CN116429709A - Spectrum detection method, spectrum detection device and computer-readable storage medium - Google Patents
Spectrum detection method, spectrum detection device and computer-readable storage medium Download PDFInfo
- Publication number
- CN116429709A CN116429709A CN202310680666.5A CN202310680666A CN116429709A CN 116429709 A CN116429709 A CN 116429709A CN 202310680666 A CN202310680666 A CN 202310680666A CN 116429709 A CN116429709 A CN 116429709A
- Authority
- CN
- China
- Prior art keywords
- spectrum
- matrix
- spectral
- detection model
- feature matrix
- 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.)
- Granted
Links
- 238000001228 spectrum Methods 0.000 title claims abstract description 430
- 238000001514 detection method Methods 0.000 title claims abstract description 166
- 238000003860 storage Methods 0.000 title claims abstract description 13
- 239000011159 matrix material Substances 0.000 claims abstract description 258
- 238000012549 training Methods 0.000 claims abstract description 54
- 238000000034 method Methods 0.000 claims abstract description 38
- 230000003595 spectral effect Effects 0.000 claims description 133
- 239000013598 vector Substances 0.000 claims description 100
- 238000004422 calculation algorithm Methods 0.000 claims description 12
- 238000005457 optimization Methods 0.000 abstract description 4
- 235000014101 wine Nutrition 0.000 description 13
- 238000004891 communication Methods 0.000 description 6
- 238000007637 random forest analysis Methods 0.000 description 5
- 238000012360 testing method Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- LFQSCWFLJHTTHZ-UHFFFAOYSA-N Ethanol Chemical compound CCO LFQSCWFLJHTTHZ-UHFFFAOYSA-N 0.000 description 2
- 230000001476 alcoholic effect Effects 0.000 description 2
- 238000002790 cross-validation Methods 0.000 description 2
- 238000003066 decision tree Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000010183 spectrum analysis Methods 0.000 description 2
- 235000020097 white wine Nutrition 0.000 description 2
- 238000001237 Raman spectrum Methods 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013145 classification model Methods 0.000 description 1
- 230000008602 contraction Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000007786 learning performance Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000004611 spectroscopical analysis Methods 0.000 description 1
- 239000000126 substance Substances 0.000 description 1
- 238000012706 support-vector machine Methods 0.000 description 1
- 230000008719 thickening Effects 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
- G01—MEASURING; TESTING
- G01J—MEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
- G01J3/00—Spectrometry; Spectrophotometry; Monochromators; Measuring colours
- G01J3/28—Investigating the spectrum
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01J—MEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
- G01J3/00—Spectrometry; Spectrophotometry; Monochromators; Measuring colours
- G01J3/28—Investigating the spectrum
- G01J2003/283—Investigating the spectrum computer-interfaced
- G01J2003/2843—Processing for eliminating interfering spectra
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Analytical Chemistry (AREA)
- Pathology (AREA)
- Immunology (AREA)
- General Health & Medical Sciences (AREA)
- Biochemistry (AREA)
- Chemical & Material Sciences (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a spectrum detection method, a device and a computer readable storage medium, wherein the method comprises the following steps: acquiring a spectrum first angle similarity matrix corresponding to a spectrum characteristic matrix, and acquiring a regularized Laplacian matrix corresponding to the spectrum first angle similarity matrix; determining a spectrum detection model based on the spectrum characteristic matrix, the regularized Laplace matrix, the auxiliary variable and the sparse constraint noise, wherein the output of the spectrum detection model is a denoising spectrum characteristic matrix; inputting the spectrum training data into the spectrum detection model for iterative training; and if the trained spectrum detection model converges, taking the trained spectrum detection model as a target spectrum detection model, and taking the denoising spectrum feature matrix of the current iteration as a target denoising spectrum feature matrix. According to the invention, the auxiliary variables are introduced to perform alternate iterative optimization on a plurality of constraints so as to extract effective information from spectrum data, and noise in a spectrum can be accurately filtered.
Description
Technical Field
The present invention relates to the field of product detection technologies, and in particular, to a spectrum detection method, a spectrum detection device, and a computer readable storage medium.
Background
Spectroscopic analysis is an effective means of substance identification. Compared with the traditional spectrum analysis method, the machine learning algorithm is applied to analyze the spectrum, so that the cost can be remarkably reduced, and the efficiency can be improved. However, the spectrum is also a typical high-dimensional data, which is rich in a lot of noise. Noise in the spectrum is filtered to obtain effective representation (presentation), which is also an element for improving the performance of the model. This relies on optimizing low-rank constraints (low-rank constraints) and manifold constraints (manifold constraint) in the spectral feature matrix. The existing combined optimization low-rank manifold constraint is mainly GNMF (Graph-regularized non-negative matrix factorization, graph regularized non-negative matrix factorization), which is sensitive to iteration initial values, has low convergence speed and low expandability, lacks adaptive learning capability for complex data, has limited capability of extracting effective information from spectrum data, and cannot accurately filter noise in a spectrum.
The foregoing is provided merely for the purpose of facilitating understanding of the technical solutions of the present invention and is not intended to represent an admission that the foregoing is prior art.
Disclosure of Invention
The invention mainly aims to provide a spectrum detection method, a spectrum detection device and a computer readable storage medium, and aims to solve the technical problem that the existing spectrum analysis is difficult to accurately filter noise in a spectrum.
In order to achieve the above object, the present invention provides a spectrum detection method, including the steps of:
acquiring a spectrum characteristic matrix corresponding to spectrum training data, acquiring a spectrum first angle similarity matrix corresponding to the spectrum characteristic matrix, and acquiring a regularized Laplacian matrix corresponding to the spectrum first angle similarity matrix;
determining a spectrum detection model based on the spectrum characteristic matrix, the regularized Laplace matrix, the auxiliary variable and the sparse constraint noise, wherein the output of the spectrum detection model is a denoising spectrum characteristic matrix;
inputting the spectrum training data into the spectrum detection model for iterative training to obtain a denoising spectrum feature matrix corresponding to the spectrum training data and a trained spectrum detection model;
if the convergence of the trained spectrum detection model is determined based on the denoising spectrum feature matrix of the current iteration and the denoising spectrum feature matrix of the last iteration, the trained spectrum detection model is taken as a target spectrum detection model, and the denoising spectrum feature matrix of the current iteration is taken as a target denoising spectrum feature matrix.
Further, the step of determining a spectrum detection model based on the spectrum feature matrix, the regularized laplace matrix, the auxiliary variable and the sparse constraint noise, wherein the output of the spectrum detection model is a denoising spectrum feature matrix comprises the following steps:
Determining the spectral feature matrix, the denoising spectral feature matrix and the sparse constraint noise, and determining a first F-norm;
determining a trace of a matrix based on the regularized laplace matrix and the spectral feature matrix;
determining a kernel norm corresponding to the auxiliary variable, and determining a 1-norm based on the sparse constraint noise;
a spectral detection model is determined based on the first F-norm, the trace of the matrix, the kernel norm, and the 1-norm.
Further, the step of inputting the spectrum training data into the spectrum detection model for iterative training to obtain a denoising spectrum feature matrix corresponding to the spectrum training data and a trained spectrum detection model includes:
for each iteration training, calculating a denoising spectral feature matrix of the current iteration corresponding to the spectral feature matrix according to a linear least square method through a spectral detection model;
calculating an auxiliary variable of the current iteration according to the singular value thresholding shrinkage operator through a spectrum detection model;
and calculating sparse constraint noise of the current iteration according to an iterative shrinkage threshold algorithm through a spectrum detection model.
Further, if it is determined that the trained spectrum detection model converges based on the denoising spectrum feature matrix of the current iteration and the denoising spectrum feature matrix of the last iteration, the step of using the trained spectrum detection model as the target spectrum detection model includes:
Determining a second F-norm based on the denoising spectral feature matrix of the current iteration and the denoising spectral feature matrix of the last iteration;
if the second F-norm is smaller than the preset error value, determining that the trained spectrum detection model converges, and taking the trained spectrum detection model as a target spectrum detection model.
Further, the step of obtaining the spectral feature matrix corresponding to the spectral training data, obtaining the first angular similarity matrix of the spectrum corresponding to the spectral feature matrix, and obtaining the regularized laplace matrix corresponding to the first angular similarity matrix of the spectrum includes:
acquiring first angle similarity between every two spectral vectors in the spectral feature matrix, and determining a spectral first angle similarity matrix based on the first angle similarity;
determining a degree matrix based on the spectrum first angular similarity matrix, wherein the degree matrix is a diagonal matrix;
based on the degree matrix, the regularized laplacian matrix is determined.
Further, if it is determined that the trained spectrum detection model converges based on the denoising spectrum feature matrix of the current iteration and the denoising spectrum feature matrix of the last iteration, the step of using the trained spectrum detection model as the target spectrum detection model further includes:
Inputting the denoising spectral feature matrix of the current iteration into an integrated learning model for prediction to obtain a prediction result corresponding to each spectrum in the spectral training data;
determining the true rate and the false positive rate corresponding to each preset threshold value based on the preset threshold values and the prediction result;
based on the true rate and the false positive rate, a ROC curve is determined, and a true rate threshold and a false positive rate threshold are determined based on the ROC curve.
Further, after the step of determining the ROC curve based on the true rate and the false positive rate and determining the true rate threshold and the false positive rate threshold based on the ROC curve, the method further includes:
the method comprises the steps of obtaining spectrum prediction data to be predicted corresponding to spectrum training data, and inputting a spectrum feature matrix to be predicted in the spectrum prediction data to be predicted into the target spectrum detection model for prediction so as to obtain a target denoising spectrum feature matrix;
determining a plurality of target spectrum vectors corresponding to each second spectrum vector in the first spectrum vector, wherein the first spectrum vector is the spectrum vector in the denoising spectrum feature matrix of the current iteration, and the second spectrum vector is the spectrum vector in the denoising spectrum feature matrix of the target;
based on the prediction result of the target spectrum vector, determining a prediction value corresponding to each second spectrum vector;
And determining classification results of the products corresponding to the spectral feature matrix to be predicted based on the predicted value, the true rate threshold and the false positive rate threshold.
Further, the step of determining a plurality of target spectral vectors corresponding to the respective second spectral vectors in the first spectral vectors includes:
obtaining second angle similarity between each first spectrum vector and each second spectrum vector;
and determining a plurality of target spectrum vectors corresponding to the second spectrum vectors in the first spectrum vectors based on the second angular similarity.
In addition, to achieve the above object, the present invention also provides a spectrum detection apparatus including: a memory, a processor and a spectrum sensing program stored on the memory and executable on the processor, which when executed by the processor, performs the steps of the spectrum sensing method of any of the preceding claims.
In addition, in order to achieve the above object, the present invention also provides a computer-readable storage medium having stored thereon a spectrum detection program which, when executed by a processor, implements the steps of the spectrum detection method of any one of the foregoing.
According to the method, a first spectrum angle similarity matrix corresponding to a spectrum characteristic matrix in spectrum training data is obtained, and a regularized Laplacian matrix corresponding to the first spectrum angle similarity matrix is obtained; then, a spectrum detection model is determined based on the spectrum feature matrix, the regularized Laplace matrix, the auxiliary variable and the sparse constraint noise, wherein the output of the spectrum detection model is a denoising spectrum feature matrix; then inputting the spectrum training data into the spectrum detection model for iterative training to obtain a denoising spectrum feature matrix corresponding to the spectrum training data and a trained spectrum detection model; and then if the convergence of the trained spectrum detection model is determined based on the denoising spectrum feature matrix of the current iteration and the denoising spectrum feature matrix of the last iteration, taking the trained spectrum detection model as a target spectrum detection model, taking the denoising spectrum feature matrix of the current iteration as the target denoising spectrum feature matrix, and carrying out alternate iteration optimization on a plurality of constraints by introducing auxiliary variables so as to extract effective information from spectrum data, so that noise in a spectrum can be accurately filtered.
Drawings
FIG. 1 is a schematic diagram of a spectrum detection device in a hardware operating environment according to an embodiment of the present invention;
FIG. 2 is a flow chart of a first embodiment of the spectrum detection method of the present invention;
FIG. 3 is a schematic diagram of a spectrum detection method according to an embodiment of the present invention;
fig. 4 is a schematic view of a scene of another embodiment of the spectrum detection method of the present invention.
The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Referring to fig. 1, fig. 1 is a schematic structural diagram of a spectrum detection device in a hardware operation environment according to an embodiment of the present invention.
The spectrum detection device of the embodiment of the invention can be a PC, a mobile terminal device with a display function such as a smart phone and a tablet personal computer. As shown in fig. 1, the spectrum detection apparatus 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 the communication bus 1002 is used to enable connected communication between these components. The user interface 1003 may include a Display, an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may further 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 stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also optionally be a storage device separate from the processor 1001 described above.
Optionally, the spectrum detection device may further include a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, and the like. Among other sensors, such as light sensors, motion sensors, and other sensors. Of course, the spectrum detection device may also be configured with other sensors such as a gyroscope, a barometer, a hygrometer, a thermometer, an infrared sensor, and the like, which are not described herein.
It will be appreciated by those skilled in the art that the terminal structure shown in fig. 1 is not limiting of the optical detection device and may include more or fewer components than shown, or may combine certain components, or a different arrangement of components.
As shown in fig. 1, an operating system, a network communication module, a user interface module, and a spectrum sensing program may be included in the memory 1005, which is a type of computer storage medium.
In the spectrum detection apparatus shown in fig. 1, the network interface 1004 is mainly used for connecting to a background server, and performing data communication with the background server; the user interface 1003 is mainly used for connecting a client (user side) and performing data communication with the client; and the processor 1001 may be used to invoke the spectrum sensing program stored in the memory 1005.
In this embodiment, the spectrum detection apparatus includes: the spectrum sensing device comprises a memory 1005, a processor 1001 and a spectrum sensing program which is stored in the memory 1005 and can be run on the processor 1001, wherein the processor 1001 executes the steps of the spectrum sensing method in the following embodiments when calling the spectrum sensing program stored in the memory 1005.
The invention also provides a spectrum detection method, referring to fig. 2, fig. 2 is a schematic flow chart of a first embodiment of the spectrum detection method of the invention.
In this embodiment, the spectrum detection method includes:
step S101, acquiring a spectrum feature matrix corresponding to spectrum training data, acquiring a spectrum first angle similarity matrix corresponding to the spectrum feature matrix, and acquiring a regularized Laplacian matrix corresponding to the spectrum first angle similarity matrix;
in this embodiment, the spectrum training data may be n spectrums, and the spectrum band degree is m, and then the size of the spectrum feature matrix X is an nxm matrix, that is, the spectrum feature matrix is a matrix formed by n spectrum vectors, where the n spectrum vectors are feature vectors corresponding to each of the n spectrums.
After the spectral feature matrix is obtained, a first angular similarity matrix of a spectrum corresponding to the spectral feature matrix is obtained, and a regularized laplace matrix corresponding to the first angular similarity matrix of the spectrum is obtained, and in a specific embodiment, the step S101 includes:
Step S1011, obtaining first angle similarity between every two spectrum vectors in the spectrum characteristic matrix, and determining a spectrum first angle similarity matrix based on the first angle similarity;
step S1012, determining a degree matrix based on the spectrum first angular similarity matrix, wherein the degree matrix is a diagonal matrix;
step S1013, determining the regularized laplacian matrix based on the degree matrix.
In this embodiment, after the spectral feature matrix is obtained, a first angular similarity between each two spectral vectors in the spectral feature matrix is obtained, and a spectral first angular similarity matrix S and matrix elements S are determined based on the first angular similarity ij Representing two spectral vectors x i 、x j The specific formula of the cosine value of the spectrum angle is:
wherein S is ij Representing two spectral vectors x i 、x j And θ is the cosine of the spectral angle of two spectral vectors x i 、x j Spectral angle, x i 、x j Two spectral vectors are respectively provided, wherein in a first angular similarity matrix S of the spectrum, when i=j, S ij =1。
Then, obtaining a first angle similarity matrix of the spectrum, and determining a degree matrix D, wherein the degree matrix is a diagonal matrix, and the elements of the degree matrix D。
And then, based on the degree matrix, determining the regularized Laplace matrix L, and further accurately obtaining the regularized Laplace matrix L, wherein the regularized Laplace matrix L has the formula:
Wherein L is regularized Laplacian matrix, D is degree matrix, and I is identity matrix.
Step S102, a spectrum detection model is determined based on the spectrum feature matrix, the regularized Laplace matrix, the auxiliary variable and the sparse constraint noise, wherein the output of the spectrum detection model is a denoising spectrum feature matrix;
in this embodiment, after the spectral feature matrix and the regularized laplace matrix are obtained, a model function of a spectral detection model is determined according to the spectral feature matrix, the regularized laplace matrix, the auxiliary variable and the sparse constraint noise, where the output of the spectral detection model is a denoising spectral feature matrix, and specifically, in a possible implementation manner, the step S102 includes:
step S1021, determining the spectral feature matrix, the denoising spectral feature matrix and the sparse constraint noise, and determining a first F-norm;
step S1022, determining the trace of the matrix based on the regularized Laplace matrix and the spectral feature matrix;
step S1023, determining a kernel norm corresponding to the auxiliary variable, and determining a 1-norm based on the sparse constraint noise;
step S1024, determining a spectrum detection model based on the first F-norm, the trace of the matrix, the kernel norm and the 1-norm.
In this embodiment, a spectral feature matrix, a denoising spectral feature matrix, and sparse constraint noise are determined, and a first F-norm is determined, where the formula of the first F-norm is: i X-U-E I F Wherein X is a spectral feature matrix, U is a denoising spectral feature matrix, and E is sparse constraint noise.
Determining the trace of the matrix based on the regularized Laplace matrix and the spectral feature matrix, wherein the formula of the trace of the determined matrix is as follows: tr (U) T LU), where U is the denoised spectral feature matrix and L is the regularized laplace matrix.
Determining a kernel norm corresponding to the auxiliary variable, determining a 1-norm based on the sparse constraint noise, and determining a model function of a spectrum detection model based on the first F-norm, the trace of the matrix, the kernel norm and the 1-norm, wherein the model function has a formula as follows:
where X is a spectral feature matrix, U is a denoising spectral feature matrix, E is sparse constraint noise, L is a regularized laplace matrix, V is an auxiliary variable, μ, λ, η are all hyper-parameters, preferably μ=λ=η=1.
Step S103, inputting the spectrum training data into the spectrum detection model for iterative training to obtain a denoising spectrum feature matrix corresponding to the spectrum training data and a trained spectrum detection model;
In this embodiment, after obtaining the spectrum detection model, the spectrum training data is input into the spectrum detection model to perform iterative training to obtain a denoising spectrum feature matrix corresponding to the spectrum training data and a trained spectrum detection model, that is, the spectrum training data is input into the spectrum detection model to perform iterative solution, and U, V, E can be obtained by each iterative solution, specifically, in one possible implementation manner, the step S103 includes:
step S1031, for each iteration training, calculating a denoising spectral feature matrix of the current iteration corresponding to the spectral feature matrix according to a linear least square method through a spectral detection model;
s1032, calculating an auxiliary variable of the current iteration according to the singular value thresholding shrinkage operator through the spectrum detection model;
step S1033, calculating sparse constraint noise of the current iteration according to an iterative shrinkage threshold algorithm through a spectrum detection model.
In this example, ADMM (Alternating Direction Method of Multipliers, alternate direction multiplier method) was introduced into ALM (Augmented Lagrangian Multiplier ), denoted as W, according to … U k →V k →E k →W k →U k+1 …, an iterative sequence is solved U, V, E.
For each (k+1th) iterative training, calculating a denoising spectral feature matrix U of the current iteration corresponding to the spectral feature matrix according to a linear least square method through a spectral detection model k+1 Using partial derivativesReduce to about U k Linear constraint of (2)Calculation of U according to the linear least squares method k+1 。
Calculating an auxiliary variable V of the current iteration according to SVT (shrinkage operator of singular value threshold, singular value thresholding contraction operator) through a spectrum detection model k+1 The sparse constraint noise E of the current iteration is calculated through ISTA (iterative shrinkage-thresholding algorithms, iterative shrinkage threshold algorithm) k+1 。
Step S104, if it is determined that the trained spectrum detection model converges based on the denoising spectrum feature matrix of the current iteration and the denoising spectrum feature matrix of the last iteration, the trained spectrum detection model is used as a target spectrum detection model, and the denoising spectrum feature matrix of the current iteration is used as a target denoising spectrum feature matrix.
In this embodiment, when the denoising spectral feature matrix of the current iteration is obtained, the denoising spectral feature matrix of the last iteration is obtained, whether the trained spectral detection model is converged is determined according to the denoising spectral feature matrix of the current iteration and the denoising spectral feature matrix of the last iteration, if so, the trained spectral detection model is used as the target spectral detection model, the denoising spectral feature matrix of the current iteration is used as the target denoising spectral feature matrix (the feature matrix of the low rank-manifold constrained spectrum), if not, the trained spectral detection model is used as the spectral detection model, and the step S103 is executed in a return manner.
Further, in one possible implementation manner, the step S104 includes:
step S1041, determining a second F-norm based on the denoising spectral feature matrix of the current iteration and the denoising spectral feature matrix of the last iteration;
in step S1042, if the second F-norm is smaller than the preset error value, the trained spectrum detection model is determined to converge, and the trained spectrum detection model is used as the target spectrum detection model.
In this embodiment, when the denoising spectral feature matrix of the current iteration is obtained, the denoising spectral feature matrix of the current iteration and the denoising spectral feature matrix of the previous iteration are based on the denoising spectral feature matrix of the current iteration, so as to determineDetermining a second F-norm, wherein the formula of the second F-norm is as follows: u is U k+1 -U k || F And determining whether the second F-norm is less than a preset error value.
If the second F-norm is smaller than the preset error value, determining that the trained spectrum detection model converges, taking the trained spectrum detection model as the target spectrum detection model, wherein the preset error value e=0.001, if the second F-norm is greater than or equal to the preset error value, determining that the trained spectrum detection model does not converge, taking the trained spectrum detection model as the spectrum detection model, and returning to execute step S103
According to the spectrum detection method, a first spectrum angle similarity matrix corresponding to a spectrum feature matrix in spectrum training data is obtained, and a regularized Laplacian matrix corresponding to the first spectrum angle similarity matrix is obtained; then, a spectrum detection model is determined based on the spectrum feature matrix, the regularized Laplace matrix, the auxiliary variable and the sparse constraint noise, wherein the output of the spectrum detection model is a denoising spectrum feature matrix; then inputting the spectrum training data into the spectrum detection model for iterative training to obtain a denoising spectrum feature matrix corresponding to the spectrum training data and a trained spectrum detection model; and then if the convergence of the trained spectrum detection model is determined based on the denoising spectrum feature matrix of the current iteration and the denoising spectrum feature matrix of the last iteration, taking the trained spectrum detection model as a target spectrum detection model, taking the denoising spectrum feature matrix of the current iteration as the target denoising spectrum feature matrix, and carrying out alternate iteration optimization on a plurality of constraints by introducing auxiliary variables so as to extract effective information from spectrum data, so that noise in a spectrum can be accurately filtered.
Based on the first embodiment, a second embodiment of the spectrum detection method of the present invention is proposed, in this embodiment, after step S104, the spectrum detection method further includes:
step S201, the denoising spectral feature matrix of the current iteration is input into an integrated learning model for prediction so as to obtain a prediction result corresponding to each spectrum in the spectral training data;
step S202, determining the true rate and the false positive rate corresponding to each preset threshold based on a plurality of preset thresholds and a prediction result;
step S203, determining an ROC curve based on the true rate and the false positive rate, and determining a true rate threshold and a false positive rate threshold based on the ROC curve.
The ensemble learning (ensemble learning) is to integrate the results of multiple base learners (typically decision trees) to improve learning performance. Representative ensemble learning algorithms are RF (random forest), XGBoost (extreme gradient boosting tree, limit gradient lift tree), lightGBM (light gradient boosting machine, lightweight gradient lift), etc. The integrated learning can adaptively select multidimensional features and multi-classifiers, so that unbalanced classification problems and regression problems can be better processed. The ensemble learning model in this embodiment may be a model employing any of RF, XGBoos, and LightGBM algorithms
In this embodiment, after the target spectrum detection model is obtained, the denoising spectral feature matrix of the current iteration is input into the ensemble learning model to predict, so as to obtain a prediction result corresponding to each spectrum in the spectrum training data, that is, the output of the ensemble learning model is the prediction result corresponding to each spectrum in the spectrum training data, and the prediction result of each spectrum is a prediction probability of a (0, 1) interval, where the ensemble learning model sets the maximum depth of the tree to be 5, and the iteration number to be 50, and can determine the classification result corresponding to each spectrum in the spectrum training data according to the prediction result.
After the prediction results are obtained, determining the TPR (true positive rate, true rate) and the FPR (false positive rate ) corresponding to each preset threshold based on a plurality of preset thresholds and the prediction results, specifically, classifying products corresponding to each spectrum through the prediction results for each preset threshold, classifying the products corresponding to the prediction results into positive classes if the prediction results are larger than the preset thresholds, otherwise, classifying the products into negative classes, obtaining the true classifications of the products corresponding to each spectrum vector of the spectrum feature matrix, classifying the classification results and the true classifications of the products corresponding to each spectrum according to the prediction results, and calculating the true rate and the false positive rate corresponding to the preset thresholds to obtain the true rate and the false positive rate corresponding to the preset thresholds.
Then, based on the true rate and the false positive rate, determining an ROC curve, and determining a true rate threshold and a false positive rate threshold based on the ROC curve, specifically, as shown in fig. 3, connecting the TPR and the FPR under different preset thresholds in a coordinate system of the TPR and the FPR to obtain an ROC curve (Receiver operating characteristic curve), wherein the area enclosed by the ROC curve is called AUROC or AUC (area under ROC curve), the larger the area is, the better the learning effect is, and the model is also shown to reduce the false detection rate and the omission rate. The TPR and FPR corresponding to the intersection of the line tpr+fpr=1 and the ROC curve are taken as the true rate threshold and false positive rate threshold.
In this embodiment, the classification result corresponding to each spectrum in the spectrum training data may be directly determined according to the prediction result. The product corresponding to the spectrum training data is a white wine sample, and the qualified product comprises 90 bottles of wine, and each bottle of wine has 6 spectrums, and the total number of the spectrums is 540. The unqualified products comprise 5 bottles of wine, and each bottle of wine has 12 spectra and 60 spectra. Five-fold cross validation was used, with one bottle of wine (12 spectra) in each fold of test set failed. The Raman spectrum has 1024 wave bands and ranges from 160 cm to 4080cm -1 Interval. The spectral feature matrix corresponding to the test set and the target denoising spectral feature matrix corresponding to the test set are input into the integrated learning to obtain two prediction results, namely, the prediction result of the spectral feature matrix and the prediction result of the target denoising spectral feature matrix, a first classification result corresponding to the test set and a second classification result of the target denoising spectral feature matrix are obtained according to the two prediction results, the AUC, the false detection rate (MDR), the False Detection Rate (FDR), the accuracy (Acc), the precision (Pre), the F1 score and the Mcc (Matthews correlation coefficient ) are calculated according to the first classification result and the second classification result respectively, evaluation is carried out to obtain the table 1 and the table 2,
TABLE 1 prediction results for input of ensemble learning model using spectral feature matrix
TABLE 2 prediction results into an ensemble learning model using a target denoising spectral feature matrix
The SVM is a support vector machine, KNN is an adjacent algorithm, RF is a random forest, XGB is a limit gradient lifting tree XGBoost, LGB is a lightweight gradient lifting machine LightGBM, and the thickening part is the best result.
In tables 1 and 2, SVM and KNN have high false detection rate despite their high accuracy and low false detection rate, indicating that these two algorithms predict almost all reject products as acceptable, whereas the ensemble learning model does not. After the target denoising spectral feature matrix is used for representing the learning model, the performance of the integrated learning model is remarkably improved. In the problem of white spirit qualified product detection, a target denoising spectral feature matrix representation learning model and a random forest, XGBoost, lightGBM and other integrated learning algorithm are fused, and excellent performance is obtained in all evaluation indexes. Because the same batch of white wine can be subjected to parallel detection of multiple sampling, the existing model can efficiently distinguish qualified products from unqualified products.
In actual production, white spirit may be disqualified because its alcoholicity is out of the specified range or because its formula is disqualified. In order to further check whether the spectrum representation learning model based on low rank-manifold constraint and the random forest, XGBoost, lightGBM and other integrated learning algorithms are fused, the unqualified types of white spirit can be effectively distinguished. In the white spirit samples used, the method comprises the following steps: 100 bottles of qualified wine, each bottle of wine has 6 spectra, and 600 spectra are total; 5 bottles of unqualified wine (3 bottles are unqualified in formula and 2 bottles are unqualified in alcohol degree) are provided by a winery, and each bottle of wine has 6 spectra, namely 30 spectra; 10 bottles of wine are prepared as unqualified products of the formula, each bottle of wine has 12 spectra, and 120 spectra are taken; 25 bottles of wine which are self-prepared unqualified products with alcoholic strength, wherein each bottle of wine has 12 spectra, and 300 spectra in total; the above, the reject includes 450 spectra in total.
Five-fold cross validation is adopted, and each folded test set comprises 20 bottles of qualified products and 120 spectra in total; total 90 spectra of reject: 1 bottle of unqualified products (6 spectrums) provided by a winery, 2 bottles of self-prepared formula unqualified products (24 spectrums), and 5 bottles of self-prepared alcoholic strength unqualified products (60 spectrums).
The unqualified type detection is realized by two-stage classification, the qualified or unqualified classification is firstly carried out on the sample, then the unqualified sample is carried out on the unqualified sample, the unqualified sample is subjected to the formula or alcohol degree is respectively classified, and the decision tree is shown in figure 4.
In addition to accuracy (Acc), the accuracy (precision) defined on the classification problem, recall (i.e., sensitivity) and F1 score can be extended to multi-classification tasks. Assume that the number of samples of the i-th type on the dataset is N i The accuracy when the sample is taken as a positive sample and the other samples are taken as negative samples is P i The weighted accuracy (weighted precision, WP) is:
similarly, a Weighted Recall (WR) may be defined, weighting the F1 score (WF 1). Note that the definition of WR is equivalent to accuracy, and thus is evaluated using three indices, accuracy, WP, WF 1.
The embodiment compares the performance of the SVM and KNN non-integrated learning methods and the random forest and XGBoost, lightGBM integrated learning methods on the problem of unqualified detection type. As shown in table 3:
TABLE 3 Table 3
According to table 3, lightgbm achieves 88% accuracy.
According to the spectrum detection method provided by the embodiment, the denoising spectrum feature matrix of the current iteration is input into the integrated learning model for prediction, so that a prediction result corresponding to each spectrum in spectrum training data is obtained; then, based on a plurality of preset thresholds and prediction results, determining the true rate and the false positive rate corresponding to each preset threshold; and then determining an ROC curve based on the real rate and the false positive rate, determining a real rate threshold value and a false positive rate threshold value based on the ROC curve, and obtaining a prediction result of each spectrum by performing integrated learning on the trained denoising spectrum feature matrix, and determining the real rate threshold value and the false positive rate threshold value according to the prediction result so as to predict the spectrum to be predicted through the denoising spectrum feature matrix, the real rate threshold value and the false positive rate threshold value of the current iteration in the follow-up process, thereby effectively solving the unbalanced classification problem.
Based on the second embodiment, a third embodiment of the spectrum detection method of the present invention is proposed, in this embodiment, after step S203, the spectrum detection method further includes:
step S101, spectrum prediction data to be predicted corresponding to spectrum training data are obtained, and a spectrum feature matrix to be predicted in the spectrum prediction data to be predicted is input into the target spectrum detection model for prediction so as to obtain a target denoising spectrum feature matrix;
step S302, determining a plurality of target spectrum vectors corresponding to each second spectrum vector in the first spectrum vectors, wherein the first spectrum vector is the spectrum vector in the denoising spectrum feature matrix of the current iteration, and the second spectrum vector is the spectrum vector in the denoising spectrum feature matrix of the target;
step S303, based on the prediction result of the target spectrum vector, determining the prediction value corresponding to each second spectrum vector;
and step S304, determining classification results of the products corresponding to the spectral feature matrix to be predicted based on the predicted value, the true rate threshold and the false positive rate threshold.
In this embodiment, after the true rate threshold and the false positive rate threshold, when product classification prediction is required, to-be-predicted spectrum prediction data corresponding to spectrum training data is obtained, where the spectrum training data and the predicted spectrum prediction data are both spectrum data of the same product, and then a to-be-predicted spectrum feature matrix in the to-be-predicted spectrum prediction data is input into the target spectrum detection model for prediction, so as to obtain a target denoising spectrum feature matrix.
Then, determining a plurality of target spectrum vectors corresponding to each second spectrum vector in the first spectrum vector, wherein the first spectrum vector is the spectrum vector in the denoising spectrum feature matrix of the current iteration, and the second spectrum vector is the spectrum vector in the denoising spectrum feature matrix of the target; specifically, the step S302 includes:
step S3021, obtaining second angular similarities between the first spectral vectors and the second spectral vectors;
step S3022, determining a plurality of target spectrum vectors corresponding to the respective second spectrum vectors from the first spectrum vectors based on the second angular similarity.
In this embodiment, the first spectral vector is a spectral vector in the denoising spectral feature matrix of the current iteration, the second spectral vector is a spectral vector in the denoising spectral feature matrix of the target, and the second angular similarity between each first spectral vector and each second spectral vector is calculated to obtain the angular similarity between each second spectral vector and each first spectral vector.
And determining a plurality of target spectrum vectors corresponding to the second spectrum vectors in the first spectrum vectors based on the second angular similarity, specifically, for each second spectrum vector, acquiring the maximum preset number of target angular similarities in the plurality of second angular similarities corresponding to the second spectrum vector, and taking the first spectrum vector corresponding to the target angular similarity as the target spectrum vector corresponding to the second spectrum vector, wherein the preset number can be 10, 20 and the like.
Then, based on the prediction result of the target spectrum vector, determining a prediction value corresponding to each second spectrum vector; specifically, for each second spectrum vector, a predicted result of the second spectrum vector corresponding to the target spectrum vector is obtained, and an average value of the predicted results of the second spectrum vector corresponding to the plurality of target spectrum vectors is used as a predicted value of the second spectrum vector.
And then, determining the classification result of each product corresponding to the spectral feature matrix to be predicted based on the predicted value, the true rate threshold and the false positive rate threshold, wherein for each predicted value, if the predicted value is larger than the true rate threshold, the classification result of the product corresponding to the predicted value is qualified, if the predicted value is smaller than the false positive rate threshold, the classification result of the product corresponding to the predicted value is unqualified, and if the predicted value is smaller than or equal to the true rate threshold and larger than or equal to the false positive rate threshold, the classification result of the product corresponding to the predicted value is intermediate.
According to the spectrum detection method, the spectrum characteristic matrix to be predicted in the spectrum prediction data to be predicted is input into the target spectrum detection model for prediction by acquiring the spectrum prediction data to be predicted corresponding to the spectrum training data, so that the target denoising spectrum characteristic matrix is obtained; then determining a plurality of target spectrum vectors corresponding to each second spectrum vector in the first spectrum vector, wherein the first spectrum vector is the spectrum vector in the denoising spectrum feature matrix of the current iteration, and the second spectrum vector is the spectrum vector in the denoising spectrum feature matrix of the target; then, based on the prediction result of the target spectrum vector, determining the prediction value corresponding to each second spectrum vector; and then determining classification results of products corresponding to the spectral feature matrix to be predicted based on the predicted value, the true rate threshold and the false positive rate threshold, selecting and integrating a learning score regression model by using the self-adaptive threshold to replace a common classification model so as to learn a sample spectrum, solving the problem of unbalanced classification of qualified product detection, being applicable to multi-classification tasks, being capable of stably, accurately, rapidly, nondestructively and high-throughput identifying unqualified products in the products, and solving the problems of long time consumption, high cost and low accuracy of the traditional unqualified product detection method. The method has high practicability and flexibility, can be effectively applied to various detection scenes, and has theoretical interpretability.
In addition, the embodiment of the invention also provides a computer readable storage medium, wherein the computer readable storage medium stores a spectrum detection program, and the spectrum detection program realizes the steps of the spectrum detection method when being executed by the processor.
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 phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article, or system that comprises the element.
The foregoing embodiment numbers of the present invention are merely for the purpose of description, and do not represent the advantages or disadvantages of the embodiments.
From the above description of the embodiments, it will be clear to those skilled in the art that the above-described embodiment method may be implemented by means of software plus a necessary general hardware platform, but of course may also be implemented by means of hardware, but in many cases the former is a preferred embodiment. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product stored in a storage medium (e.g. ROM/RAM, magnetic disk, optical disk) as described above, comprising instructions for causing a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) to perform the method according to the embodiments of the present invention.
The foregoing description is only of the preferred embodiments of the present invention, and is not intended to limit the scope of the invention, but rather is intended to cover any equivalents of the structures or equivalent processes disclosed herein or in the alternative, which may be employed directly or indirectly in other related arts.
Claims (10)
1. A method of spectral detection, the method comprising the steps of:
acquiring a spectrum characteristic matrix corresponding to spectrum training data, acquiring a spectrum first angle similarity matrix corresponding to the spectrum characteristic matrix, and acquiring a regularized Laplacian matrix corresponding to the spectrum first angle similarity matrix;
determining a spectrum detection model based on the spectrum characteristic matrix, the regularized Laplace matrix, the auxiliary variable and the sparse constraint noise, wherein the output of the spectrum detection model is a denoising spectrum characteristic matrix;
inputting the spectrum training data into the spectrum detection model for iterative training to obtain a denoising spectrum feature matrix corresponding to the spectrum training data and a trained spectrum detection model;
if the convergence of the trained spectrum detection model is determined based on the denoising spectrum feature matrix of the current iteration and the denoising spectrum feature matrix of the last iteration, the trained spectrum detection model is taken as a target spectrum detection model, and the denoising spectrum feature matrix of the current iteration is taken as a target denoising spectrum feature matrix.
2. The method of spectral detection according to claim 1, wherein the step of determining a spectral detection model based on the spectral feature matrix, regularized laplacian matrix, auxiliary variables, and sparsely constrained noise, wherein the output of the spectral detection model is a denoised spectral feature matrix comprises:
determining the spectral feature matrix, the denoising spectral feature matrix and the sparse constraint noise, and determining a first F-norm;
determining a trace of a matrix based on the regularized laplace matrix and the spectral feature matrix;
determining a kernel norm corresponding to the auxiliary variable, and determining a 1-norm based on the sparse constraint noise;
a spectral detection model is determined based on the first F-norm, the trace of the matrix, the kernel norm, and the 1-norm.
3. The method for spectrum detection as claimed in claim 1, wherein the step of inputting the spectrum training data into the spectrum detection model for iterative training to obtain a denoised spectrum feature matrix corresponding to the spectrum training data and a trained spectrum detection model comprises:
for each iteration training, calculating a denoising spectral feature matrix of the current iteration corresponding to the spectral feature matrix according to a linear least square method through a spectral detection model;
Calculating an auxiliary variable of the current iteration according to the singular value thresholding shrinkage operator through a spectrum detection model;
and calculating sparse constraint noise of the current iteration according to an iterative shrinkage threshold algorithm through a spectrum detection model.
4. The method for spectrum detection as claimed in claim 1, wherein the step of taking the trained spectrum detection model as the target spectrum detection model if it is determined that the trained spectrum detection model converges based on the denoised spectrum feature matrix of the current iteration and the denoised spectrum feature matrix of the last iteration comprises:
determining a second F-norm based on the denoising spectral feature matrix of the current iteration and the denoising spectral feature matrix of the last iteration;
if the second F-norm is smaller than the preset error value, determining that the trained spectrum detection model converges, and taking the trained spectrum detection model as a target spectrum detection model.
5. The method of spectrum detection as claimed in claim 1, wherein the steps of obtaining a spectral feature matrix corresponding to the spectral training data, obtaining a spectral first angular similarity matrix corresponding to the spectral feature matrix, and obtaining a regularized laplace matrix corresponding to the spectral first angular similarity matrix include:
Acquiring first angle similarity between every two spectral vectors in the spectral feature matrix, and determining a spectral first angle similarity matrix based on the first angle similarity;
determining a degree matrix based on the spectrum first angular similarity matrix, wherein the degree matrix is a diagonal matrix;
based on the degree matrix, the regularized laplacian matrix is determined.
6. The method according to any one of claims 1 to 5, wherein if it is determined that the trained spectrum detection model converges based on the denoised spectrum feature matrix of the current iteration and the denoised spectrum feature matrix of the last iteration, the step of taking the trained spectrum detection model as the target spectrum detection model further comprises:
inputting the denoising spectral feature matrix of the current iteration into an integrated learning model for prediction to obtain a prediction result corresponding to each spectrum in the spectral training data;
determining the true rate and the false positive rate corresponding to each preset threshold value based on the preset threshold values and the prediction result;
based on the true rate and the false positive rate, a ROC curve is determined, and a true rate threshold and a false positive rate threshold are determined based on the ROC curve.
7. The method of spectral detection according to claim 6, wherein after the steps of determining a ROC curve based on a true rate and a false positive rate and determining a true rate threshold and a false positive rate threshold based on the ROC curve, further comprising:
the method comprises the steps of obtaining spectrum prediction data to be predicted corresponding to spectrum training data, and inputting a spectrum feature matrix to be predicted in the spectrum prediction data to be predicted into the target spectrum detection model for prediction so as to obtain a target denoising spectrum feature matrix;
determining a plurality of target spectrum vectors corresponding to each second spectrum vector in the first spectrum vector, wherein the first spectrum vector is the spectrum vector in the denoising spectrum feature matrix of the current iteration, and the second spectrum vector is the spectrum vector in the denoising spectrum feature matrix of the target;
based on the prediction result of the target spectrum vector, determining a prediction value corresponding to each second spectrum vector;
and determining classification results of the products corresponding to the spectral feature matrix to be predicted based on the predicted value, the true rate threshold and the false positive rate threshold.
8. The spectral detection method of claim 7, wherein the step of determining a plurality of target spectral vectors corresponding to respective second spectral vectors in the first spectral vectors comprises:
Obtaining second angle similarity between each first spectrum vector and each second spectrum vector;
and determining a plurality of target spectrum vectors corresponding to the second spectrum vectors in the first spectrum vectors based on the second angular similarity.
9. A spectral detection device, the spectral detection device comprising: a memory, a processor and a spectrum sensing program stored on the memory and executable on the processor, which when executed by the processor, implements the steps of the spectrum sensing method as claimed in any one of claims 1 to 8.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored thereon a spectrum detection program which, when executed by a processor, implements the steps of the spectrum detection method according to any of claims 1 to 8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310680666.5A CN116429709B (en) | 2023-06-09 | 2023-06-09 | Spectrum detection method, spectrum detection device and computer-readable storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310680666.5A CN116429709B (en) | 2023-06-09 | 2023-06-09 | Spectrum detection method, spectrum detection device and computer-readable storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116429709A true CN116429709A (en) | 2023-07-14 |
CN116429709B CN116429709B (en) | 2023-09-12 |
Family
ID=87085798
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310680666.5A Active CN116429709B (en) | 2023-06-09 | 2023-06-09 | Spectrum detection method, spectrum detection device and computer-readable storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116429709B (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN111028172A (en) * | 2019-12-10 | 2020-04-17 | 浙江工业大学 | Hyperspectral image denoising method based on non-convex low-rank matrix approximation without parameters |
CN111860612A (en) * | 2020-06-29 | 2020-10-30 | 西南电子技术研究所(中国电子科技集团公司第十研究所) | Unsupervised hyperspectral image hidden low-rank projection learning feature extraction method |
CN112069919A (en) * | 2020-08-17 | 2020-12-11 | 浙江工业大学 | Hyperspectral image denoising method based on non-convex low-rank matrix approximation and total variation regularization |
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 |
CN114820352A (en) * | 2022-04-08 | 2022-07-29 | 南京邮电大学 | Hyperspectral image denoising method and device and storage medium |
-
2023
- 2023-06-09 CN CN202310680666.5A patent/CN116429709B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN111028172A (en) * | 2019-12-10 | 2020-04-17 | 浙江工业大学 | Hyperspectral image denoising method based on non-convex low-rank matrix approximation without parameters |
CN111860612A (en) * | 2020-06-29 | 2020-10-30 | 西南电子技术研究所(中国电子科技集团公司第十研究所) | Unsupervised hyperspectral image hidden low-rank projection learning feature extraction method |
CN112069919A (en) * | 2020-08-17 | 2020-12-11 | 浙江工业大学 | Hyperspectral image denoising method based on non-convex low-rank matrix approximation and total variation regularization |
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 |
CN114820352A (en) * | 2022-04-08 | 2022-07-29 | 南京邮电大学 | Hyperspectral image denoising method and device and storage medium |
Non-Patent Citations (1)
Title |
---|
薛志祥 等: "局部超图拉普拉斯约束的高光谱影像低秩表示去噪方法", 《光学学报》, vol. 37, no. 5, pages 0510001 - 1 * |
Also Published As
Publication number | Publication date |
---|---|
CN116429709B (en) | 2023-09-12 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11282295B2 (en) | Image feature acquisition | |
US10984289B2 (en) | License plate recognition method, device thereof, and user equipment | |
Zhang et al. | Hyperspectral band selection based on multi-objective optimization with high information and low redundancy | |
Chen et al. | A comprehensive approach to mode clustering | |
CN111915437B (en) | Training method, device, equipment and medium of money backwashing model based on RNN | |
CN107563442B (en) | Hyperspectral image classification method based on sparse low-rank regular graph tensor embedding | |
CN105740912B (en) | The recognition methods and system of low-rank image characteristics extraction based on nuclear norm regularization | |
Yadav et al. | Multiresolution local binary pattern variants based texture feature extraction techniques for efficient classification of microscopic images of hardwood species | |
CN109858575B (en) | Data classification method based on convolutional neural network | |
CN111476138B (en) | Construction method, identification method and related equipment for building drawing component identification model | |
CN106650765B (en) | High-spectral data based on convolutional neural networks turns the Hyperspectral data classification method of grayscale image | |
CN114707598A (en) | Mixed gas identification method, system, terminal equipment and readable storage medium | |
Taşcı | A meta-ensemble classifier approach: Random rotation forest | |
CN114463587A (en) | Abnormal data detection method, device, equipment and storage medium | |
Juola et al. | Classification of tree species based on hyperspectral reflectance images of stem bark | |
CN116429709B (en) | Spectrum detection method, spectrum detection device and computer-readable storage medium | |
US11397868B2 (en) | Fungal identification by pattern recognition | |
CN115479905B (en) | Spectral analysis method, spectral analysis device, terminal equipment and medium | |
Dinç et al. | Evaluation of hyperspectral image classification using random forest and Fukunaga-Koontz transform | |
CN111860601A (en) | Method and device for predicting large fungus species | |
CN111400663A (en) | Risk identification method, device, equipment and computer readable storage medium | |
Nanni et al. | Ensemble to improve gesture recognition | |
Hou et al. | FMRSS Net: Fast Matrix Representation‐Based Spectral‐Spatial Feature Learning Convolutional Neural Network for Hyperspectral Image Classification | |
Işık et al. | Consensus rule for wheat cultivar classification on VL, VNIR and SWIR imaging | |
CN118035755A (en) | Spectrum detection method, device, equipment and 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 |