CN115479905B - Spectral analysis method, spectral analysis device, terminal equipment and medium - Google Patents
Spectral analysis method, spectral analysis device, terminal equipment and medium Download PDFInfo
- Publication number
- CN115479905B CN115479905B CN202211409528.5A CN202211409528A CN115479905B CN 115479905 B CN115479905 B CN 115479905B CN 202211409528 A CN202211409528 A CN 202211409528A CN 115479905 B CN115479905 B CN 115479905B
- Authority
- CN
- China
- Prior art keywords
- matrix
- batch
- spectral
- spectral data
- spectral analysis
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000010183 spectrum analysis Methods 0.000 title claims abstract description 123
- 238000000034 method Methods 0.000 title claims abstract description 70
- 239000011159 matrix material Substances 0.000 claims abstract description 260
- 230000003595 spectral effect Effects 0.000 claims abstract description 119
- 230000000694 effects Effects 0.000 claims description 104
- 238000001228 spectrum Methods 0.000 claims description 45
- 238000004422 calculation algorithm Methods 0.000 claims description 17
- 238000012545 processing Methods 0.000 claims description 15
- 239000002253 acid Substances 0.000 claims description 14
- 150000002148 esters Chemical class 0.000 claims description 14
- LFQSCWFLJHTTHZ-UHFFFAOYSA-N Ethanol Chemical compound CCO LFQSCWFLJHTTHZ-UHFFFAOYSA-N 0.000 claims description 10
- 238000012163 sequencing technique Methods 0.000 claims description 10
- 238000000354 decomposition reaction Methods 0.000 claims description 9
- 238000005457 optimization Methods 0.000 claims description 9
- 239000013598 vector Substances 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 7
- 238000004611 spectroscopical analysis Methods 0.000 claims description 4
- 238000004891 communication Methods 0.000 description 6
- 238000007636 ensemble learning method Methods 0.000 description 6
- 239000000126 substance Substances 0.000 description 6
- 238000010801 machine learning Methods 0.000 description 5
- 238000010586 diagram Methods 0.000 description 4
- 239000012636 effector Substances 0.000 description 4
- 238000005259 measurement Methods 0.000 description 4
- 238000012706 support-vector machine Methods 0.000 description 4
- 238000001237 Raman spectrum Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 3
- 238000004590 computer program Methods 0.000 description 3
- 238000002790 cross-validation Methods 0.000 description 3
- 238000003066 decision tree Methods 0.000 description 3
- 238000007637 random forest analysis Methods 0.000 description 3
- 239000000796 flavoring agent Substances 0.000 description 2
- 235000019634 flavors Nutrition 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000001476 alcoholic effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000005416 organic matter Substances 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Pure & Applied Mathematics (AREA)
- Mathematical Optimization (AREA)
- Mathematical Analysis (AREA)
- Computational Mathematics (AREA)
- Biochemistry (AREA)
- Life Sciences & Earth Sciences (AREA)
- Immunology (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Analytical Chemistry (AREA)
- Chemical & Material Sciences (AREA)
- Pathology (AREA)
- Algebra (AREA)
- Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a spectral analysis method, a device, a terminal device and a medium, wherein the method comprises the following steps: acquiring spectral data of a sample to be analyzed, and constructing a corresponding adjacency matrix according to the spectral data; removing batch errors of the spectral data based on the adjacency matrix; and performing ensemble learning on the spectral data with batch errors removed to obtain a corresponding spectral analysis result. The invention can improve the accuracy of the spectral analysis of the sample.
Description
Technical Field
The present invention relates to the field of data analysis technologies, and in particular, to a spectral analysis method, an apparatus, a terminal device, and a computer-readable storage medium.
Background
Spectroscopic analysis is an effective means of performing substance identification. Compared with the traditional spectrum analysis method, the method has the advantages that the machine learning algorithm is applied to analyze the spectrum, so that the cost can be obviously reduced, and the efficiency can be improved. However, when the sample is a complex mixture of multiple substances, the spectral information is the superposition of the spectra of the multiple substances. This makes it difficult for a single learner to extract the useful information contained in the spectral data when applying machine learning algorithms to spectral analysis. Moreover, in practical applications, the spectra of the samples are often not measured in the same batch. The error caused by the different measurement batches is called batch effect. For some spectral data with strong batch effect, it is critical to effectively remove or reduce the influence of the batch effect for spectral analysis by using a machine learning algorithm.
Therefore, the accuracy of the spectral analysis result is greatly reduced by the existing mode of using a single learner to perform spectral analysis, and in addition, the influence of batch effect on the spectral analysis result is ignored, so that the obtained spectral analysis result is more unreliable.
Disclosure of Invention
The invention mainly aims to provide a spectral analysis method, a spectral analysis device, terminal equipment and a computer readable storage medium, and aims to improve the accuracy of spectral analysis of spectral data.
To achieve the above object, the present invention provides a spectral analysis method, comprising the steps of:
acquiring spectral data of a sample to be analyzed, and constructing a corresponding adjacency matrix according to the spectral data;
removing batch errors of the spectral data based on the adjacency matrix;
and performing ensemble learning on the spectral data with the batch errors removed to obtain a corresponding spectral analysis result.
Optionally, the spectral data includes spectral features, and the step of constructing a corresponding adjacency matrix from the spectral data includes:
determining a similarity graph among the spectrums of the samples to be analyzed based on the spectrum characteristics, and determining each node in the similarity graph;
sequencing the distances between the characteristic vectors corresponding to the nodes to obtain a sequencing result;
and constructing a node set corresponding to each node based on the sequencing result, and constructing an adjacency matrix based on a plurality of node sets.
Optionally, after the step of constructing a node set corresponding to each node based on the sorting result, and constructing an adjacency matrix based on a plurality of node sets, the method further includes:
and determining a corresponding regularized Laplace matrix according to the adjacency matrix, and removing batch errors of the spectral data of the sample to be analyzed based on the regularized Laplace matrix.
Optionally, the step of removing batch errors of the spectral data based on the adjacency matrix comprises:
determining an initial feature matrix corresponding to the spectral data of the sample to be analyzed through a preset optimization algorithm based on the regularized Laplace matrix corresponding to the adjacency matrix, wherein the initial feature matrix is a matrix for removing batch effect and noise influence;
acquiring a corresponding initial batch effect factor matrix based on the initial characteristic matrix;
and determining a target initial characteristic matrix and a target batch effect factor matrix based on the initial batch effect factor matrix so as to remove the batch errors of the spectral data.
Optionally, the step of determining a target initial feature matrix and a target batch-wise effect factor matrix based on the initial batch-wise effect factor matrix to perform a batch error removal process on the spectral data includes:
performing non-negative matrix decomposition on the initial batch of effect factor matrixes to obtain a first non-negative matrix and a second non-negative matrix which are different in size;
acquiring a target batch effect factor matrix based on the first non-negative matrix and the second non-negative matrix, and determining whether the target batch effect factor matrix meets a preset low-rank constraint;
if yes, determining a corresponding target characteristic matrix based on the target batch effect factor matrix;
and removing the batch errors of the spectral data according to the target characteristic matrix and the target batch effect factor matrix.
Optionally, the step of performing ensemble learning on the spectral data with batch errors removed to obtain a corresponding spectral analysis result includes:
performing ensemble learning on the spectral data with batch errors removed through a preset ensemble learning model to obtain a corresponding spectral analysis result, wherein the ensemble learning model comprises: lightweight gradient hoist LightGBM.
Optionally, the step of performing ensemble learning on the spectrum data with the batch errors removed through a preset ensemble learning model to obtain a corresponding spectrum analysis result includes:
inputting the spectral data of the sample to be analyzed, from which the batch errors are removed, into the LightGBM, and obtaining a spectral analysis result output by the LightGBM, where the spectral analysis result includes, when the sample to be analyzed is the white spirit to be analyzed: one or more of alcohol content, total amount of acid ester, and taste index.
To achieve the above object, the present invention also provides a spectral analysis apparatus including:
the system comprises a construction module, a data acquisition module and a data processing module, wherein the construction module is used for acquiring spectral data of a sample to be analyzed and constructing a corresponding adjacency matrix according to the spectral data;
a removal processing module for removing batch errors of the spectral data based on the adjacency matrix;
and the integrated learning module is used for performing integrated learning on the spectral data with batch errors removed to obtain a corresponding spectral analysis result.
To achieve the above object, the present invention also provides a terminal device, which includes a memory, a processor, and a spectrum analysis program stored on the memory and executable on the processor, wherein the spectrum analysis program, when executed by the processor, implements the steps of the spectrum analysis method as described above.
Furthermore, to achieve the above object, the present invention also provides a computer readable storage medium, on which a spectrum analysis program is stored, the spectrum analysis program implementing the steps of the spectrum analysis method as described above when executed by a processor.
To achieve the above object, the present invention also provides a computer program product comprising a computer program which, when executed by a processor, implements the steps of the spectral analysis method as described above.
The invention provides a spectral analysis method, a device, a terminal device, a computer readable storage medium and a computer program product, which are used for constructing a corresponding adjacency matrix according to spectral data of a sample to be analyzed by acquiring the spectral data; removing batch errors of the spectral data based on the adjacency matrix; and performing ensemble learning on the spectral data with batch errors removed to obtain a corresponding spectral analysis result.
Compared with the spectral analysis mode that a single learner ignores batch effects in the prior art, the method and the device remove batch errors of the spectral data of the sample to be analyzed in advance, and then perform integrated learning on the spectral data with the batch effects removed, so that the spectral data of the sample is analyzed and processed. Therefore, the spectrum analysis method can perform spectrum analysis in an integrated learning mode after batch errors of the spectrum data are removed, and avoids the influence of the batch errors on the spectrum data.
Drawings
FIG. 1 is a schematic diagram of a hardware operating environment according to an embodiment of the present invention;
FIG. 2 is a first flowchart of an embodiment of a spectral analysis method according to the present invention;
FIG. 3 is a second flow chart of an embodiment of the spectral analysis method of the present invention;
FIG. 4 is a diagram illustrating test results of an embodiment of the spectral analysis method of the present invention;
FIG. 5 is a functional block diagram of an embodiment of a spectral analysis apparatus according to the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the invention.
As shown in fig. 1, fig. 1 is a schematic device structure diagram of a hardware operating environment according to an embodiment of the present invention.
The terminal device of the embodiment of the invention can be a smart phone, a computer, a server, other network devices and the like, and can be used for realizing spectral analysis of a sample.
As shown in fig. 1, the terminal device may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, a communication bus 1002. Wherein a communication bus 1002 is used to enable connective communication between these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1005 may be a high-speed RAM memory or a non-volatile memory such as a disk memory. The memory 1005 may alternatively be a storage device separate from the processor 1001.
Those skilled in the art will appreciate that the device configuration shown in fig. 1 does not constitute a limitation of the spectroscopic analysis device and may include more or fewer components than those shown, or some components may be combined, or a different arrangement of components.
As shown in fig. 1, the memory 1005, which is a kind of computer storage medium, may include therein an operation, a network communication module, a user interface module, and a spectrum analysis program. The operations are programs that manage and control the hardware and software resources of the device, supporting the operation of the spectral analysis program as well as other software or programs. In the device shown in fig. 1, the user interface 1003 is mainly used for data communication with a client; the network interface 1004 is mainly used for establishing communication connection with a server; and the processor 1001 may be configured to invoke the spectral analysis program stored in the memory 1005 and perform the following operations:
acquiring spectral data of a sample to be analyzed, and constructing a corresponding adjacency matrix according to the spectral data;
removing batch errors of the spectral data based on the adjacency matrix;
and performing ensemble learning on the spectral data with batch errors removed to obtain a corresponding spectral analysis result.
Further, the processor 1001 may be configured to call the spectral analysis program stored in the memory 1005 and perform the following operations:
determining a similarity graph among the spectrums of the samples to be analyzed based on the spectrum characteristics, and determining each node in the similarity graph;
sorting the distances between the characteristic vectors corresponding to the nodes to obtain a sorting result;
and constructing a node set corresponding to each node based on the sequencing result, and constructing an adjacency matrix based on a plurality of node sets.
Further, after the steps of constructing the node sets corresponding to the nodes based on the sorting result and constructing the adjacency matrix based on a plurality of node sets, the processor 1001 may be configured to invoke a spectrum analysis program stored in the memory 1005 and perform the following operations:
and determining a corresponding regularized Laplace matrix according to the adjacency matrix, and removing batch errors of the spectral data of the sample to be analyzed based on the regularized Laplace matrix.
Further, the processor 1001 may be configured to call the spectral analysis program stored in the memory 1005 and perform the following operations:
determining an initial feature matrix corresponding to the spectral data of the sample to be analyzed through a preset optimization algorithm based on the regularized Laplace matrix corresponding to the adjacency matrix, wherein the initial feature matrix is a matrix for removing batch effect and noise influence;
acquiring a corresponding initial batch effect factor matrix based on the initial characteristic matrix;
and determining a target initial characteristic matrix and a target batch effect factor matrix based on the initial batch effect factor matrix so as to remove the batch errors of the spectral data.
Further, the processor 1001 may be configured to call the spectral analysis program stored in the memory 1005 and perform the following operations:
carrying out non-negative matrix decomposition on the initial batch of effect factor matrixes to obtain a first non-negative matrix and a second non-negative matrix which are different in size;
acquiring a target batch effect factor matrix based on the first non-negative matrix and the second non-negative matrix, and determining whether the target batch effect factor matrix meets a preset low-rank constraint;
if yes, determining a corresponding target characteristic matrix based on the target batch effect factor matrix;
and removing the batch errors of the spectral data according to the target characteristic matrix and the target batch effect factor matrix.
Further, the processor 1001 may be configured to invoke a spectral analysis program stored in the memory 1005 and perform the following operations:
performing ensemble learning on the spectral data with batch errors removed through a preset ensemble learning model to obtain a corresponding spectral analysis result, wherein the ensemble learning model comprises: lightweight gradient hoist LightGBM.
Further, the processor 1001 may be configured to invoke a spectral analysis program stored in the memory 1005 and perform the following operations:
inputting the spectral data of the sample to be analyzed with the batch errors removed into the LightGBM, and obtaining a spectral analysis result output by the LightGBM, wherein when the sample to be analyzed is the white spirit to be analyzed, the spectral analysis result includes: one or more of alcohol content, total amount of acid ester, and taste index.
Referring to fig. 2, fig. 2 is a schematic flow chart of a spectral analysis method according to a first embodiment of the present invention.
While a logical order is shown in the flow chart, in some cases, the steps shown or described may be performed in an order different than presented herein.
Considering the existing substance spectrum analysis method, on one hand, only a single learner is used for spectrum analysis, and on the other hand, because the spectra of the samples are often not measured in the same batch, the error caused by the different measurement batches is called batch effect (batch effect), and for some spectrum data with strong batch effect, the accuracy of spectrum analysis using a machine learning algorithm is greatly reduced due to the fact that the influence of the batch effect cannot be effectively removed or reduced.
In view of the above problems, the present invention provides a method for analyzing a spectrum, which is intended to perform a precise analysis on a spectrum of a sample, as shown in fig. 3, and includes: a batch effect removal (batch effect removal) method based on data-adaptive shrinkage and clustering (DASC) effectively removes (or reduces) batch effects among spectra of samples, and performs ensemble learning on samples to be analyzed with batch effects removed to obtain corresponding spectral analysis results through an ensemble learning model. Therefore, the spectral analysis precision is greatly improved.
It is worth noting that in the present invention, the removal principle of the batch effect is as follows:
assuming that n samples exist in the data set, the characteristic dimension of each sample is m, and the size of a characteristic matrix X obtained through measurement is n × m. Suppose that:
wherein, U represents a characteristic matrix for removing the batch effect and the noise influence, B represents a batch effect factor, and E represents noise. The matrix U satisfies the manifold constraint, i.e. makes similar features of different batches as close as possible in the manifold. Matrix B satisfies the low rank constraint, i.e., the batch effect factors affecting the experimental results are as low as possible (the basis of matrix B is as low as possible). The noise matrix E satisfies the sparsity constraint.
Further, optimization issues can be considered:
wherein,
the Frobenius norm (F-norm) of the representation matrix,
represents the kernel norm of the matrix, tr (-) represents the traces of the matrix, and L is an n-th order square matrix representing the regularized Laplace matrix of the features.
On this basis, if the batch effect of the sample spectrum (i.e. the batch error in this embodiment) is removed, the following solution is required:
obtaining initial U, enabling B = X-U to obtain initial B, and performing non-Negative Matrix Factorization (NMF) on the initial B to obtain optimized low-rank B * Finally, U * =U-B * To achieve removal of batch errors.
Thus, in the present invention, as shown in fig. 3, in order to apply DASC to perform batch effect removal, it is first required to model the features of n samples into a graph (graph) having n nodes, and then calculate the laplacian matrix of the graph. After batch effect removal using DASC, the spectra were learned using LightGBM to obtain spectral analysis results (classification or regression).
Specifically, the spectral analysis method in the present embodiment includes the following steps:
s10, acquiring spectral data of a sample to be analyzed, and constructing a corresponding adjacency matrix according to the spectral data;
in this embodiment, the sample to be analyzed may be a sample of any substance, for example, in order to measure parameters such as alcohol degree and total amount of acid ester of white spirit, a certain brand of white spirit to be analyzed may be used as the sample to be analyzed. The spectral data in this embodiment may be the spectral characteristics of the sample to be analyzed.
On this basis, the terminal device first obtains spectral data (specifically, spectral characteristics) of a sample to be analyzed, and then establishes a corresponding adjacency matrix according to the spectral characteristics.
It should be noted that, in the present embodiment, the constructed adjacency matrix is used for removing the subsequent batch effect (i.e. the batch error in the present embodiment) of the spectral data.
Step S20, removing batch errors of the spectral data based on the adjacency matrix;
and step S30, performing ensemble learning on the spectral data with batch errors removed to obtain a corresponding spectral analysis result.
After the terminal device obtains the adjacency matrix corresponding to the spectral characteristics, the batch errors of the samples to be analyzed are removed according to the adjacency matrix, and further, the integrated learning is performed on the tube spectral characteristics with the batch errors removed, for example, the spectral characteristics with the batch errors removed are subjected to spectral analysis through an integrated learning model, and a corresponding spectral analysis result is obtained.
In this embodiment, the terminal device first obtains a spectral feature of a sample to be analyzed, and then establishes an adjacency matrix corresponding to the spectral data according to the spectral data. And then. According to the adjacency matrix, batch errors of the samples to be analyzed are removed, and further, ensemble learning is carried out on tube spectral features with the batch errors removed.
Compared with the spectral analysis mode that a single learner ignores batch effects in the prior art, the method and the device remove the batch effects of the spectral data of the sample in advance, further perform ensemble learning on the spectral data with the batch effects removed, and realize analysis processing on the spectral data of the sample. Therefore, the spectrum analysis method can perform spectrum analysis in an integrated learning mode after the batch effect of the spectrum data is removed, so that the influence of the batch effect on the spectrum data is avoided, and the integrated learning mode is adopted, so that the spectrum analysis precision and the spectrum analysis efficiency are improved compared with a single learner, and meanwhile, compared with the traditional spectrum detection and analysis method, the spectrum analysis cost is reduced based on the machine learning algorithm of the integrated learning.
Further, a second embodiment of the spectral analysis method of the present invention is proposed based on the first embodiment of the spectral analysis method of the present invention.
In this embodiment, in the step S10, "constructing a corresponding adjacency matrix according to the spectral data" may include:
step S101, determining a similarity graph among spectrums of the sample to be analyzed based on the spectrum characteristics, and determining each node in the similarity graph;
s102, sequencing the distances among the characteristic vectors of the nodes to obtain a sequencing result;
and step S103, constructing a node set corresponding to each node based on the sequencing result, and constructing an adjacency matrix based on a plurality of node sets.
In this embodiment, in order to remove the batch effect of the spectral data, it is necessary to construct an adjacency matrix of the samples to be analyzed in advance, which includes: and acquiring a corresponding similarity graph of the sample to be analyzed based on the spectrum data, and further constructing an adjacency matrix corresponding to the spectrum characteristic based on nodes in the similarity graph.
Specifically, for example, the terminal device constructs a similarity graph between spectra of the samples to be analyzed in advance according to spectral features of the samples to be analyzed, acquires each node in the similarity graph, calculates a distance between feature vectors corresponding to each node on the basis of the similarity graph (in this embodiment, the distance between the feature vectors may be an euclidean distance), and sorts the distances to obtain a sorting result.
Furthermore, according to the sorting result, t nearest nodes except the node i are determined, the node set of the t nodes forms a neighborhood N (i), and further, based on the N (i), a (self-loop) adjacency matrix is constructed:
wherein, the "" indicates a Hadamard product (Hadamard product), the matrix C satisfies: if j ∈ N (i), C ij =1; otherwise, C ij =0. In this embodiment, t may be set to 10, and in addition, t may also be set to other values, and the value of t is not specifically limited in this embodiment.
Through the operation, the terminal device determines an adjacent matrix corresponding to the spectral characteristics of the sample to be analyzed, so as to further perform the batch effect removal or reduction operation based on the adjacent matrix.
Further, after the step S103, the method may further include:
step S104, determining a corresponding regularized Laplace matrix according to the adjacency matrix, and removing batch errors of the spectral data based on the regularized Laplace matrix.
After the terminal device acquires the adjacency matrix corresponding to the spectral data of the sample to be analyzed, the regularized laplacian matrix corresponding to the adjacency matrix is acquired, so that the regularized laplacian matrix is utilized to further remove batch effects.
Specifically, as described above, the terminal device acquires the adjacency matrix
Then, the corresponding regularized laplacian matrix is obtained as follows:
wherein I is an identity matrix, D is a degree matrix, and the degree matrix is a diagonal matrix defined as
In this way, the terminal device determines a corresponding regularized laplacian matrix from the adjacency matrix.
Further, in the step S20, the removing the batch error of the spectral data based on the adjacency matrix may include:
step S201, determining an initial characteristic matrix corresponding to the spectral data of the sample to be analyzed through a preset optimization algorithm based on the regularized Laplace matrix corresponding to the adjacency matrix, wherein the initial characteristic matrix is a matrix for removing batch effect and noise influence;
step S202, acquiring a corresponding initial batch effect factor matrix based on the initial characteristic matrix;
step S203, determining a target initial feature matrix and a target batch effect factor matrix based on the initial batch effect factor matrix, so as to remove the batch error of the spectral data.
In this embodiment, as described above:
assuming that n spectra are in a data set of a sample to be analyzed, and the number of wave bands of each spectrum is m, the size of a measured spectral feature matrix X is n × m, and an initial feature matrix U corresponding to spectral data is solved through a preset optimization algorithm (i.e., the above formula):
it is noted that, in order to reduce the amount of calculation and improve the efficiency of spectral analysis, the expression of the initial feature matrix U may be used
Is approximately as
。
It can be appreciated that because the eigenvalues of the regularized laplacian matrix are distributed over the interval [1, 2), at λ <0.5, there is:
therefore, the error of approximating (I + λ L) -1 by (I- λ L) is not more than
. Therefore, when λ is small, it will be
Is approximately as
The method is reasonable, and not only can reduce the complexity of spectral analysis but also can improve the efficiency of spectral analysis on the basis of not influencing the precision of spectral analysis. In the present embodiment, λ is set to 0.01 specifically, and in addition to this, λ may be determined to other values.
Further, by the above formula
Acquiring a corresponding initial batch effect factor matrix B:
thus, the terminal device obtains the initial feature matrix, and obtains the corresponding initial batch effect factor matrix according to the approximated initial feature matrix, so as to further determine the final target initial feature matrix and the target batch effect factor matrix based on the initial batch effect factor matrix B to perform the batch error removal processing of the spectral data.
Further, in step S203, the determining a target initial characteristic matrix and a target batch-effect factor matrix based on the initial batch-effect factor matrix to perform a removal process on the batch errors of the spectral data may include:
step S2031, carrying out non-negative matrix decomposition on the effect factor matrix of the initial batch to obtain a first non-negative matrix and a second non-negative matrix with different sizes;
step S2032, acquiring a target batch effect factor matrix based on the first non-negative matrix and the second non-negative matrix, and determining whether the target batch effect factor matrix meets a preset low-rank constraint;
step S2033, if yes, determining a corresponding target characteristic matrix based on the target batch effect factor matrix;
and S2034, removing batch errors of the spectral data according to the target characteristic matrix and the target batch effect factor matrix.
After the terminal equipment acquires the initial batch of effector factor matrix B, non-Negative Matrix Factorization (NMF) is carried out on the initial batch of effector factor matrix B through an iterative algorithm to obtain a first non-negative matrix with the size of n × k and a second non-negative matrix with the size of k × m, wherein n, k and m represent rows and columns of the matrix, in addition,
。
at this point, the demand solution optimization problem:
to achieve batch effect removal of spectral data.
Therefore, based on the above formula
Determining a target batch effector matrix B = WH consisting of:
it can be known that the rank of B is not greater than k, and the low rank constraint is satisfied, and therefore, the terminal device obtains an optimized low rank B.
On this basis, the target feature matrix U = X-B. Thus, the terminal device determines the target feature matrix U and the target batch effect factor matrix B, and realizes the batch effect of the sample spectral data to be analyzed. In this embodiment, the value of k and the number of iterations of the initial batch effector matrix B are not specifically limited, for example, k may be 32, and the number of iterations is 20.
Further, in the step S30, "performing ensemble learning on the spectrum data with the batch errors removed to obtain the corresponding spectrum analysis result" may include:
step S301, performing ensemble learning on the spectrum data with batch errors removed through a preset ensemble learning model to obtain a corresponding spectrum analysis result, wherein the ensemble learning model comprises: lightweight gradient hoist LightGBM.
It should be noted that, in this embodiment, the terminal device pre-constructs the ensemble learning model, and the ensemble learning module used in this embodiment may be a lightweight gradient elevator LightGBM, which reduces memory consumption by optimizing feature storage and increases operation speed by a parallel computing architecture, compared with other ensemble learning algorithms, thereby increasing time efficiency and space efficiency of the ensemble learning algorithm, and thus further increasing spectral analysis efficiency. In addition, other ensemble learning models capable of spectral analysis may be used. The LightGBM construction method in this embodiment may refer to an existing model construction method, and is not particularly limited, nor is the LightGBM construction method the key content of the present invention.
Therefore, the terminal equipment can utilize the constructed LightGBM to perform ensemble learning on the spectral data with batch errors removed to obtain a corresponding spectral analysis result.
Further, the step S301 may include:
step S3011, outputting the spectral data of the sample to be analyzed with batch errors removed to the ensemble learning model, and obtaining a spectral analysis result output by the ensemble learning model, wherein when the sample to be analyzed is white spirit to be analyzed, the spectral analysis result comprises: alcohol content, total amount of acid ester, and taste index.
It should be noted that, in this embodiment, the integrated learning model is used to analyze the spectrum data after the batch effect is removed, so that the spectrum analysis efficiency is greatly improved, and the spectrum analysis accuracy is also improved.
Specifically, for example, the terminal device further outputs the spectral data with the batch effect removed to the LightGBM, and obtains the spectral analysis result output by the LightGBM, it can be understood that the type of the sample to be analyzed is different, and the corresponding spectral analysis result is also different, for example, when the sample to be analyzed is a certain brand of white spirit, the spectral analysis result includes: alcohol content, total amount of acid ester, taste index, etc.
In this embodiment, the terminal device constructs a similarity graph between spectra of samples to be analyzed in advance according to spectral features of the samples to be analyzed, acquires each node in the similarity graph, calculates distances between feature vectors corresponding to each node on the basis, and sorts the distances to obtain a sorting result. And then, according to the sequencing result, determining t nearest nodes except the node i of any node i, wherein the node set of the t nodes forms a neighborhood matrix. The terminal equipment acquires an initial characteristic matrix, acquires a corresponding initial batch effect factor matrix according to the approximated initial characteristic matrix, and carries out non-negative matrix decomposition on the initial batch effect factor matrix B to obtain a first non-negative matrix and a second non-negative matrix. And further, a target batch effect factor matrix and a target characteristic matrix are determined, and the batch effect of the spectral data of the sample to be analyzed is realized. Finally, the terminal device outputs the spectral data with the batch effect removed to the constructed LightGBM, and obtains a spectral analysis result output by the LightGBM. Therefore, the spectrum analysis method can perform spectrum analysis in an ensemble learning mode after batch effect of the spectrum data is removed, influence of the batch effect on the spectrum data is avoided, and due to the fact that the LightGBM ensemble learning model is adopted and is a parallel computing framework, time efficiency and space efficiency of an ensemble learning algorithm are improved.
Further, a third embodiment of the spectral analysis of the present invention is proposed based on the first and second embodiments of the spectral analysis method of the present invention.
In this embodiment, the spectral analysis method in the above embodiment is used to analyze each parameter of the white spirit, so as to test the accuracy of the spectral analysis result obtained by using the spectral analysis method of the present invention. For example, the alcohol content is one of the most critical indexes of white spirit products, in this embodiment, 55 bottles of white spirit produced by a certain brewery are used as samples to be analyzed, 9 times of repeated experiments are performed on each bottle of white spirit to measure the spectrum of the white spirit, 55 × 9=495 spectrums are obtained in total, the number of bands of each spectrum is 1024, and the size of the spectral feature matrix X obtained by measurement is 495 × 1024.
On the basis, the alcohol content of the white spirit is predicted by analyzing and processing the Raman spectrum of the white spirit. The proof of the sample was within the range of 29.2 to 29.6% vol, and the five-fold cross validation was performed for the regression prediction of the proof of white spirit. When the prediction error with respect to the alcoholic strength is within. + -. 0.1% vol, it is regarded as correct. Furthermore, the spectral analysis result of the invention is compared with three non-ensemble learning methods of SVM, KNN and decision tree and four ensemble learning methods of AdaBoost, random forest, XGBoost and LightGBM, and the prediction accuracy of the different methods is shown in fig. 4. Therefore, when the alcohol degree of the white spirit is predicted, the integrated learning method is generally superior to the non-integrated learning method, the spectral analysis method disclosed by the invention is superior to other methods, and the accuracy of over 80% is achieved.
In addition, the total amount of acid ester in the white spirit is predicted by Raman spectrum of the white spirit. The total amount of acid ester is an important index of the white spirit product, and has obvious influence on the flavor of the white spirit. In this embodiment, the total amount of acid ester in the white spirit is predicted by using the white spirit sample and the spectrum data thereof. The total amount of acid ester in the sample was in the range of 19.9-21.7 mmol/L. During the regression prediction of the total amount of acid ester in white spirit, five-fold cross validation is adopted. When the prediction error of the total amount of the acid ester is within +/-0.3 mmol/L, the prediction is considered to be correct. Furthermore, the spectrum analysis result of the present invention is compared with three non-integrated learning methods, namely SVM (support vector machines), KNN (K-Nearest Neighbor algorithm) and decision tree, and four integrated learning methods, namely AdaBoost, random forest, boost (eXtreme Gradient boost) and Light Gradient boost Machine (distributed Gradient boost), so that the prediction accuracy of different methods is shown in fig. 4. Therefore, when the total amount of acid ester in the white spirit is predicted, the ensemble learning method is generally superior to the non-ensemble learning method, and the spectral analysis method disclosed by the invention is superior to other methods, so that the accuracy of about 73% is achieved.
Besides, the taste index of the white spirit is predicted through the Raman spectrum of the white spirit. The flavor of the white spirit needs to be measured through various taste indexes, wherein the fullness, sweetness and softness are three key taste indexes of the white spirit. In this embodiment, the taste index of the white spirit is predicted by using the white spirit sample and the spectrum data thereof. The quantitative result of the taste index is obtained by that an expert in a winery scores the liquor after tasting the liquor, the score is accurate to one decimal place, and the score is 5. The sample has a fullness of 3.8-4.3, a sweetness of 4.2-4.5 and a softness of 4.0-4.5. When the taste indexes of the white spirit are subjected to regression prediction, five-fold cross validation is adopted. When the prediction error of the taste index is within +/-0.1 minute, the prediction is considered to be correct. Furthermore, the spectral analysis result of the present invention is compared with three non-ensemble learning methods, i.e., SVM, KNN and decision tree, and four ensemble learning methods, i.e., adaBoost, random forest, XGBoost and LightGBM, to determine the prediction accuracy of different methods, as shown in fig. 4. Therefore, although the taste indexes are manually graded and have certain errors, the prediction accuracy is lower than the prediction result of the organic matter content of the white spirit, the spectral analysis method of the invention is still superior to other methods in the prediction of the three taste indexes.
Through the above description, it can be seen that the accuracy of the spectral analysis of the sample to be analyzed in the present invention is significantly higher than that in other manners, and the present invention can be generally applied to the spectral analysis scenarios of various substances.
In addition, an embodiment of the present invention further provides a spectrum analysis apparatus, and with reference to fig. 5, the spectrum analysis apparatus includes:
the building module 10 is configured to obtain spectral data of a sample to be analyzed, and build a corresponding adjacency matrix according to the spectral data;
a removal processing module 20, configured to remove batch errors of the spectral data based on the adjacency matrix;
and the ensemble learning module 30 is configured to perform ensemble learning on the spectrum data with batch errors removed to obtain a corresponding spectrum analysis result.
Further, the building block 10 includes:
the node determining unit is used for determining a similarity graph among the spectrums of the samples to be analyzed based on the spectrum characteristics and determining each node in the similarity graph;
the sorting unit is used for sorting the distances between the characteristic vectors corresponding to the nodes to obtain a sorting result;
and the constructing module is used for constructing a node set corresponding to each node based on the sequencing result and constructing an adjacency matrix based on the node sets.
Further, the building block 10 includes:
the first matrix determining unit is configured to determine a corresponding regularized laplacian matrix according to the adjacency matrix, so as to remove batch errors of the spectral data of the sample to be analyzed based on the regularized laplacian matrix.
Further, the removal processing module 20 includes:
the second matrix determining unit is used for determining an initial characteristic matrix corresponding to the spectral data of the sample to be analyzed through a preset optimization algorithm based on the regularized Laplace matrix corresponding to the adjacent matrix, wherein the initial characteristic matrix is a matrix for removing batch effect and noise influence;
the matrix acquisition unit is used for acquiring a corresponding initial batch effect factor matrix based on the initial characteristic matrix;
a removal processing unit comprising: and determining a target initial characteristic matrix and a target batch effect factor matrix based on the initial batch effect factor matrix so as to remove the batch errors of the spectral data.
Further, the removal processing unit includes:
the decomposition subunit is used for carrying out non-negative matrix decomposition on the initial batch of effect factor matrixes to obtain a first non-negative matrix and a second non-negative matrix which are different in size;
the determining subunit is used for acquiring a target batch effect factor matrix based on the first non-negative matrix and the second non-negative matrix and determining whether the target batch effect factor matrix meets a preset low-rank constraint;
the matrix determining subunit is used for determining a corresponding target characteristic matrix based on the target batch effect factor matrix;
and the removing subunit is used for removing the batch errors of the spectral data according to the target characteristic matrix and the target batch effect factor matrix.
Further, the ensemble learning module 30 includes:
the spectrum analysis unit is used for performing ensemble learning on the spectrum data with the batch errors removed through a preset ensemble learning model to obtain a corresponding spectrum analysis result, wherein the ensemble learning model comprises: lightweight gradient hoist LightGBM.
Further, the spectral analysis unit includes:
the spectrum analysis subunit is configured to input, to the LightGBM, spectrum data of a sample to be analyzed from which batch errors are removed, and obtain a spectrum analysis result output by the LightGBM, where the spectrum analysis result includes, when the sample to be analyzed is white spirit to be analyzed: one or more of alcohol content, total amount of acid ester, and taste index.
The development of the specific embodiment of the spectrum analysis system of the present invention is substantially the same as that of each embodiment of the spectrum analysis method described above, and is not described herein again.
Furthermore, an embodiment of the present invention further provides a computer-readable storage medium, on which a spectrum analysis program is stored, and the spectrum analysis program, when executed by a processor, implements the steps of the spectrum analysis method as described below.
The embodiments of the spectral analysis apparatus and the computer-readable storage medium of the present invention can refer to the embodiments of the spectral analysis method of the present invention, and are not described herein again.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or system. Without further limitation, an element defined by the phrases "comprising one of 8230; \8230;" 8230; "does not exclude the presence of additional like elements in a process, method, article, or system that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are only for description, and do not represent the advantages and disadvantages of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solution of the present invention essentially or contributing to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal device (such as a smart phone, a computer, a server, and other network devices) to execute the method according to the embodiments of the present invention.
The above description is only a preferred embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes, which are made by using the contents of the present specification and the accompanying drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (8)
1. A method of spectral analysis, the method comprising:
acquiring spectral data of a sample to be analyzed, and constructing a corresponding adjacency matrix according to the spectral data;
removing batch errors of the spectral data based on the adjacency matrix;
performing ensemble learning on the spectral data with batch errors removed to obtain a corresponding spectral analysis result;
the step of removing batch errors of the spectral data based on the adjacency matrix comprises:
determining an initial feature matrix corresponding to the spectral data of the sample to be analyzed through a preset optimization algorithm based on the regularized Laplace matrix corresponding to the adjacency matrix, wherein the initial feature matrix is a matrix for removing batch effect and noise influence;
acquiring a corresponding initial batch effect factor matrix based on the initial characteristic matrix;
determining a target initial characteristic matrix and a target batch effect factor matrix based on the initial batch effect factor matrix so as to remove batch errors of the spectral data;
the step of determining a target initial feature matrix and a target batch-wise effect factor matrix based on the initial batch-wise effect factor matrix to remove batch errors of the spectral data includes:
performing non-negative matrix decomposition on the initial batch of effect factor matrixes to obtain a first non-negative matrix and a second non-negative matrix which are different in size;
acquiring a target batch effect factor matrix based on the first non-negative matrix and the second non-negative matrix, and determining whether the target batch effect factor matrix meets a preset low-rank constraint;
if yes, determining a corresponding target characteristic matrix based on the target batch effect factor matrix;
and removing the batch errors of the spectral data according to the target characteristic matrix and the target batch effect factor matrix.
2. A method of spectral analysis according to claim 1, wherein said spectral data includes spectral features, and said step of constructing a corresponding adjacency matrix from said spectral data includes:
determining a similarity graph among the spectrums of the sample to be analyzed based on the spectrum characteristics, and determining each node in the similarity graph;
sorting the distances between the characteristic vectors corresponding to the nodes to obtain a sorting result;
and constructing a node set corresponding to each node based on the sequencing result, and constructing an adjacency matrix based on a plurality of node sets.
3. The method for spectral analysis according to claim 2, wherein after said step of constructing a set of nodes corresponding to said respective nodes based on said sorting result and constructing an adjacency matrix based on a plurality of said sets of nodes, further comprising:
and determining a corresponding regularized Laplace matrix according to the adjacency matrix, and removing batch errors of the spectral data of the sample to be analyzed based on the regularized Laplace matrix.
4. The method for spectral analysis according to claim 1, wherein the step of performing ensemble learning on the spectral data with the batch errors removed to obtain the corresponding spectral analysis result comprises:
the integrated learning method includes the steps that integrated learning is conducted on spectral data with batch errors removed through a preset integrated learning model to obtain a corresponding spectral analysis result, wherein the integrated learning model comprises the following steps: lightweight gradient hoist LightGBM.
5. The method for spectral analysis according to claim 4, wherein the step of performing ensemble learning on the spectral data with batch errors removed by a predetermined ensemble learning model to obtain the corresponding spectral analysis result comprises:
inputting the spectral data of the sample to be analyzed with the batch errors removed into the LightGBM, and obtaining a spectral analysis result output by the LightGBM, wherein when the sample to be analyzed is the white spirit to be analyzed, the spectral analysis result includes: one or more of alcohol content, total amount of acid ester, and taste index.
6. A spectroscopic analysis apparatus, wherein the spectroscopic analysis apparatus comprises:
the system comprises a construction module, a data acquisition module and a data processing module, wherein the construction module is used for acquiring spectral data of a sample to be analyzed and constructing a corresponding adjacency matrix according to the spectral data;
a removal processing module for removing batch errors of the spectral data based on the adjacency matrix;
the integrated learning module is used for performing integrated learning on the spectral data with batch errors removed to obtain a corresponding spectral analysis result;
the removal processing module includes:
the matrix determining unit is used for determining an initial characteristic matrix corresponding to the spectral data of the sample to be analyzed through a preset optimization algorithm based on the regularized Laplace matrix corresponding to the adjacency matrix, wherein the initial characteristic matrix is a matrix for removing batch effect and noise influence;
the matrix acquisition unit is used for acquiring a corresponding initial batch effect factor matrix based on the initial characteristic matrix;
a removal processing unit, configured to determine a target initial feature matrix and a target batch effect factor matrix based on the initial batch effect factor matrix, so as to perform removal processing on a batch error of the spectral data;
the removal processing unit includes:
the decomposition subunit is used for carrying out non-negative matrix decomposition on the initial batch of effect factor matrixes to obtain a first non-negative matrix and a second non-negative matrix which are different in size;
the determining subunit is used for acquiring a target batch effect factor matrix based on the first non-negative matrix and the second non-negative matrix and determining whether the target batch effect factor matrix meets a preset low-rank constraint;
the matrix determining subunit is used for determining a corresponding target characteristic matrix based on the target batch effect factor matrix;
and the removing subunit is used for removing the batch errors of the spectral data according to the target characteristic matrix and the target batch effect factor matrix.
7. A terminal device, characterized in that the terminal device comprises a memory, a processor and a spectral analysis program stored on the memory and executable on the processor, the spectral analysis program, when executed by the processor, implementing the steps of the spectral analysis method according to any one of claims 1-5.
8. A computer-readable storage medium, having a spectral analysis program stored thereon, which when executed by a processor implements the steps of the spectral analysis method according to any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211409528.5A CN115479905B (en) | 2022-11-11 | 2022-11-11 | Spectral analysis method, spectral analysis device, terminal equipment and medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211409528.5A CN115479905B (en) | 2022-11-11 | 2022-11-11 | Spectral analysis method, spectral analysis device, terminal equipment and medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115479905A CN115479905A (en) | 2022-12-16 |
| CN115479905B true CN115479905B (en) | 2023-03-14 |
Family
ID=84396654
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211409528.5A Active CN115479905B (en) | 2022-11-11 | 2022-11-11 | Spectral analysis method, spectral analysis device, terminal equipment and medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115479905B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118013258B (en) * | 2024-04-08 | 2024-06-14 | 山东星芭克生物科技有限公司 | Information acquisition method of intelligent water-soluble fertilizer production line |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109115692A (en) * | 2018-07-04 | 2019-01-01 | 北京格致同德科技有限公司 | A kind of spectral data analysis method and device |
| CN109508440A (en) * | 2018-11-28 | 2019-03-22 | 武汉轻工大学 | Construction method, device, equipment and the storage medium of spectrum analysis model |
| CN113160069A (en) * | 2021-02-26 | 2021-07-23 | 桂林电子科技大学 | Hyperspectral image denoising method based on image signal |
| CN114548282A (en) * | 2022-02-23 | 2022-05-27 | 杭州师范大学 | Sorting method and device based on fluorescence spectrum and readable storage medium |
| CN114611582A (en) * | 2022-02-16 | 2022-06-10 | 温州大学 | Method and system for analyzing substance concentration based on near infrared spectrum technology |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA3131843A1 (en) * | 2020-09-25 | 2022-03-25 | Royal Bank Of Canada | System and method for structure learning for graph neural networks |
-
2022
- 2022-11-11 CN CN202211409528.5A patent/CN115479905B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109115692A (en) * | 2018-07-04 | 2019-01-01 | 北京格致同德科技有限公司 | A kind of spectral data analysis method and device |
| CN109508440A (en) * | 2018-11-28 | 2019-03-22 | 武汉轻工大学 | Construction method, device, equipment and the storage medium of spectrum analysis model |
| CN113160069A (en) * | 2021-02-26 | 2021-07-23 | 桂林电子科技大学 | Hyperspectral image denoising method based on image signal |
| CN114611582A (en) * | 2022-02-16 | 2022-06-10 | 温州大学 | Method and system for analyzing substance concentration based on near infrared spectrum technology |
| CN114548282A (en) * | 2022-02-23 | 2022-05-27 | 杭州师范大学 | Sorting method and device based on fluorescence spectrum and readable storage medium |
Non-Patent Citations (1)
| Title |
|---|
| 基于集成学习的酒样光谱识别的研究;李大同;《中国优秀硕士学位论文全文数据库 工程科技Ⅰ辑》;20200115(第1期);摘要、第7-49页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115479905A (en) | 2022-12-16 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11244151B2 (en) | Computer-implemented method of recognizing facial expression, apparatus for recognizing facial expression, method of pre-training apparatus for recognizing facial expression, computer-program product for recognizing facial expression | |
| CN105335756B (en) | A kind of image classification method and image classification system based on Robust Learning model | |
| CN109829481B (en) | Image classification method and device, electronic equipment and readable storage medium | |
| CN115479905B (en) | Spectral analysis method, spectral analysis device, terminal equipment and medium | |
| CN116361801B (en) | Malicious software detection method and system based on semantic information of application program interface | |
| CN104820841B (en) | Hyperspectral classification method based on low order mutual information and spectrum context waveband selection | |
| WO2019195971A1 (en) | Spectral analysis method, apparatus, electronic device, and computer-readable storage medium | |
| CN114943674A (en) | Defect detection method, electronic device and storage medium | |
| WO2023096708A1 (en) | System and method for balancing sparsity in weights for accelerating deep neural networks | |
| CN113822207A (en) | Hyperspectral remote sensing image identification method and device, electronic equipment and storage medium | |
| CN115357699A (en) | Text extraction method, device, equipment and storage medium | |
| Liu et al. | Development of a mobile application for identification of grapevine (Vitis vinifera L.) cultivars via deep learning | |
| CN115687980A (en) | Desensitization classification method of data table, and classification model training method and device | |
| CN114398871A (en) | Financial entity extraction method, device, equipment and computer readable storage medium | |
| CN113160886A (en) | Cell type prediction system based on single cell Hi-C data | |
| US20200279148A1 (en) | Material structure analysis method and material structure analyzer | |
| CN117607120A (en) | Food additive Raman spectrum detection method and device based on improved Resnext model | |
| CN109253981B (en) | Method and device for establishing quantitative analysis model based on infrared spectrum | |
| CN111860601A (en) | Method and device for predicting large fungus species | |
| CN115563985A (en) | Statement analysis method, statement analysis device, statement analysis apparatus, storage medium, and program product | |
| CN116680401A (en) | Document processing method, document processing device, apparatus and storage medium | |
| CN115541531A (en) | Method for predicting protein content in feed based on two-dimensional correlation spectrum | |
| Li et al. | Y-SPCR: A new dimensionality reduction method for gene expression data classification | |
| WO2023220891A1 (en) | Resolution-switchable segmentation networks | |
| CN116429709B (en) | Spectrum detection method, spectrum detection device and computer-readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |