algorithm.

[0083] IVA full-spectrum retrieval method separation map, such as Figure 2 As shown, Figure 2 (a) in the image represents the original cotton spectrum. Figure 2 (b) in the image shows the separated cotton spectrum. Figure 2 (c) in the image represents the original pure polyester spectrum. Figure 2 In the figure, (d) represents the separation spectrum, and the expandable parameters are shown in Table 1:

[0084] Table 1

[0085]

[0086] NMF (Nonnegative Matrix Factorization):

[0087] Data Preprocessing: Mean Centering: Perform mean-reduction operation on each row of the pure cotton reference spectrum (cotton_ref), the pure polyester reference spectrum (poly_ref), and the mixed spectrum matrix X: cotton_ref = cotton_ref - mean(cotton_ref), where mean is a function for calculating the average; poly_ref = poly_ref - mean(poly_ref); X = mix_data - mean(mix_data, 2), where mix_data represents mixed data, which can eliminate baseline drift and improve the stability of NMF decomposition.

[0088] Nonnegativity transformation: Since NMF requires the input data to be nonnegative, a translation operation is performed on the centered mixed spectral matrix X: nonneg mixed spectral matrix X_nonneg, X_nonneg=X-min(X(:))+1e-6; where 1e-6 is a small positive offset to avoid zero values ​​causing numerical instability;

[0089] Nonnegative matrix factorization (NMF) execution: The core parameters are set to 2, with the number of components num_components=2%; the number of separated components (corresponding to cotton and polyester components); the repetition coefficient num_replicates=15%; and the number of NMF repetitions (which can be extended to 30~50 times). Multiple repetitions can effectively reduce the risk of local optima caused by random initial values.

[0090] Decomposition Algorithm: Call a custom or standard NMF function: [source, A] = NMF(X_nonneg, num_components, num_replicates); where source is a 2×L matrix, each row corresponding to a separated pure component spectrum; A is a 2×2 mixing coefficient matrix, representing the relative weight of each component in the original observation; num_components represents the number of components, indicating the number of pure components that the NMF algorithm needs to separate from the mixed data. In this embodiment, it is set to 2, corresponding to the two components to be separated: pure cotton and pure polyester; num_replicates represents the number of repetitions, indicating the number of times the NMF algorithm runs independently. Since the results of NMF may get stuck in local optima due to the influence of initial values, multiple repetitions (15 times in this embodiment, which can be extended to 30~50 times) can reduce this risk and improve the reliability of the results; Output verification: If the number of rows in source is less than 2, an error "insufficient separated components" is reported to ensure the effectiveness of the algorithm; Result matching and post-processing: Construct a 2×2 correlation coefficient matrix corr_matr ix, where the element is the absolute value of the Pearson correlation coefficient between the separated spectrum and the reference spectrum: corr_matrix(i,j)=|corr(source(i,:)',ref(j)')|, where i and j represent the indices of the separated spectrum and the reference spectrum, respectively. source(i,:)' represents all elements in the i-th row of the separated spectrum matrix source, i.e., the i-th separated pure component spectrum, which is a row vector. ref(j)' represents all elements in the j-th row of the reference spectrum matrix ref, i.e., the j-th index... The quasi-reference spectrum, such as the standard spectrum of pure cotton or pure polyester, is a row vector. The ' in the upper right corner indicates a transpose operation, converting the row vector into a column vector. The optimal correspondence is automatically matched by maximizing the sum of the diagonal elements (corr_matrix(1,1)+corr_matrix(2,2) or corr_matrix(1,2)+corr_matrix(2,1)), requiring no manual intervention. Amplitude correction: To avoid amplitude distortion in the NMF results, linear regression is used for amplitude and mean correction. The separated pure cotton spectrum... Where std is a function for calculating the standard deviation, and the NMF full-spectrum search separation plot is shown below. Figure 3 As shown, Figure 3 (a) in the image is the reference pure cotton spectrum. Figure 3 (b) in the image shows the separated cotton component spectrum. Figure 3 (c) in the image is the reference pure polyester spectrum. Figure 3 In the image, (d) represents the separated peak spectrum, matched using the NMF algorithm peak position matching method, such as... Figure 4 As shown, Figure 4 (a) in the image represents the original spectrum. Figure 4(b) in the table shows the peak position matching details, and the expandable parameters are shown in Table 2:

[0091] Table 2

[0092]

[0093] SCA (Sparse Component Analysis):

[0094] The algorithm parameters are set as shown in Table 3:

[0095] Table 3

[0096]

[0097] Core process of sparse component analysis: Random orthogonal initialization: At the beginning of each iteration, a random matrix E∈ ²×², and orthogonalize it using the linear algebra method QR decomposition: [E,~]=QRQR(E'); E=E'; ensure that the initial separation matrix is ​​full rank and has a good condition number, where E' represents the transpose of E;

[0098] Improved natural gradient update: Separate signal estimation: S=E·X; Sparsity promotes nonlinear functions: Construct a sparsity metric using a log-hyperbolic function. Where tanh() is the hyperbolic tangent function, and the threshold of 0.1 can be adjusted to 0.05–0.2 depending on the noise level; natural gradient calculation and orthogonalization: gradient grad=(I-φ·S / T)·W, where I represents the identity matrix, S T Let E_new be the transpose of the separated signal estimate S; T represents the number of samples, and the superscript T indicates the transpose of the matrix. The updated matrix E_new = E + learn_rate (learning rate) · grad; then, symmetric orthogonalization is performed using compact SVD (singular value decomposition): [U,~,V] = svd(E_new,'econ'); where 'econ' is a parameter in the SVD function, representing "economic". E = U · V Where U represents the left singular matrix and V represents the right singular matrix. This step ensures that the separation matrices remain orthogonal after each iteration, reducing accumulated error;

[0099] Convergence criterion: Terminate early when the difference in F-norms ||E-E_new||_F < tolerance, thus saving computational resources;

[0100] Symbol alignment and similarity assessment: Bimodal symbol correction:

[0101] To prevent the inherent sign ambiguity of ICA / SCA, the following correlation coefficients are calculated separately: positive correlation coefficient corr_pos1 = corr(+source1, pure_cotton); negative correlation coefficient corr_neg1 = corr(-source1, pure_cotton). Similarly, source2 and pure_polyester are processed, and the sign configuration is maximized to achieve a total similarity of total_similarity = |corr_cotton| + |corr_polyester|, where pure_polyester represents the pure polyester reference spectrum and corr_polyester represents the pure polyester correlation coefficient. The true value of the correlation coefficient (including the sign) is recorded.

[0102] Normalization: Divide the best estimated source by its own L2 norm to eliminate amplitude ambiguity: the estimated source signal source_i = source_i / ||source_i||2; ||source_i||2 represents the L2 norm (also known as the Euclidean norm) of source_i. The SCA full-spectrum search separation graph is shown below. Figure 5 As shown, Figure 5 (a) in the image represents the original pure cotton spectrum. Figure 5 (b) in the image shows the separated cotton spectrum. Figure 5 (c) in the image represents the original pure polyester spectrum. Figure 5 (d) represents the separated distillation spectrum, and the extended and adjustable parameters are shown in Table 4:

[0103] Table 4

[0104]

[0105] Time-Frequency Masking Analysis (TFMBSS, blind source separation based on time-frequency masking):

[0106] The core algorithm parameters are shown in Table 5:

[0107] Table 5

[0108]

[0109] Core steps of the proximal alternating direction multiplier method (Proximal ADMM-like):

[0110] Random initialization, main variable w∈ ^{N²}, where Let N represent the set of real numbers. 2 Let be a 3D real vector space, which follows a standard normal distribution. Where randn refers to the random number generation function, and the subsequent reshape function reshapes W∈ ^{N×N}; auxiliary variable y∈ ^{N×T_half×K}, where, ^{N×T_half×K}: Represents a three-dimensional real space, where the three parameters correspond to the three dimensions of the three-dimensional matrix.

[0111] N: The size of the first dimension (consistent with the dimension of the matrix W after the main variable w is reshaped);

[0112] T_half: The size of the second dimension (inferred from the context, it is "half of the time / sample length", possibly the length after dividing the full sample length T into two parts);

[0113] K: The size of the third dimension (representing the total number of segments the data is divided into).

[0114] Similarly, random initialization is used to introduce sparse constraints;

[0115] Gradient calculation and near-end update:

[0116] For each segment k=1…K: x_segment=Y(:,:,k); where x_segment represents the k-th segment of data in segmented data Y, y_segment=y(:,:,k); where y_segment represents the k-th segment of data in auxiliary variable y corresponding to x_segment, grad+=μ1μ2·(x_segment(:)) ·y_segment(:)); Efficient matrix multiplication is achieved through flattening and vectorization; l2,1 proximal operator (group sparsity): w_out=prox_l21(w_in,λ), which performs soft thresholding on the column vector group to achieve inter-group sparsity and intra-group joint sparsity; where grad represents the gradient (short for gradient), the partial derivative of the objective function with respect to the parameters, used to guide the direction of parameter update.

[0117] μ1, μ2: Two scalar hyperparameters (usually denoted as mu1, mu2) used to adjust the scaling factors of the gradient;

[0118] x_segment: The k-th segment of data in segmented data Y (segment represents a data segment).

[0119] y_segment: The k-th segment of data in the auxiliary variable y that corresponds to x_segment;

[0120] l2,1 proximal operator: l2,1 norm proximal operator, an optimization tool for solving problems with l2,1 norm regularization;

[0121] prox_l21: A functional representation of the l2,1 norm proximal operator used to implement group sparsity constraints;

[0122] w_in and w_out: Input and output variables of the proximal operator, representing the main variables before and after processing, respectively;

[0123] λ: Regularization parameter (lambda), which controls the strength of sparsity constraints. The larger the value, the stronger the sparsity.

[0124] Update of auxiliary variable y and mask sparsification:

[0125] Calculate intermediate values: z = y_segment + (2·reshape(hat_w,N,N) - reshape(w,N,N)) • x_segment; Maximum response mask: M = generateMaxMask(z); Keep the element with the largest absolute value in each column, and set the rest to zero to further enhance the sparse representation;

[0126] Momentum update: α represents the weight that controls the historical information and the current estimate;

[0127] Momentum update for main variable w: w = α·hat_w + (1-α)·w; ensuring stable convergence of the iteration;

[0128] hat_w: An estimate of the main variable w (usually represented as w with a hat), used for intermediate calculations;

[0129] generateMaxMask: A function that generates the maximum response mask, used to filter important elements;

[0130] M: Maximum response mask matrix, which enhances sparsity by keeping the element with the largest absolute value in each column and setting the rest to zero;

[0131] α: Momentum parameter (alpha), which controls the weight ratio of historical information in the current update (0<α<1);

[0132] T_half: Half of the time / sample dimension (may indicate that the data is divided into two parts).

[0133] K: The total number of segments, indicating that the overall data will be divided into K segments for processing;

[0134] N: The dimension of the matrix, here indicating that the dimension of the square matrix W is N×N;

[0135] TFMBSS full-spectrum retrieval separation diagram, such as Figure 6 As shown, Figure 6 (a) in the image represents the original pure cotton spectrum. Figure 6 (b) in the image shows the separated cotton spectrum. Figure 6 (c) in the image represents the original pure polyester spectrum. Figure 6 (d) in the figure represents the separation spectrum. The scalable parameters and applicable scenarios are shown in Table 6.

[0136] Table 6

[0137]

[0138] S3, construct a standard spectral library, and calibrate the standard spectra in the standard library and the separated spectra;

[0139] Specifically, step S3 includes step S31:

[0140] S31, the standard spectra in the standard library and the separated spectra are baseline corrected using the downsampling baseline correction method;

[0141] It should be explained that a standard spectral library is constructed, and each standard spectral record in the library includes the name of the pure component; the position of the key absorption peak; and the full-spectrum vector. In this embodiment, the correction process employs a downsampling baseline correction method to perform baseline correction on both the standard spectrum and the separated spectrum.

[0142] S4, detect the local extrema of each of the source spectra to obtain the peak set;

[0143] S5, Match the peak position set with the standard peak positions of each pure component in the standard spectral library, and determine the number of intersections;

[0144] S6. If the number of intersections is greater than a preset value, it is determined that the component exists, and a list of existing components and the corresponding matching degree or similarity score are output.

[0145] Understandably, the standard fractions of each pure component in the standard spectral library are matched, and if the intersection number... In this embodiment, if there are ≥3 peaks in N_min, the component is determined to exist, and a "list of existing components" and the corresponding matching or similarity score are output, without providing a concentration value.

[0146] It is worth noting that the cosine similarity, Euclidean distance, or correlation coefficient between each pure component in the standard spectral library and the peak set can also be calculated, and the existence of the component can be determined based on the cosine similarity, Euclidean distance, or correlation coefficient.

[0147] Specific application scenarios, for example, Scenario 1: Deploy a portable NIR spectrometer at the receiving point of the weaving workshop to scan each roll of fabric labeled "pure cotton" for 2 seconds. The BSS module outputs two source spectra in real time: if one of them matches the "polyester characteristic peaks 1050nm and 1415nm double peaks" in the library with a matching degree r≥0.85, a red light will immediately illuminate and the signboard will display "contains more than 3% polyester". On-site workers can then take photos as evidence and refuse to accept the fabric.

[0148] Scenario 2: Customs officers conduct random inspections on a batch of containers declared as "aspirin raw materials." A handheld Raman spectroscopy device completes spectral acquisition within 5 seconds, and the BSS separates 3 source spectra; one of them shows a match of r=0.92 with the standard acetaminophen spectrum. The system immediately indicates "8% acetaminophen adulteration," and simultaneously generates a timestamped PDF report and uploads it to the regulatory platform.

[0149] Scenario 3: Gasoline blending monitoring. A fiber optic probe is installed on the top of the blending tank to collect the NIR spectrum of the blended gasoline in real time. The BSS decomposes the spectrum into three source spectra: "aromatics," "olefins," and "alkanes." The system calculates the RON (Research Octane Number) in real time based on the source intensity. When the octane number deviation exceeds 0.3 units, the additive flow rate is automatically adjusted.

[0150] During the implementation of the example:

[0151] 1. Sample preparation

[0152] Fold the knitted jersey fabric into 4 layers (thickness ≈ 2mm) and place it in the center of the portable near-infrared spectrometer (NIR) compression device.

[0153] Five different locations were randomly selected for each sample for measurement, and the average spectrum was used as the raw data.

[0154] 2. Blind Source Separation (BSS) Parameters – SCA Settings

[0155] Algorithm: Sparse Component Analysis (SCA), based on l2,1 regularization + Proximal Alternating Direction Multiplier Method (ProximalADMM-like).

[0156] Source number k: 3 (cotton, polyester, and spandex components).

[0157] Sparse threshold λ: 0.01 (adjustable range 5×10⁻⁶) - ³–1×10 - ¹, based on signal-to-noise ratio (adaptive).

[0158] Step size parameters μ1 and μ2: both set to 0.01.

[0159] Momentum / relaxation weight α: 0.5.

[0160] Maximum number of iterations: 100 rounds (internal ADMM loop), external random initialization for 10 rounds to select the optimal value.

[0161] Preprocessing methods:

[0162] – Normalization: Vector normalization (L2-norm=1).

[0163] – Baseline correction: Polynomial baseline (3rd order fit).

[0164] –SNV (Standard Normal Variable) transformation to eliminate scattering differences

[0165] 3. Standard spectral library, as shown in Table 7.

[0166] Table 7

[0167]

[0168] 4. Test Procedure (SCA Process)

[0169] (1) Collect the original spectrum: obtain 5 spectra at 5 different locations and calculate the average.

[0170] (2) Constructing the observation matrix: Divide the 2-channel spectrum into two segments to form a three-dimensional tensor X∈ ²×T / 2×2.

[0171] (3) Run SCA:

[0172] a. Randomly initialize the main variable w∈ 4 and auxiliary variable y∈ ²×T / 2×2;

[0173] b. Iteratively update the maximum response mask by near-end l2,1 contraction until convergence;

[0174] c. Output three sparse source spectra S1, S2, and S3.

[0175] (4) Compare with the image library:

[0176] – Calculate the Pearson correlation coefficient r between each source spectrum and the standard spectrum in the library;

[0177] –Discrimination threshold: r ≥ 0.85 is considered a successful match.

[0178] (5) Result determination:

[0179] –S1 and the cotton spectrum r=0.90, identified as cotton;

[0180] –S2 and the polyester spectrum r=0.92, indicating it is polyester;

[0181] –S3 and spandex spectra r=0.87, identified as spandex. Identification of cotton-polyester binary blends;

[0182] Results: Three source spectra were separated, one of which was located at 3300 and 1730 cm⁻¹. - ¹Successfully matched with spandex, indicating the presence of spandex. The spectral diagram of the cotton-polyester fabric is shown below. Figure 7 As shown.

[0183] In summary, the model-free intelligent identification method for mixture components in the above embodiments of the present invention employs four methods—IVA (Independent Vector Analysis), NMF (Non-negative Matrix Factorization), SCA (Sparse Algorithm Analysis), and TFMBSS (Time-Frequency Masking-Based Blind Source Separation)—to separate cotton-polyester textile mixtures. It then uses peak matching or full-spectrum retrieval to match the spectra of standard pure cotton and pure polyester, outputting existence conclusions within seconds. This solves the pain point of traditional methods relying on modeling and time-consuming verification, achieving non-destructive, rapid, and accurate qualitative verification of specific components in textile mixtures.

[0184] Example 2

[0185] The second embodiment of the present invention also provides a model-free intelligent identification system for mixture components. Please refer to [link to relevant documentation]. Figure 8 The system includes:

[0186] The acquisition and preprocessing module 10 is used to acquire several mixture spectra of the sample to be tested and to preprocess the several mixture spectra.

[0187] The output module 20 is used to arrange the preprocessed spectra of the mixture into a matrix and separate and output several source spectra.

[0188] A calibration module 30 is constructed to construct a standard spectral library and to calibrate the standard spectra in the standard library and the separated spectra.

[0189] The detection module 40 is used to detect the local extrema of each of the source spectra to obtain a set of peak positions;

[0190] The matching and judgment module 50 is used to match the peak position set with the standard peak positions of each pure component in the standard spectral library and determine the number of intersections.

[0191] The determination output module 60 is used to determine the existence of a component if the number of intersections is greater than a preset value, and output a list of existing components and the corresponding matching degree or similarity score.

[0192] In some optional embodiments, the acquisition preprocessing module 10 includes:

[0193] The acquisition unit is used to acquire the original spectral vector of the sample to be tested based on the spectrometer, so as to obtain several spectra of the mixture;

[0194] The processing unit is used to sequentially perform baseline correction, normalization, smoothing, and noise reduction on several of the mixture spectra.

[0195] In some alternative embodiments, the arrangement output module 20 includes:

[0196] Arrangement unit, used to arrange several spectra of the mixture, matching the number of separated components, into a matrix;

[0197] The analysis output unit is used to perform independent vector analysis, nonnegative matrix decomposition, constrained nonnegative spectroscopy, and sparse component analysis based on spectral characteristics, and to analyze the matrix using spectral sparsity or based on time-frequency masking to output several source spectra.

[0198] In some alternative embodiments, the construction correction module 30 includes:

[0199] The correction unit is used to perform baseline correction on the standard spectra in the standard library and the separated spectra using a downsampling baseline correction method.

[0200] In some alternative embodiments, the detection module 40 includes:

[0201] The calculation unit is used to calculate the cosine similarity, Euclidean distance, or correlation coefficient between each pure component in the standard spectral library and the peak set, and to determine whether the component exists based on the cosine similarity, the Euclidean distance, or the correlation coefficient.

[0202] The functions or operation steps implemented by the above modules and units are largely the same as those in the above method embodiments, and will not be repeated here.

[0203] The model-free intelligent identification system for mixture components provided in this embodiment of the invention has the same implementation principle and technical effects as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the system embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0204] Example 3

[0205] The third embodiment of the present invention also proposes an electronic device, please refer to [link / reference]. Figure 9 The image shows an electronic device according to a third embodiment of the present invention.

[0206] The electronic device may include a processor 71 and a memory 72 storing computer program instructions.

[0207] Specifically, the processor 71 may include a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement this application.

[0208] The memory 72 may include a mass storage device for data or instructions. For example, and not limitingly, the memory 72 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 72 may include removable or non-removable (or fixed) media. Where appropriate, the memory 72 may be internal or external to a data processing device. In a particular embodiment, the memory 72 is non-volatile memory. In a particular embodiment, the memory 72 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random-Access Memory (FPMDRAM), Extended Data Out Dynamic Random-Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0209] The memory 72 can be used to store or cache various data files that need to be processed and / or communicated, as well as possible computer program instructions executed by the processor 71.

[0210] The processor 71 reads and executes the computer program instructions stored in the memory 72 to implement the model-free intelligent identification method for mixture components described in Embodiment 1.

[0211] In some embodiments, the electronic device may further include a communication interface 73 and a bus 70. For example, Figure 9 As shown, the processor 71, memory 72, and communication interface 73 are connected through bus 70 and complete communication with each other.

[0212] The communication interface 73 is used to enable communication between the various modules, devices, units, and / or equipment in this application. The communication interface 73 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0213] Bus 70 includes hardware, software, or both, that couples components of a device together. Bus 70 includes, but is not limited to, at least one of the following: data bus, address bus, control bus, expansion bus, and local bus. For example, and not as a limitation, bus 70 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 70 may include one or more buses. Although this application describes and illustrates a specific bus, this application considers any suitable bus or interconnection.

[0214] The electronic device can acquire a model-free intelligent identification system for mixture components and execute the model-free intelligent identification method for mixture components in this embodiment.

[0215] Furthermore, in conjunction with the model-free intelligent identification method for mixture components in Embodiment 1 above, this application can provide a storage medium for implementation. This storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement the model-free intelligent identification method for mixture components in Embodiment 1 above.

[0216] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0217] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A modeling-free mixture component intelligent identification method, characterized in that, The method comprises: a plurality of mixture spectra of a sample to be tested are collected, and the plurality of mixture spectra are preprocessed; the preprocessed plurality of mixture spectra are arranged into a matrix, and a plurality of source spectra are separated and output, the step specifically comprising: a plurality of the mixture spectra matching the number of separated components are arranged into a matrix; independent vector analysis, non-negative matrix factorization, constrained non-negative spectra, and sparse component analysis are used to analyze the matrix according to spectral characteristics, and spectral sparsity or time-frequency mask-based analysis is used to output the plurality of source spectra and a source spectrum matrix; a standard spectrum library is constructed, and standard spectra in the standard spectrum library and the separated spectra are corrected, the step specifically comprising: a baseline correction method of down-sampling is used to perform baseline correction on the standard spectra in the standard spectrum library and the separated spectra, wherein each standard spectrum comprises a pure component name, a key absorption peak position, and a full spectrum vector; local extreme values of each source spectrum are detected to obtain a peak position set; the peak position set is matched with each pure component standard peak position in the standard spectrum library, and the number of intersections is determined; if the number of intersections is greater than a preset value, it is determined that a component exists, and an existing component list and a corresponding matching degree or similarity score are output.

2. The modeling-free mixture component intelligent identification method according to claim 1, characterized in that, The step of collecting a plurality of mixture spectra of a sample to be tested and preprocessing the plurality of mixture spectra comprises: an original spectrum vector of the sample to be tested is obtained based on a spectrometer to obtain the plurality of mixture spectra; baseline correction, normalization, smoothing, and denoising are sequentially performed on the plurality of mixture spectra.

3. The modeling-free mixture component intelligent identification method according to claim 1, characterized in that, After the step of detecting local extreme values of each source spectrum to obtain a peak position set, the method further comprises: cosine similarity, Euclidean distance, or correlation coefficient of each pure component in the standard spectrum library and the peak position set is calculated, and whether a component exists is determined according to the cosine similarity, the Euclidean distance, or the correlation coefficient.

4. A modeling-free mixture component intelligent identification system, characterized in that, The system comprises: a collection and preprocessing module configured to collect a plurality of mixture spectra of a sample to be tested and preprocess the plurality of mixture spectra; an arrangement and output module configured to arrange the preprocessed plurality of mixture spectra into a matrix and separate and output a plurality of source spectra; the arrangement and output module comprises: an arrangement unit configured to arrange a plurality of the mixture spectra matching the number of separated components into a matrix; an analysis and output unit configured to use independent vector analysis, non-negative matrix factorization, constrained non-negative spectra, and sparse component analysis to analyze the matrix according to spectral characteristics, and use spectral sparsity or time-frequency mask-based analysis to output the plurality of source spectra and a source spectrum matrix; a construction and correction module configured to construct a standard spectrum library and correct standard spectra in the standard spectrum library and the separated spectra; the construction and correction module comprises: A correction unit is configured to correct the standard spectra in the standard spectrum library and the separated spectra by using a method of down-sampling baseline correction, wherein each standard spectrum comprises a pure component name, key absorption peak positions, and a full spectrum vector; A detection module is configured to detect local extreme values of each source spectrum to obtain a peak position set; A matching judgment module is configured to match the peak position set with standard peak positions of each pure component in the standard spectrum library, and determine the number of intersections; A judgment output module is configured to determine that a component exists if the number of intersections is greater than a preset value, and output an existing component list and corresponding matching degree or similarity score.

5. The modeling-free mixture component intelligent identification system according to claim 4, wherein, The acquisition and preprocessing module comprises: An acquisition unit is configured to acquire original spectrum vectors of the sample to be tested based on a spectrometer to obtain a plurality of mixture spectra; A processing unit is configured to sequentially perform baseline correction, normalization, smoothing, and denoising processing on the plurality of mixture spectra.

6. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the computer program to implement the modeling-free intelligent identification method of mixture components according to any one of claims 1-3.

7. A storage medium having stored thereon a computer program, characterized in that The program is executed by the processor to implement the modeling-free intelligent identification method of mixture components according to any one of claims 1-3.

Images