CN109146003B - Hyperspectral imaging-based classification identification method, device and system - Google Patents
Hyperspectral imaging-based classification identification method, device and system Download PDFInfo
- Publication number
- CN109146003B CN109146003B CN201811156074.9A CN201811156074A CN109146003B CN 109146003 B CN109146003 B CN 109146003B CN 201811156074 A CN201811156074 A CN 201811156074A CN 109146003 B CN109146003 B CN 109146003B
- Authority
- CN
- China
- Prior art keywords
- characteristic
- factor
- hyperspectral
- identification
- images
- 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
- 238000000034 method Methods 0.000 title claims abstract description 63
- 238000000701 chemical imaging Methods 0.000 title claims abstract description 40
- 230000009467 reduction Effects 0.000 claims abstract description 41
- 238000010801 machine learning Methods 0.000 claims abstract description 20
- 238000012545 processing Methods 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 11
- 239000011159 matrix material Substances 0.000 claims description 41
- 238000000556 factor analysis Methods 0.000 claims description 19
- 238000007476 Maximum Likelihood Methods 0.000 claims description 8
- 230000003595 spectral effect Effects 0.000 claims description 7
- 230000008859 change Effects 0.000 abstract description 16
- 238000012360 testing method Methods 0.000 abstract description 8
- 239000000126 substance Substances 0.000 description 19
- 238000005516 engineering process Methods 0.000 description 18
- 238000001228 spectrum Methods 0.000 description 16
- 238000001514 detection method Methods 0.000 description 11
- 235000013372 meat Nutrition 0.000 description 10
- 238000004364 calculation method Methods 0.000 description 9
- 238000005259 measurement Methods 0.000 description 9
- 235000013305 food Nutrition 0.000 description 6
- 230000008676 import Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000010998 test method Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 4
- 238000000605 extraction Methods 0.000 description 4
- 238000009499 grossing Methods 0.000 description 4
- 238000010521 absorption reaction Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000004069 differentiation Effects 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 239000003344 environmental pollutant Substances 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000004519 manufacturing process Methods 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
- 238000012634 optical imaging Methods 0.000 description 1
- 231100000719 pollutant Toxicity 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000002310 reflectometry Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
- Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention provides a hyperspectral imaging-based classification identification method, a hyperspectral imaging-based classification identification device and a hyperspectral imaging-based classification identification system, wherein the method comprises the following steps: classifying the sample; collecting multi-frame hyperspectral images of various samples and carrying out noise reduction treatment to obtain characteristic images; analyzing the characteristic images of various samples and extracting common factors; carrying out zeroing processing on the common factors to obtain characteristic subintervals, and arranging according to the weight; importing the sorted characteristic subinterval data into a machine learning model for training; inputting a hyperspectral image of a sample to be identified into a trained machine learning model for identification, and outputting an identification result; the method is simple, the calculated amount is effectively reduced, and the hyperspectral images are collected for multiple times and subjected to noise reduction, so that the change caused by the change of the environment, the testing method and the testing personnel is avoided, and a more effective and more stable identification algorithm can be obtained; and the effective feature subspace is extracted and segmented, so that the features identifying different attributes can be quickly positioned.
Description
Technical Field
The invention belongs to the field of data processing, and particularly relates to a hyperspectral imaging-based classification identification method, device and system.
Background
The hyperspectral imaging technology integrates the spectrum technology and the image technology into a system, a large number of optical images of a sample at each wavelength are scanned and acquired in an electromagnetic wave range, and an image of the spatial position of the sample to be detected is acquired while a plurality of continuous spectrums are measured, so that the hyperspectral imaging technology is a modern optical imaging technology with high spectrum and spatial resolution. The combination of spectrum and imaging enables the system to obtain physical information of a tested sample and provide chemical spectral analysis of a substance, and due to the difference of absorption rate, reflectivity and the like influenced by different chemical and physical characteristics, the system has a corresponding absorption value at a specific wavelength, so that unique attributes of an object can be calculated according to an absorption peak value at the specific wavelength, and the accuracy of substance classification and detection work is improved. A large amount of information is stored in the obtained data, and a model with high precision and fast response is trained by combining the relevant technology of machine learning, so that the enterprise cost is effectively reduced, and the production efficiency is improved.
The hyperspectral imaging technology provides higher resolution, the number of acquired wave bands is large, experimental data is huge, so that the whole data matrix is huge, and strong correlation or noise data between adjacent wave band information can cause interference to the model precision and the analysis speed to a certain extent. Conventional data processing methods, such as the moving average smoothing method, may cause information loss due to errors in the selected width or fail to achieve the desired noise reduction goal, and even though the savtzky-Golay convolution smoothing method is used, the least square fitting filter function is used, the irrelevant information cannot be effectively extracted. Inputting such samples into a machine-learned model not only greatly increases the computational load of data but also interferes with training, affecting model accuracy.
Disclosure of Invention
The invention provides a hyperspectral imaging-based classification identification method, device and system, and aims to solve the technical problems of large data volume and large calculation amount of the existing hyperspectral imaging-based identification technology.
The invention provides a classification identification method based on hyperspectral imaging, which comprises the following steps:
classifying the sample;
receiving multi-frame hyperspectral images of various samples and performing noise reduction processing to obtain characteristic images;
analyzing the characteristic images of various samples, and extracting common factors;
zeroing the public factors to obtain characteristic subintervals, and arranging according to the weights;
importing the sorted characteristic subinterval data into a machine learning model for training;
and inputting the hyperspectral image of the sample to be recognized into the trained machine learning model for recognition, and outputting a recognition result.
According to one embodiment of the invention, the analyzing of the characteristic images of various samples and the extracting of common factors comprise:
establishing a factor analysis model;
solving a correlation matrix, an eigenvalue and an eigenvector of the normalized data;
acquiring the number of characteristic values of which the characteristic values are larger than a preset threshold value;
solving a factor load matrix according to a maximum likelihood method;
performing factor rotation according to a variance maximization method;
calculating factor scores according to the rotated factor load matrix;
and reducing the data dimension and obtaining a common factor.
According to one embodiment of the invention, the factor analysis model is as follows:
wherein, x is sample data, mu is a sample mean value, f is a common factor, epsilon is a special factor, and a is a coefficient.
According to an embodiment of the invention, the preset threshold is 1.
According to an embodiment of the invention, the weight is determined according to a length ratio of the characteristic subintervals.
According to an embodiment of the invention, the weights are determined according to variance ratio values within the feature subintervals.
A second object of the present invention is to provide a classification identifying apparatus based on hyperspectral imaging, comprising a processor and a memory, wherein the memory is used for storing a plurality of instructions, and the processor is used for reading the instructions and executing:
receiving multi-frame hyperspectral images of various samples and performing noise reduction processing to obtain characteristic images;
analyzing the characteristic images of various samples and extracting common factors;
zeroing the public factors to obtain characteristic subintervals, and arranging according to the weights;
importing the sorted characteristic subinterval data into a machine learning model for training;
and inputting the hyperspectral image of the sample to be recognized into the trained machine learning model for recognition, and outputting a recognition result.
According to an embodiment of the invention, the processor is further configured to perform:
establishing a factor analysis model;
solving a correlation matrix, an eigenvalue and an eigenvector of the normalized data;
acquiring the number of characteristic values of which the characteristic values are larger than a preset threshold value;
solving a factor load matrix according to a maximum likelihood method;
performing factor rotation according to a variance maximization method;
calculating factor scores according to the rotated factor load matrix;
and reducing the data dimension and obtaining a common factor.
According to an embodiment of the invention, the processor is further configured to perform:
and determining the weight according to the variance ratio in the characteristic subinterval or the length ratio of the characteristic subinterval.
The invention also provides a hyperspectral imaging-based classification and identification system, which comprises the above classification and identification device and a hyperspectral image acquisition device, wherein the hyperspectral image acquisition device is connected with the processor, and the hyperspectral image acquisition device is used for receiving multi-frame hyperspectral images of various samples.
The hyperspectral imaging-based classification identification method, device and system provided by the embodiment at least have the following beneficial effects:
(1) the method is simple, the calculation flow is effectively saved, the calculation amount is reduced, and the hyperspectral images are collected for multiple times and subjected to noise reduction, so that the change caused by the change of the environment, the test method and the test personnel is effectively avoided, and a more effective and more stable identification algorithm can be obtained; the effective feature subspace is extracted and segmented, and features which identify different attributes can be rapidly positioned;
(2) the characteristic quantity dimensionality reduction is adopted, the main characteristics are used as different weights, and when the neural network is used for classification, the problems caused by non-important characteristics or noise with the same characteristics can be effectively avoided;
(3) obtaining intrinsic characteristics after characteristic dimension reduction and characteristic extraction, extracting significant subspace of the characteristic space, obtaining an arrangement of weight distribution, and then adopting semi-supervised learning to effectively identify and classify so as to obtain a more accurate identification effect;
(4) compared with the traditional physical and chemical method, the detection speed is improved; meanwhile, in the measurement of the content of each component of an article to be measured, the content of a certain chemical component in food is measured by utilizing the traditional chemical and spectral technology, the obtained numerical value is the average numerical value to be measured in a sample, and the hyperspectral imaging technology can obtain the information of each point in space and can obtain more precise component distribution information, thereby improving certain accuracy.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
Fig. 1 is a flowchart of a classification identification method based on hyperspectral imaging according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of noise reduction for a two-dimensional hyperspectral image in the hyperspectral imaging-based classification and identification method provided by the embodiment of the invention.
Fig. 3 is a schematic diagram of denoising a three-dimensional hyperspectral image in the hyperspectral imaging-based classification and identification method provided by the embodiment of the invention.
Fig. 4 is a flowchart of analyzing feature images of various samples in the classification and identification method based on hyperspectral imaging according to the embodiment of the present invention.
Fig. 5 is a schematic diagram of exploratory factor analysis performed on p types of samples to obtain common factors and special factors in the classification and identification method based on hyperspectral imaging according to the embodiment of the invention.
Fig. 6 is a schematic structural diagram of a classification and identification device based on hyperspectral imaging according to an embodiment of the present invention.
Fig. 7 is a schematic structural diagram of a classification and identification system based on hyperspectral imaging according to an embodiment of the present invention.
Detailed Description
In order to make the technical problems, technical solutions and advantageous effects solved by the present invention more clearly apparent, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Example one
Referring to fig. 1, the present embodiment provides a classification and identification method based on hyperspectral imaging, including:
step S101, classifying samples;
step S102, receiving multi-frame hyperspectral images of various samples, and performing noise reduction processing to obtain characteristic images;
step S103, analyzing the characteristic images of various samples and extracting common factors;
step S104, carrying out zeroing processing on the common factors to obtain characteristic subintervals, and arranging the characteristic subintervals according to weights;
step S105, importing the sorted characteristic subinterval data into a machine learning model for training;
and S106, inputting the hyperspectral image of the sample to be recognized into the trained machine learning model for recognition, and outputting a recognition result.
Specifically, step S101 is first executed to classify the samples, and the samples share p classes, which are respectively denoted as a1，a2，......，ap。
Executing the step S102, receiving multi-frame hyperspectral images of various samples, preferably, collecting the same sample for multiple times, performing noise reduction processing, and obtaining a characteristic image a after noise reduction for a two-dimensional hyperspectral image noise reduction reference image 2 and a three-dimensional hyperspectral image noise reduction image 31 *，a2 *，......ap *。
If classification and identification are directly carried out on the basis of the characteristic spectrum, great changes can be generated due to changes of environments, test methods and test personnel, so that effective noise reduction needs to be carried out on the characteristic spectrum, and a more effective and more stable identification algorithm can be obtained. The method comprises the steps of adopting a single-point hyperspectral detector for detection, determining corresponding labels for samples to be classified in order to obtain high-quality noise reduction, dividing the samples into a plurality of groups, respectively carrying out repeated hyperspectral band detection in each group, and removing data noise by adopting a smoothing and filtering mode to obtain a smooth curve in each group.
Step S103 is executed, and for various samples, preferably, a exploratory factor analysis method is adopted to analyze the characteristic images, common factors are searched, and the essential structure of the multi-perspective observation value is found out, so that a series of bands with large differences are screened out.
Referring to fig. 4, the characteristic images of various types of samples are analyzed to extract common factors, including:
step S1031, establishing a factor analysis model;
step S1032, solving a correlation matrix, an eigenvalue and an eigenvector of the normalized data;
step S1033, acquiring the number of the characteristic values of which the characteristic values are larger than a preset threshold;
step S1034, solving a factor load matrix according to a maximum likelihood method;
step S1035, performing factor rotation according to the variance maximization method;
step S1036, calculating factor scores according to the rotated factor load matrix;
step S1037, data dimension is reduced, and a common factor is obtained.
Specifically, in step S1031, a factor analysis model is established, which is as follows:
wherein, x is sample data, mu is a sample mean value, f is a common factor, epsilon is a special factor, and a is a coefficient.
In step S1032, the correlation matrix, eigenvalue, and eigenvector of the normalized data are obtained:
the correlation matrix R of the normalized data is as follows:
solving the eigenvalue and eigenvector of R:
let | R- λ I | ═ 0;
wherein, λ is a characteristic value, I is a unit matrix, λ may have p values, and the number of p characteristic values greater than 0 is the number of factors;
solving the equation (R-lambda)iI)Xi＝0，XiIs λiThe feature vector of (2).
In step S1033, the number of eigenvalues whose eigenvalues are greater than the preset threshold is obtained, and as a preferred embodiment, the preset threshold is 1.
In step S1034, the factor load matrix is solved according to the maximum likelihood method:
the above factor analysis model can be abbreviated as:
X＝μ+Aρ+ε；
it can be demonstrated that: r is A. A*+D；
Wherein D is a diagonal matrix;
the likelihood function is
Solving an equation:
solving the unique solution to obtain a factor load matrix A*。
In step S1035, factor rotation is performed according to the variance maximization method:
A＝A*·T；
note the book
Wherein,
hicontributes to the variance;
factor load matrix A*Has a relative variance of the square of the jth column element of
Selecting an orthogonal matrix T, a factor load matrix A*Sum of relative variances of the rotated m column element sums of squares:
V＝V1+V2+...Vm
obtaining a matrix A after rotation*·T。
In step S1036, a factor score is calculated from the rotated factor load matrix:
the pole value is calculated by differentiation:
the corresponding factors are divided into:
wherein,
as an estimate, XjIs the sample data.
Referring to fig. 5, exploratory factor analysis is performed on the p-type samples, the common factor is interval II, and the special factors include interval III and interval I.
The exploration factor analysis method used in this embodiment may be replaced by other methods, and a method of performing multiple measurements and preprocessing on a single category and then performing dimension reduction processing on all categories may be adopted.
Further, referring to fig. 6, step S104 is executed to perform zeroing processing on the common factor, so as to obtain feature subintervals, and arrange the feature subintervals according to weights, where the weights are determined according to length ratios of the feature subintervals or according to variance ratio values within the feature subintervals.
And (3) obtaining the wave bands with minimum change and reflecting public information in different classes of characteristic spectra by adopting methods such as dimension reduction analysis (EFA, PCA) and the like for the characteristic images of different classes. The method can obtain a plurality of subintervals in the whole characteristic spectrum by negating the interval on the whole characteristic spectrum interval, the subintervals record the most obvious difference information between the classes, the drastic change of the subintervals directly reflects the characteristics of the current class of objects, and the space in other wave band ranges has little change, so that images which can obviously reflect the subintervals need to be reordered in the corresponding distribution ranges through characteristic dimension reduction and certain measurement indexes such as variance contribution rate, weight and the like, and the characteristic spectrum is converted into another characteristic space.
Further, step S106 is executed to import the sorted feature subinterval data into a machine learning Model for training, and import the Attention Model and import the weight information.
Further, step S107 is executed, the hyperspectral image of the sample to be recognized is input to the trained machine learning model for recognition, and a recognition result can be obtained.
The hyperspectral imaging-based classification identification method provided by the embodiment at least has the following beneficial effects:
(1) the method is simple, the calculation flow is effectively saved, the calculation amount is reduced, and the hyperspectral images are collected for multiple times and subjected to noise reduction, so that the change caused by the change of the environment, the test method and the test personnel is effectively avoided, and a more effective and more stable identification algorithm can be obtained; the effective feature subspace is extracted and segmented, and features which identify different attributes can be rapidly positioned;
(2) the characteristic quantity dimensionality reduction is adopted, the main characteristics are used as different weights, and when the neural network is used for classification, the problems caused by non-important characteristics or noise with the same characteristics can be effectively avoided;
(3) the intrinsic features are obtained after feature dimension reduction and feature extraction, the significant subspace of the feature space is extracted, a weight distribution arrangement is obtained, and then effective identification and classification are carried out by adopting semi-supervised learning, so that a more accurate identification effect is obtained.
(4) Compared with the traditional physical and chemical method, the detection speed is improved; meanwhile, in the measurement of the content of each component of an article to be measured, the content of a certain chemical component in food is measured by utilizing the traditional chemical and spectral technology, the obtained numerical value is the average numerical value to be measured in a sample, and the hyperspectral imaging technology can obtain the information of each point in space and can obtain more precise component distribution information, thereby improving certain accuracy.
Example two
Referring to fig. 6, the present embodiment provides a classification identifying apparatus based on hyperspectral imaging, including a processor 201 and a memory 202, where the memory 202 is configured to store a plurality of instructions, and the processor 201 is configured to read the instructions and execute:
receiving multi-frame hyperspectral images of various samples and performing noise reduction processing to obtain characteristic images;
analyzing the characteristic images of various samples and extracting common factors;
zeroing the public factors to obtain characteristic subintervals, and arranging according to the weights;
importing the sorted characteristic subinterval data into a machine learning model for training;
and inputting the hyperspectral image of the sample to be recognized into the trained machine learning model for recognition, and outputting a recognition result.
Specifically, the processor 201 is configured to receive multiple frames of hyperspectral images of various samples, preferably, collect the same sample for multiple times, perform noise reduction processing, and obtain a feature image a after performing noise reduction on a two-dimensional hyperspectral image noise reduction reference image 2 and a three-dimensional hyperspectral image noise reduction image 31 *，a2 *，......ap *。
If classification and identification are directly carried out on the basis of the characteristic spectrum, huge changes can be generated due to changes of environments, test methods and test personnel, so that effective noise reduction needs to be carried out on the characteristic spectrum, and a more effective and more stable identification algorithm can be obtained. The method comprises the steps of adopting a single-point hyperspectral detector for detection, determining corresponding labels for samples to be classified in order to obtain high-quality noise reduction, dividing the samples into a plurality of groups, respectively carrying out repeated hyperspectral band detection in each group, and removing data noise by adopting a smoothing and filtering mode to obtain a smooth curve in each group.
Further, the processor 201 analyzes the characteristic image by using an exploratory factor analysis method, searches for a common factor, and finds out a plurality of distant observation values and an essential structure, thereby screening a series of bands with large differences. The processor 201 is further configured to perform:
establishing a factor analysis model;
solving a correlation matrix, an eigenvalue and an eigenvector of the normalized data;
acquiring the number of characteristic values of which the characteristic values are larger than a preset threshold value;
solving a factor load matrix according to a maximum likelihood method;
performing factor rotation according to a variance maximization method;
calculating factor scores according to the rotated factor load matrix;
and reducing the data dimension and obtaining a common factor.
Specifically, a factor analysis model is established, which is as follows:
wherein, x is sample data, mu is a sample mean value, f is a common factor, epsilon is a special factor, and a is a coefficient.
And (3) solving a correlation matrix, an eigenvalue and an eigenvector of the normalized data:
the correlation matrix R of the normalized data is as follows:
solving the eigenvalue and eigenvector of R:
let | R- λ I | ═ 0;
wherein, λ is a characteristic value, I is a unit matrix, λ may have p values, and the number of p characteristic values greater than 0 is the number of factors;
solving the equation (R-lambda)iI)xi＝0，XiIs λiThe feature vector of (2).
As a preferred embodiment, the number of the eigenvalues of which the eigenvalues are greater than the preset threshold is obtained, and the preset threshold is 1.
Solving the factor load matrix according to a maximum likelihood method:
the above factor analysis model can be abbreviated as:
X＝μ+Aρ+ε；
it can be demonstrated that: r is A. A*+D；
Wherein D is a diagonal matrix;
the likelihood function is
Solving an equation:
solving the unique solution to obtain a factor load matrix A*。
Factor rotation is performed according to the variance maximization method:
A＝A*·T；
note book
Wherein,
hicontributes to the variance;
factor load matrix A*Has a relative variance of the square of the jth column element of
Selecting an orthogonal matrix T, a factor load matrix A*Sum of relative variances of the rotated m column element sums of squares:
V＝V1+V2+...Vm
obtaining a matrix A after rotation*.T。
Calculating a factor score according to the rotated factor load matrix:
the pole value is calculated by differentiation:
the corresponding factors are divided into:
wherein,
as an estimate, XjIs the sample data.
Referring to fig. 5, exploratory factor analysis is performed on the p-type samples, the common factor is interval II, and the special factors include interval III and interval I.
The exploration factor analysis method used in this embodiment may be replaced by other methods, and a method of performing multiple measurements and preprocessing on a single category and then performing dimension reduction processing on all categories may be adopted.
Further, the processor 201 is further configured to: and performing zeroing processing on the common factor to obtain a characteristic subinterval, and arranging the characteristic subintervals according to weight, wherein the weight is determined according to the length ratio of the characteristic subinterval or the variance ratio in the characteristic subinterval.
And (3) adopting methods such as dimension reduction analysis (EFA, PCA) and the like to the feature images of different classes to obtain the wave bands with minimum change in the feature spectra of different classes and reflecting the public information. The method can obtain a plurality of subintervals in the whole characteristic spectrum by negating the interval on the whole characteristic spectrum interval, the subintervals record the most obvious difference information between the classes, the drastic change of the subintervals directly reflects the characteristics of the current class of objects, and the space in other wave band ranges has little change, so that images which can obviously reflect the subintervals need to be reordered in the corresponding distribution ranges through characteristic dimension reduction and certain measurement indexes such as variance contribution rate, weight and the like, and the characteristic spectrum is converted into another characteristic space.
Further, the processor 201 is further configured to import the sorted feature subinterval data into a machine learning Model for training, and import an Attention Model and import weight information.
Further, the processor 201 is further configured to input the hyperspectral image of the sample to be recognized into the trained machine learning model for recognition, so as to obtain a recognition result.
The classification and identification device based on hyperspectral imaging provided by the embodiment at least has the following beneficial effects:
(1) the calculation process is effectively saved, the calculation amount is reduced, and the hyperspectral images are collected for multiple times and subjected to noise reduction, so that the change caused by the change of the environment, the testing method and testing personnel is effectively avoided, and a more effective and more stable identification algorithm can be obtained; the effective feature subspace is extracted and segmented, and features which identify different attributes can be rapidly positioned;
(2) the characteristic quantity dimensionality reduction is adopted, the main characteristics are used as different weights, and when the neural network is used for classification, the problems caused by non-important characteristics or noise with the same characteristics can be effectively avoided;
(3) the intrinsic features are obtained after feature dimension reduction and feature extraction, the significant subspace of the feature space is extracted, a weight distribution arrangement is obtained, and then effective identification and classification are carried out by adopting semi-supervised learning, so that a more accurate identification effect is obtained.
(4) Compared with the traditional physical and chemical method, the detection speed is improved; meanwhile, in the measurement of the content of each component of an article to be measured, the content of a certain chemical component in food is measured by utilizing the traditional chemical and spectral technology, the obtained numerical value is the average numerical value to be measured in a sample, and the hyperspectral imaging technology can obtain the information of each point in space and can obtain more precise component distribution information, thereby improving certain accuracy.
EXAMPLE III
Referring to fig. 7, the present embodiment provides a classification and identification system based on hyperspectral imaging, which includes a classification and identification device 301, and further includes a hyperspectral image acquisition device 302, where the hyperspectral image acquisition device 302 is connected to a processor in the classification and identification device 301, and the hyperspectral image acquisition device 302 is configured to receive multiframe hyperspectral images of various samples.
For the working principle of the classification recognition apparatus 301, please refer to embodiment two, which is not repeated herein.
The classification recognition system provided by the embodiment at least has the following beneficial effects:
(1) the calculation process is effectively saved, the calculation amount is reduced, and the hyperspectral images are collected for multiple times and subjected to noise reduction, so that the change caused by the change of the environment, the test method and the tester is effectively avoided, and a more effective and more stable identification algorithm can be obtained; the effective feature subspace is extracted and segmented, and features which identify different attributes can be rapidly positioned;
(2) the characteristic quantity dimensionality reduction is adopted, the main characteristics are used as different weights, and when the neural network is used for classification, the problems caused by non-important characteristics or noise with the same characteristics can be effectively avoided;
(3) the intrinsic features are obtained after feature dimension reduction and feature extraction, the significant subspace of the feature space is extracted, a weight distribution arrangement is obtained, and then effective identification and classification are carried out by adopting semi-supervised learning, so that a more accurate identification effect is obtained.
(4) Compared with the traditional physical and chemical method, the detection speed is improved; meanwhile, in the measurement of the content of each component of an article to be measured, the content of a certain chemical component in food is measured by utilizing the traditional chemical and spectral technology, the obtained numerical value is the average numerical value to be measured in a sample, and the hyperspectral imaging technology can obtain the information of each point in space and can obtain more precise component distribution information, thereby improving certain accuracy.
Example four
The embodiment provides a specific application scenario, and further describes the classification recognition device.
The method is applied to the field of safety detection of food industry, high-quality meat and inferior meat are firstly classified, multi-frame hyperspectral images of different types of meat are collected through a hyperspectral image collection device, noise reduction processing is carried out, and characteristic images are obtained; analyzing the characteristic images of various samples and extracting common factors; and carrying out zeroing processing on the common factors to obtain characteristic subintervals, arranging according to weights, importing the sorted characteristic subinterval data into a machine learning model for training, and obtaining a trained model.
In actual inspection, high-quality meat and inferior meat are mixed together, and a multi-frame hyperspectral image of meat to be detected is acquired through a hyperspectral image acquisition device and is sent to a processor, and characteristic factors such as components, quality and pollutant guarantee period of the meat to be detected can be detected out by inputting the trained identification model, so that the high-quality meat and the inferior meat are identified. Compared with the traditional physical and chemical method, the detection speed is improved; meanwhile, in the measurement of the content of each component of the meat, the content of a certain chemical component in the food is measured by utilizing the traditional chemical and spectral technology, the obtained numerical value is the average numerical value to be measured in the sample, and the hyperspectral imaging technology can obtain the information of each point in space and can obtain more precise component distribution information, thereby improving certain accuracy.
In the description of the present invention, it is to be understood that the positional or orientational relationships indicated by the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", "axial", "radial", "circumferential", and the like are based on the positional or orientational relationships shown in the drawings and are intended to facilitate the description of the invention and to simplify the description, but do not indicate or imply that the referenced device or element must have a particular orientation, be constructed and operated in a particular orientation, and thus are not to be construed as limiting the invention. Furthermore, a feature defined as "first", "second" may explicitly or implicitly include one or more of the feature. In the description of the present invention, "a plurality" means two or more unless otherwise specified.
In the description of the present invention, it is to be noted that, unless otherwise explicitly specified or limited, the terms "mounted", "connected" and "connected" are to be construed broadly, e.g., as meaning either a fixed connection, a removable connection, or an integral connection; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood in specific cases to those skilled in the art.
In the description herein, reference to the description of the terms "one embodiment," "some embodiments," "an illustrative embodiment," "an example," "a specific example," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the claims and their equivalents.
Claims (8)
1. A classification identification method based on hyperspectral imaging is characterized by comprising the following steps:
classifying the sample;
receiving multi-frame hyperspectral images of various samples and performing noise reduction processing to obtain characteristic images;
analyzing the characteristic images of various samples and extracting common factors;
zeroing the public factors to obtain characteristic subintervals, and arranging according to the weights;
importing the sorted characteristic subinterval data into a machine learning model for training;
inputting a hyperspectral image of a sample to be identified into a trained machine learning model for identification, and outputting an identification result;
performing zeroing processing on the common factor, including: negating the common factor over the entire characteristic spectral interval;
analyzing the characteristic images of various samples, and extracting common factors, wherein the common factors comprise:
establishing a factor analysis model;
solving a correlation matrix, an eigenvalue and an eigenvector of the normalized data;
acquiring the number of characteristic values of which the characteristic values are larger than a preset threshold value;
solving a factor load matrix according to a maximum likelihood method;
performing factor rotation according to a variance maximization method;
calculating factor scores according to the rotated factor load matrix;
and reducing the data dimension and obtaining a common factor.
2. The hyperspectral imaging-based classification and identification method according to claim 1 is characterized in that: the factor analysis model is as follows:
wherein, x is sample data, mu is a sample mean value, f is a common factor, epsilon is a special factor, and a is a coefficient.
3. The hyperspectral imaging-based classification and identification method according to claim 2, characterized in that: the preset threshold is 1.
4. The hyperspectral imaging-based classification and identification method according to claim 3 is characterized in that: the weight is determined according to the length ratio of the characteristic subintervals.
5. The hyperspectral imaging-based classification and identification method according to claim 3, characterized in that: the weight is determined according to the variance ratio in the characteristic subinterval.
6. The utility model provides a categorised recognition device based on hyperspectral imaging which characterized in that: comprises a processor and a memory, wherein the memory is used for storing a plurality of instructions, and the processor is used for reading the instructions and executing:
receiving multi-frame hyperspectral images of various samples and performing noise reduction processing to obtain characteristic images;
analyzing the characteristic images of various samples and extracting common factors;
zeroing the public factors to obtain characteristic subintervals, and arranging according to the weights;
importing the sorted characteristic subinterval data into a machine learning model for training;
inputting a hyperspectral image of a sample to be identified into a trained machine learning model for identification, and outputting an identification result;
performing zeroing processing on the common factor, including: negating the common factor over the entire characteristic spectral interval;
the processor is further configured to perform:
establishing a factor analysis model;
solving a correlation matrix, an eigenvalue and an eigenvector of the normalized data;
acquiring the number of characteristic values of which the characteristic values are larger than a preset threshold value;
solving a factor load matrix according to a maximum likelihood method;
performing factor rotation according to a variance maximization method;
calculating factor scores according to the rotated factor load matrix;
and reducing the data dimension and obtaining a common factor.
7. The hyperspectral imaging-based classification and identification device according to claim 6, wherein: the processor is further configured to perform:
and determining the weight according to the ratio of the variance sizes in the characteristic subintervals or the length ratio of the characteristic subintervals.
8. A classification recognition system based on hyperspectral imaging is characterized in that: the classification and identification device comprises the classification and identification device according to any one of claims 6 to 7, and further comprises a hyperspectral image acquisition device, wherein the hyperspectral image acquisition device is connected with the processor, and the hyperspectral image acquisition device is used for receiving multiframe hyperspectral images of various samples.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811156074.9A CN109146003B (en) | 2018-09-30 | 2018-09-30 | Hyperspectral imaging-based classification identification method, device and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811156074.9A CN109146003B (en) | 2018-09-30 | 2018-09-30 | Hyperspectral imaging-based classification identification method, device and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109146003A CN109146003A (en) | 2019-01-04 |
| CN109146003B true CN109146003B (en) | 2022-06-07 |
Family
ID=64814086
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811156074.9A Active CN109146003B (en) | 2018-09-30 | 2018-09-30 | Hyperspectral imaging-based classification identification method, device and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109146003B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110991458B (en) * | 2019-11-25 | 2023-05-23 | 创新奇智(北京)科技有限公司 | Image feature-based artificial intelligent recognition result sampling system and sampling method |
| CN111310571B (en) * | 2020-01-17 | 2022-11-22 | 中国科学院长春光学精密机械与物理研究所 | Hyperspectral image classification method and device based on spatial-spectral-dimensional filtering |
| CN115618051B (en) * | 2022-12-20 | 2023-03-21 | 楠楠聚智信息科技有限责任公司 | Internet-based smart campus monitoring video storage method |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101487892A (en) * | 2009-02-23 | 2009-07-22 | 北京航空航天大学 | High-spectrum data dimensionality reduction method based on factor analysis model |
| CN102592134A (en) * | 2011-11-28 | 2012-07-18 | 北京航空航天大学 | Multistage decision fusing and classifying method for hyperspectrum and infrared data |
| CN102799903A (en) * | 2012-07-18 | 2012-11-28 | 中国科学院自动化研究所 | High-spectrum automatic cloud detection method based on space and spectral information |
| CN106960221A (en) * | 2017-03-14 | 2017-07-18 | 哈尔滨工业大学深圳研究生院 | A kind of hyperspectral image classification method merged based on spectral signature and space characteristics and system |
| CN107644212A (en) * | 2017-10-19 | 2018-01-30 | 中国地质大学（武汉） | A kind of Classification of hyperspectral remote sensing image method, equipment and storage device |
| CN107958216A (en) * | 2017-11-27 | 2018-04-24 | 沈阳航空航天大学 | Based on semi-supervised multi-modal deep learning sorting technique |
| CN108182449A (en) * | 2017-12-25 | 2018-06-19 | 南京信息工程大学 | A kind of hyperspectral image classification method |
| CN108388917A (en) * | 2018-02-26 | 2018-08-10 | 东北大学 | A kind of hyperspectral image classification method based on improvement deep learning model |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8315472B2 (en) * | 2009-05-29 | 2012-11-20 | Raytheon Company | System and method for reducing dimensionality of hyperspectral images |
| CN104239902B (en) * | 2014-09-12 | 2018-04-24 | 西安电子科技大学 | Hyperspectral image classification method based on non local similitude and sparse coding |
| CN104700109B (en) * | 2015-03-24 | 2018-04-10 | 清华大学 | The decomposition method and device of EO-1 hyperion intrinsic image |
| CN108154094B (en) * | 2017-12-14 | 2020-04-24 | 浙江工业大学 | Hyperspectral image unsupervised waveband selection method based on subinterval division |
-
2018
- 2018-09-30 CN CN201811156074.9A patent/CN109146003B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101487892A (en) * | 2009-02-23 | 2009-07-22 | 北京航空航天大学 | High-spectrum data dimensionality reduction method based on factor analysis model |
| CN102592134A (en) * | 2011-11-28 | 2012-07-18 | 北京航空航天大学 | Multistage decision fusing and classifying method for hyperspectrum and infrared data |
| CN102799903A (en) * | 2012-07-18 | 2012-11-28 | 中国科学院自动化研究所 | High-spectrum automatic cloud detection method based on space and spectral information |
| CN106960221A (en) * | 2017-03-14 | 2017-07-18 | 哈尔滨工业大学深圳研究生院 | A kind of hyperspectral image classification method merged based on spectral signature and space characteristics and system |
| CN107644212A (en) * | 2017-10-19 | 2018-01-30 | 中国地质大学（武汉） | A kind of Classification of hyperspectral remote sensing image method, equipment and storage device |
| CN107958216A (en) * | 2017-11-27 | 2018-04-24 | 沈阳航空航天大学 | Based on semi-supervised multi-modal deep learning sorting technique |
| CN108182449A (en) * | 2017-12-25 | 2018-06-19 | 南京信息工程大学 | A kind of hyperspectral image classification method |
| CN108388917A (en) * | 2018-02-26 | 2018-08-10 | 东北大学 | A kind of hyperspectral image classification method based on improvement deep learning model |
Non-Patent Citations (3)
| Title |
|---|
| Hyperspectral Image Classification With Deep Feature Fusion Network;Weiwei Song,et al.;《IEEE》;20180207;第3173-3184页 * |
| 网络安全态势评估指标体系优化模型研究;吴果 等;《计算机工程与科学》;20170531;第39卷(第5期);第861-869页 * |
| 高光谱图像分类方法的若干研究;杨仁欣;《中国优秀硕士学位论文全文数据库 信息科技辑》;20170315;第4-5、8、15、19、22-23页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109146003A (en) | 2019-01-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109146003B (en) | Hyperspectral imaging-based classification identification method, device and system | |
| CN109142317B (en) | Raman spectrum substance identification method based on random forest model | |
| CN110717368A (en) | Qualitative classification method for textiles | |
| CN109858477A (en) | The Raman spectrum analysis method of object is identified in complex environment with depth forest | |
| CN105424641B (en) | A kind of near infrared spectrum recognition methods of crude oil type | |
| CN103235095A (en) | Water-injected meat detection method and device | |
| CN1831516A (en) | Method for nondistructive discriminating variety and true and false of cigarette using visible light and near-infrared spectrum technology | |
| CN112669915B (en) | Pear nondestructive testing method based on neural network and near infrared spectrum | |
| CN107247033B (en) | Identify the method for Huanghua Pear maturity based on rapid decay formula life cycle algorithm and PLSDA | |
| CN107679569A (en) | Raman spectrum substance automatic identifying method based on adaptive hypergraph algorithm | |
| CN103674854A (en) | Apple grading method based on synergic interval partial least square discriminant analysis | |
| CN111044483A (en) | Method, system and medium for determining pigment in cream based on near infrared spectrum | |
| CN101498661A (en) | Infrared spectrum characteristic extraction method for high-precision discriminating variety, producing area and growth mode of traditional Chinese medicinal materials | |
| CN110749565A (en) | Method for rapidly identifying storage years of Pu' er tea | |
| CN110378374A (en) | A kind of tealeaves near infrared light profile classification method that fuzzy authentication information extracts | |
| CN109685098B (en) | Tea variety classification method for fuzzy inter-cluster separation and clustering | |
| CN107632010A (en) | A kind of quantitative approach of combination LIBS to steel samples | |
| CN115905881B (en) | Yellow pearl classification method and device, electronic equipment and storage medium | |
| CN115420726B (en) | Method for rapidly identifying target object by utilizing reconstructed SERS spectrum | |
| CN102937575A (en) | Watermelon sugar degree rapid modeling method based on secondary spectrum recombination | |
| CN111523587A (en) | Woody plant species spectrum identification method based on machine learning | |
| CN111693484A (en) | Sesame oil variety identification method based on terahertz time-domain spectroscopy | |
| CN113418889B (en) | Real-time detection method for water content and total colony count of dried vegetables based on deep learning | |
| CN104297205A (en) | Method for quickly and nondestructively identifying edible oil | |
| CN111881738B (en) | Near infrared spectrum classification method for tea leaves through nuclear fuzzy orthogonal discriminant analysis |
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 |