CN109508647A - A kind of spectra database extended method based on generation confrontation network - Google Patents
A kind of spectra database extended method based on generation confrontation network Download PDFInfo
- Publication number
- CN109508647A CN109508647A CN201811228596.5A CN201811228596A CN109508647A CN 109508647 A CN109508647 A CN 109508647A CN 201811228596 A CN201811228596 A CN 201811228596A CN 109508647 A CN109508647 A CN 109508647A
- Authority
- CN
- China
- Prior art keywords
- spectrum
- network
- spectra
- initial experiments
- data
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/194—Terrestrial scenes using hyperspectral data, i.e. more or other wavelengths than RGB
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Multimedia (AREA)
- Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)
Abstract
The present invention relates to a kind of based on the spectra database extended method for generating confrontation network, belongs to technical field of spectral detection, is widely used in the extension of the spectra databases such as near infrared spectrum, Raman spectrum, laser induced breakdown spectroscopy, fluorescence spectrum and Terahertz.Tested sample is measured using spectral measurement system to obtain a small amount of experimental spectrum.Spectrum is originally generated according to the dimension of experimental spectrum obtained generation same dimension.Construction generates network G and differentiates network D, to generation network and differentiates that network carries out the interactive training of shared parameter formula.Above interactive training step is repeated, is increased with interactive training number, spectrum is generated and gradually levels off to Initial experiments spectrum.Using but be not limited to unsupervised learning clustering method, the similarity for generating spectrum and Initial experiments spectrum is judged.If being unsatisfactory for requiring, repeated interaction training step;If meeting the requirements, spectra database is constituted together with generation spectrum and original spectrum.
Description
Technical field
The present invention relates to a kind of based on the spectra database extended method for generating confrontation network, belongs to spectrum detection technique neck
Domain.
Background technique
Spectral analysis technique, such as near infrared spectrum (Near Infrared Spectroscopy), Raman spectrum
(Raman Spectroscopy), laser induced breakdown spectroscopy (Laser-induced breakdown spectroscopy,
LIBS), fluorescence spectrum (Fluorescence Spectroscopy) and tera-hertz spectra (Terahertz spectroscopy)
Deng, at present explosive detection, clinical medicine pattern detection, cultural heritage identification, alloy process, space probation and
Agricultural is widely studied and applies with fields such as food analyses.According to the spectrum of acquisition, discriminant analysis can be carried out to substance
Or Classification and Identification.Due to spectral technique have it is quickly real-time, convenient accurate and can online in situ detection advantage, will know in classification
Other field plays an increasingly important role.
When carrying out identification classification using spectroscopic data, in order to construct disaggregated model, first have to establish spectra database.Cause
It can all be influenced by environment and instrument itself performance in acquisition process for every spectrogram, it is not complete for leading to every spectrogram
Equally, but there is certain randomness.Therefore, only with a small amount of spectroscopic data building spectra database be it is inappropriate, need
It obtains a large amount of spectrum to every kind of substance to be used to build library, to cover randomness range in class, so that spectroscopic data and inhomogeneity in class
The differentiation of spectroscopic data is more accurate.
However in certain special occasions and field, it tends to be difficult to obtain enough spectroscopic datas for building library.For example, LIBS
Technology is so that sample is generated plasma using High Power Laser Pulses ablation tested sample surface, and radiation issues characteristic spectrum.
LIBS technology is not the detection technique completely lossless to sample, when being detected to precious sample or the difficult sample obtained,
Cause to be difficult to obtain a large amount of spectroscopic datas since sample is at high cost.And Raman spectral technique is needed when obtaining and often shining modal data
The longer time of integration is wanted, particularly with the Raman device excited using pulse laser, it is often necessary to multiple laser pulses
Excitation accumulation could obtain the preferable Raman spectrum of a signal-to-background ratio.This to obtain time that enough spectrum needs significantly
Increase, to increase manpower and material resources.These all give the Classification and Identification application of many special dimensions and occasion to bring difficulty, compel
The one kind to be found that is essential can complete spectra database extension based on a small amount of experimental spectrum, and then carry out modeling and realize spectral classification
Know method for distinguishing.
Summary of the invention
The purpose of the present invention is to solve being difficult to obtain enough spectroscopic datas for establishing the problem of database, one is provided
Kind can be widely used near infrared spectrum, Raman spectrum, laser based on the spectra database extended method for generating confrontation network
The extension of the spectra databases such as induced breakdown spectroscopy, fluorescence spectrum and Terahertz adapts to above method every kind of substance of detection
Spectrum can carry out analogue simulation, solve the problems, such as to be difficult to obtain a large amount of spectroscopic datas in special dimension and occasion.
Technical scheme is as follows:
A kind of spectra database extended method based on generation confrontation network, it is characterized in that this method comprises the following steps:
1. carrying out spectrum data gathering.
Spectra collection experimental provision is built, the spectrum data gathering of sample to be tested is carried out.
2. building generates confrontation network (including generate network G and differentiate network D).
It is constructed using artificial nerve network model and generates network model G, according to used Initial experiments spectrum, generate phase
With the specified mean value of dimension and the normal distribution random number sequence of variance, as being originally generated spectrum;
3. using spectrum and experimental spectrum building differentiation network model D is generated.
Differentiate that network model D is the neural network model of one two classification, the minimum number of plies that is arranged is 3 layers, and the last layer is
Output layer, output layer number of nodes are 1.To Initial experiments spectrum and spectrum progress discriminant analysis is generated using D network, by original reality
Optometry spectrum and generation spectrum are considered as inhomogeneity sample, obtain the training parameter of D network.To setting when differentiating that network D is trained
Loss function is as follows:
Wherein, D (x) is the output for differentiating network on original real spectrum data set, and x~pdata is original true light
The probability distribution of spectrum data set, D (G (z)) are to differentiate that network is generating the output on the simulated spectra data set that network generates, z
~pz (z) makes a living into the spectroscopic data collection probability distribution of network simulation, and z is random vector, EX~pdata(x) [logD (x)] is indicated
Differentiate that network can distinguish true experiment gained spectrum and emulation gained spectroscopic data, EZ~pz (z)[log (1-D (G (z)))] table
Show that generating network can be generated the simulated spectra data for differentiating that network is indistinguishable.
4. the training parameter of shared D network is trained to network model G is generated.
Initial experiments spectrum and generation spectrum are considered as similar sample at this time, damaged to setting when network G is trained is generated
It is as follows to lose function:
Wherein every meaning is identical as meaning every in the loss function for differentiating network D.
Loss function can combine is defined as:
5. generating new simulated spectra according to the new generation network G that training obtains.
6. repeat step 3-5, makes to generate spectrum and gradually level off to Initial experiments spectrum.
7. pair generation spectrum and Initial experiments spectrum carry out similarity judgement.
Using but be not limited to unsupervised learning clustering method, to generate the similarity of spectrum and Initial experiments spectrum into
Row judgement.If being unsatisfactory for requiring, repeatedly step 3-5;If meeting the requirements, light is constituted together with generation spectrum and original spectrum
Modal data library.
Beneficial effect
Spectral method of detection is that have one of potential analysis method in current material identification and detection.But many special
Different application, it is difficult to quickly and easily obtain mass data for establishing spectra database.A kind of base proposed by the invention
It can be constructed compared with prior art by a small amount of experimental spectrum data in the spectra database extended method for generating confrontation network
The largely generation spectroscopic data very much like with experimental spectrum, completes the extension of spectra database, solves for being difficult to obtain
Enough samples and being difficult to measure enough time etc., and spectroscopic data builds the problem of library hardly possible in special circumstances, detects to promote spectral matching factor
Method provides possibility in the application of these special dimensions.
Detailed description of the invention
Fig. 1 is the method for the present invention flow diagram;
Fig. 2 is that a kind of laser induced breakdown spectroscopy measuring system constitutes schematic diagram;
Fig. 3 is that a kind of Raman spectrum measurement system constitutes schematic diagram;
The LIBS spectrum that Fig. 4 is typical hazard product high explosive CL-20 in embodiment 1 generates spectrum and original spectrum compares
Figure, in which:
It (a) (b) is experimental spectrum to generate spectrum;
The LIBS spectrum that Fig. 5 is typical hazard product high explosive CL-20 in embodiment 1 generates spectrum and original spectrum uses
The cluster analysis result of unsupervised learning method PCA;
The LIBS spectrum that Fig. 6 is typical hazard product high explosive CL-20 in embodiment 1 generates spectrum and original spectrum uses
The cluster analysis result of unsupervised learning method K-means;
The Raman spectrum that Fig. 7 is typical hazard product high explosive CL-20 in embodiment 2 generates spectrum and original spectrum pair
Than figure, in which:
It (a) (b) is experimental spectrum to generate spectrum;
The Raman spectrum that Fig. 8 is typical hazard product high explosive CL-20 in embodiment 2 generates spectrum and adopts with original spectrum
With the cluster analysis result of unsupervised learning method PCA;
The Raman spectrum that Fig. 9 is typical hazard product high explosive CL-20 in embodiment 2 generates spectrum and adopts with original spectrum
With the cluster analysis result of unsupervised learning method K-means.
Specific embodiment
Objects and advantages in order to better illustrate the present invention, with reference to the accompanying drawings and examples to the content of present invention make into
One step explanation.
Embodiment 1
By taking the LIBS spectrum of typical hazard product high explosive CL-20 as an example, illustrate to lure based on the laser for generating confrontation network
Lead the application of breakdown spectral and Raman spectrum data extension modeling method in LIBS spectrum.
Step 1: acquisition LIBS spectrum
Laser induced breakdown spectroscopy experimental provision as shown in Figure 2 is built, wherein 1 being laser, 2 being reflecting mirror, 3 be poly-
Focus lens, 4 be laser induced plasma, 5 be photodetector, 6 be three-dimensional sample platform, 7 be optical fiber, 8 be delayer
DG535,9 are spectrometer, and 10 be computer.
CL-20 powder is adhered to slide surface with double-sided adhesive, is placed on sample stage 6, LIBS spectrum, laser are acquired
Device uses Nd:YAG laser fundamental frequency 1064nm output wavelength, and frequency 1Hz adjusts three-dimensional sample platform, makes the acquisition of every luminous spectrum
Point is located at different location on sample, totally to avoid the ablation of surface C L-20 sample, excites the spectrum of glass slide.Removal focuses
Former luminous modal datas of adjustment collect 100 hair LIBS spectrum altogether.
Step 2: building generates confrontation network (including generate network G and differentiate network D).
Construction generates network model G, simulated spectra is generated for the first time, according to used CL-20 Initial experiments spectrum (4092
Dimension), the normal distribution random number sequence that the specified mean value for generating 100 hair identical dimensionals is 0 and variance is 1.
Step 3: using spectrum and experimental spectrum building differentiation network model D is generated.
It is constructed using artificial nerve network model and differentiates network model D, D is two Classification Neurals, and the number of plies is arranged
It is 3 layers, the first-level nodes number is mutually all 4092 with dimension, and middle layer is hidden layer, and number of nodes 200, the last layer is output
Layer, output layer number of nodes are 1, to Initial experiments spectrum and spectrum progress discriminant analysis are generated using D network, by Initial experiments light
Spectrum and generation spectrum are considered as inhomogeneity sample, obtain the training parameter of D network, lose to setting when differentiating that network D is trained
Function is as follows:
Wherein D (x) is the output for differentiating network on original real spectrum data set, and x~pdata is original real spectrum
The probability distribution of data set, D (G (z)) be differentiate network generate network generate simulated spectra data set on output, z~
Pz (z) makes a living into the spectroscopic data collection probability distribution of network simulation, and z is random vector, EX~pdata(x) [logD (x)] expression is sentenced
Other network can distinguish true experiment gained spectrum and emulation gained spectroscopic data, EZ~pz (z)[log (1-D (G (z)))] is indicated
Generating network can be generated the simulated spectra data for differentiating that network is indistinguishable.
Step 4: the training parameter of shared D network, is trained to network model G is generated.
Initial experiments spectrum and generation spectrum are considered as similar sample at this time, damaged to setting when network G is trained is generated
It is as follows to lose function:
Wherein every meaning is identical as meaning every in the loss function for differentiating network D.
Loss function, which can combine, to be defined as:
Step 5: generating new simulated spectra according to the new generation network G that training obtains.
Step 6: repeat step 3~step 5, makes to generate spectrum and gradually level off to Initial experiments spectrum.Interactive training
When number is 180000 times, generates spectrum and Initial experiments spectrum is as shown in Figure 4.
Step 7: to spectrum and the progress similarity judgement of Initial experiments spectrum is generated.
Cluster point is carried out to true experimental data and emulation data using two kinds of unsupervised learning methods of PCA, K-means
Analysis, judges whether simulated spectra and experimental spectrum mix as one kind, cannot be distinguished.The LIBS of typical hazard product high explosive CL-20
It is as shown in Figure 5 using the clustering figure of unsupervised learning method PCA with original spectrum that spectrum generates spectrum;Typical hazard product are high
The LIBS spectrum of bursting charge CL-20 generates spectrum and original spectrum uses the clustering figure of unsupervised learning method K-means
As shown in Figure 6.
Analyzing result proves that generation spectrum is not only intuitively apparently extremely similar with Initial experiments spectrum, by typical non-
Supervised Clustering Methods PCA and K-means can not distinguish the two, and generating spectrum can be used for constructing database, complete
Database extension, to include but is not limited to that the supervised learnings such as SVM method carries out modeling and provides sufficient LIBS spectroscopic data.
Embodiment 2
By taking the Raman spectrum of typical hazard product high explosive CL-20 as an example, illustrate to lure based on the laser for generating confrontation network
Lead the application of breakdown spectral and Raman spectrum data extension modeling method in Raman spectrum.
Step 1: acquisition Raman spectrum
Raman spectrum experimental provision as shown in Figure 2 is built, wherein 11 be hand-held semiconductor laser and receiver, 12
It is the computer of integrated form Raman spectrometer for sample bottle, 13.
CL-20 powder is contained with sample bottle, sample bottle side wall is directed at using hand-held semiconductor laser and receiver,
It is directly measured through sample bottle.The continuous laser that laser output is wavelength 785nm, time of integration 1125ms, removal focus adjustment
Former luminous modal datas, collect 100 hair Raman altogether along side wall.
Step 2: building generates confrontation network (including generate network G and differentiate network D).
Construction generates network model G, simulated spectra is generated for the first time, according to used CL-20 Initial experiments spectrum (1987
Dimension), the normal distribution random number sequence that the specified mean value for generating 100 hair identical dimensionals is 0 and variance is 1.
Step 3: using spectrum and experimental spectrum building differentiation network model D is generated.
It is constructed using artificial nerve network model and differentiates network model D, D is two Classification Neurals, and the number of plies is arranged
It is 3 layers, the first-level nodes number is mutually all 1987 with dimension, and middle layer is hidden layer, and number of nodes 200, the last layer is output
Layer, output layer number of nodes are 1, to Initial experiments spectrum and spectrum progress discriminant analysis are generated using D network, by Initial experiments light
Spectrum and generation spectrum are considered as inhomogeneity sample, obtain the training parameter of D network, lose to setting when differentiating that network D is trained
Function is as follows:
Wherein D (x) is the output for differentiating network on original real spectrum data set, and x~pdata is original real spectrum
The probability distribution of data set, D (G (z)) be differentiate network generate network generate simulated spectra data set on output, z~
Pz (z) makes a living into the spectroscopic data collection probability distribution of network simulation, and z is random vector, EX~pdata(x) [logD (x)] expression is sentenced
Other network can distinguish true experiment gained spectrum and emulation gained spectroscopic data, EZ~pz (z)[log (1-D (G (z)))] is indicated
Generating network can be generated the simulated spectra data for differentiating that network is indistinguishable.
Step 4: the training parameter of shared D network, is trained to network model G is generated.
Initial experiments spectrum and generation spectrum are considered as similar sample at this time, damaged to setting when network G is trained is generated
It is as follows to lose function:
Wherein every meaning is identical as meaning every in the loss function for differentiating network D.
Loss function, which can combine, to be defined as:
Step 5: generating new simulated spectra according to the new generation network G that training obtains.
Step 6: repeat step 3~step 5, makes to generate spectrum and gradually level off to Initial experiments spectrum.Interactive training
When number is 180000 times, generates spectrum and Initial experiments spectrum is as shown in Figure 7.
Step 7: to spectrum and the progress similarity judgement of Initial experiments spectrum is generated.
Cluster point is carried out to true experimental data and emulation data using two kinds of unsupervised learning methods of PCA, K-means
Analysis, judges whether simulated spectra and experimental spectrum mix as one kind, cannot be distinguished.The Raman of typical hazard product high explosive CL-20
It is as shown in Figure 8 using the clustering figure of unsupervised learning method PCA with original spectrum that spectrum generates spectrum;Typical hazard product are high
The Raman spectrum of bursting charge CL-20 generates spectrum and original spectrum uses the clustering figure of unsupervised learning method K-means
As shown in Figure 9.
Analyzing result proves that generation spectrum is not only intuitively apparently extremely similar with Initial experiments spectrum, by typical non-
Supervised Clustering Methods PCA and K-means can not distinguish the two, and generating spectrum can be used for constructing database, complete
Database extension, to include but is not limited to that the supervised learnings such as SVM method carries out modeling and provides sufficient Raman spectroscopic data.
Claims (5)
1. a kind of based on the spectra database extended method for generating confrontation network, it is characterised in that: specific step is as follows:
1) spectrum data gathering is carried out.Spectra collection experimental provision is built, a small amount of spectroscopic data of determinand is acquired;
2) building generates confrontation network (including generate network G and differentiate network D);
3) network model D is differentiated using generation spectrum and experimental spectrum building;
4) training parameter for sharing D network is trained to network model G is generated;
5) the new generation network G obtained according to training, generates new simulated spectra;
6) step 3-5 is repeated, makes to generate spectrum and gradually levels off to Initial experiments spectrum;
7) similarity judgement is carried out to generation spectrum and Initial experiments spectrum.
2. according to claim 1 a kind of based on the spectra database extended method for generating confrontation network, it is characterised in that:
The data base extension method that is itd is proposed while being suitable near infrared spectrum (Near Infrared Spectroscopy), Raman light
Compose (Raman Spectroscopy), laser induced breakdown spectroscopy (Laser-induced breakdown spectroscopy,
Abbreviation LIBS), fluorescence spectrum (Fluorescence Spectroscopy) and tera-hertz spectra (Terahertz
Spectroscopy) etc., the spectrum for adapting to every kind of substance of detection to above method can carry out analogue simulation, be to measurement
System and sample are not particularly limited.
3. according to claim 1 a kind of based on the spectra database extended method for generating confrontation network, it is characterised in that:
The step 2)~6) in generate network G and differentiate network D shared parameter interactive training model comprising the following specific steps
1) construction generate network model G, generate simulated spectra for the first time, according to used Initial experiments spectrum (LIBS spectrum or
Raman spectrum), generate the specified mean value of identical dimensional and the normal distribution random number sequence of variance;
2) differentiate that network model D, D are two Classification Neurals using artificial nerve network model construction, layer is at least set
Number is 3 layers, and the last layer is output layer, and output layer number of nodes is 1, using D network to Initial experiments spectrum and generate spectrum into
Initial experiments spectrum and generation spectrum are considered as inhomogeneity sample, obtain the training parameter of D network by row discriminant analysis;
3) training parameter for sharing D network is trained to network model G is generated, at this time by Initial experiments spectrum and generation light
Spectrum is considered as similar sample;
4) the new generation network G obtained according to training generates new simulated spectra, repeats above step, with interactive training
Number increases, and generates spectrum and gradually levels off to Initial experiments spectrum.
4. existing according to claim 1 with a kind of spectra database extended method based on generation confrontation network, feature described in 3
Meet following training condition during generating the confrontation network optimization:
1) differentiate the loss function setting of network D are as follows:
2) the loss function setting of network G is generated are as follows:
3) complex optimum objective function are as follows:
Wherein D (x) is the output for differentiating network on original real spectrum data set, and x~pdata is original real spectrum data
The probability distribution of collection, D (G (z)) are to differentiate that network is generating the output on the simulated spectra data set that network generates, z~pz (z)
The spectroscopic data collection probability distribution of network simulation is made a living into, z is random vector, EX~pdata(x) [logD (x)] indicates to differentiate network
True experiment gained spectrum and emulation gained spectroscopic data, E can be distinguishedZ~pz (z)[log (1-D (G (z)))] indicates to generate net
The simulated spectra data for differentiating that network is indistinguishable can be generated in network.
5. it is according to claim 1 a kind of based on the spectra database extended method for generating confrontation network, it is right in step 7)
Generate spectrum and Initial experiments spectrum and carry out similarity judgement and specifically refer to for the simulated spectra data generated, using but it is unlimited
True experimental data and emulation data are divided in unsupervised learning method (such as PCA, K-means clustering method etc.) etc.
Analysis, determines when simulated spectra and experimental spectrum has met similarity requirement, cannot be distinguished.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811228596.5A CN109508647A (en) | 2018-10-22 | 2018-10-22 | A kind of spectra database extended method based on generation confrontation network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201811228596.5A CN109508647A (en) | 2018-10-22 | 2018-10-22 | A kind of spectra database extended method based on generation confrontation network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109508647A true CN109508647A (en) | 2019-03-22 |
Family
ID=65745868
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201811228596.5A Pending CN109508647A (en) | 2018-10-22 | 2018-10-22 | A kind of spectra database extended method based on generation confrontation network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109508647A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110070124A (en) * | 2019-04-15 | 2019-07-30 | 广州小鹏汽车科技有限公司 | A kind of image amplification method and system based on production confrontation network |
CN110163179A (en) * | 2019-05-29 | 2019-08-23 | 浙江啄云智能科技有限公司 | A kind of contraband detecting recognition methods, system, equipment and its storage medium based on deep learning |
CN111126452A (en) * | 2019-12-03 | 2020-05-08 | 中国科学院国家空间科学中心 | Ground feature spectral curve expansion method and system based on principal component analysis |
CN111597751A (en) * | 2020-03-24 | 2020-08-28 | 自然资源部第一海洋研究所 | Crude oil film absolute thickness inversion method based on self-expansion depth confidence network |
CN112712857A (en) * | 2020-12-08 | 2021-04-27 | 北京信息科技大学 | Method for generating biological Raman spectrum data based on WGAN (WGAN) antagonistic generation network |
CN113378680A (en) * | 2021-06-01 | 2021-09-10 | 厦门大学 | Intelligent database building method for Raman spectrum data |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106845548A (en) * | 2017-01-24 | 2017-06-13 | 清华大学 | Model transfer method based on high-spectral data |
CN106997380A (en) * | 2017-03-21 | 2017-08-01 | 北京工业大学 | Imaging spectrum safe retrieving method based on DCGAN depth networks |
CN107016406A (en) * | 2017-02-24 | 2017-08-04 | 中国科学院合肥物质科学研究院 | The pest and disease damage image generating method of network is resisted based on production |
CN107220600A (en) * | 2017-05-17 | 2017-09-29 | 清华大学深圳研究生院 | A kind of Picture Generation Method and generation confrontation network based on deep learning |
CN107392255A (en) * | 2017-07-31 | 2017-11-24 | 深圳先进技术研究院 | Generation method, device, computing device and the storage medium of minority class picture sample |
CN107506770A (en) * | 2017-08-17 | 2017-12-22 | 湖州师范学院 | Diabetic retinopathy eye-ground photography standard picture generation method |
CN107563355A (en) * | 2017-09-28 | 2018-01-09 | 哈尔滨工程大学 | Hyperspectral abnormity detection method based on generation confrontation network |
US20180218497A1 (en) * | 2017-01-27 | 2018-08-02 | Arterys Inc. | Automated segmentation utilizing fully convolutional networks |
CN108399625A (en) * | 2018-02-28 | 2018-08-14 | 电子科技大学 | A kind of SAR image orientation generation method generating confrontation network based on depth convolution |
CN108491809A (en) * | 2018-03-28 | 2018-09-04 | 百度在线网络技术(北京)有限公司 | The method and apparatus for generating model for generating near-infrared image |
CN108537742A (en) * | 2018-03-09 | 2018-09-14 | 天津大学 | A kind of panchromatic sharpening method of remote sensing images based on generation confrontation network |
CN108564039A (en) * | 2018-04-16 | 2018-09-21 | 北京工业大学 | A kind of epileptic seizure prediction method generating confrontation network based on semi-supervised deep layer |
-
2018
- 2018-10-22 CN CN201811228596.5A patent/CN109508647A/en active Pending
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106845548A (en) * | 2017-01-24 | 2017-06-13 | 清华大学 | Model transfer method based on high-spectral data |
US20180218497A1 (en) * | 2017-01-27 | 2018-08-02 | Arterys Inc. | Automated segmentation utilizing fully convolutional networks |
CN107016406A (en) * | 2017-02-24 | 2017-08-04 | 中国科学院合肥物质科学研究院 | The pest and disease damage image generating method of network is resisted based on production |
CN106997380A (en) * | 2017-03-21 | 2017-08-01 | 北京工业大学 | Imaging spectrum safe retrieving method based on DCGAN depth networks |
CN107220600A (en) * | 2017-05-17 | 2017-09-29 | 清华大学深圳研究生院 | A kind of Picture Generation Method and generation confrontation network based on deep learning |
CN107392255A (en) * | 2017-07-31 | 2017-11-24 | 深圳先进技术研究院 | Generation method, device, computing device and the storage medium of minority class picture sample |
CN107506770A (en) * | 2017-08-17 | 2017-12-22 | 湖州师范学院 | Diabetic retinopathy eye-ground photography standard picture generation method |
CN107563355A (en) * | 2017-09-28 | 2018-01-09 | 哈尔滨工程大学 | Hyperspectral abnormity detection method based on generation confrontation network |
CN108399625A (en) * | 2018-02-28 | 2018-08-14 | 电子科技大学 | A kind of SAR image orientation generation method generating confrontation network based on depth convolution |
CN108537742A (en) * | 2018-03-09 | 2018-09-14 | 天津大学 | A kind of panchromatic sharpening method of remote sensing images based on generation confrontation network |
CN108491809A (en) * | 2018-03-28 | 2018-09-04 | 百度在线网络技术(北京)有限公司 | The method and apparatus for generating model for generating near-infrared image |
CN108564039A (en) * | 2018-04-16 | 2018-09-21 | 北京工业大学 | A kind of epileptic seizure prediction method generating confrontation network based on semi-supervised deep layer |
Non-Patent Citations (2)
Title |
---|
ANTREAS ANTONIOU 等: "DATA AUGMENTATION GENERATIVE ADVERSARIAL NETWORKS", 《ARXIV:1711.04340V3》 * |
IAN J. GOODFELLOW 等: "Generative Adversarial Nets", 《ARXIV:1406.2661V1》 * |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110070124A (en) * | 2019-04-15 | 2019-07-30 | 广州小鹏汽车科技有限公司 | A kind of image amplification method and system based on production confrontation network |
CN110163179A (en) * | 2019-05-29 | 2019-08-23 | 浙江啄云智能科技有限公司 | A kind of contraband detecting recognition methods, system, equipment and its storage medium based on deep learning |
CN111126452A (en) * | 2019-12-03 | 2020-05-08 | 中国科学院国家空间科学中心 | Ground feature spectral curve expansion method and system based on principal component analysis |
CN111126452B (en) * | 2019-12-03 | 2023-05-23 | 中国科学院国家空间科学中心 | Feature spectrum curve expansion method and system based on principal component analysis |
CN111597751A (en) * | 2020-03-24 | 2020-08-28 | 自然资源部第一海洋研究所 | Crude oil film absolute thickness inversion method based on self-expansion depth confidence network |
CN111597751B (en) * | 2020-03-24 | 2023-10-24 | 自然资源部第一海洋研究所 | Crude oil film absolute thickness inversion method based on self-expanding depth confidence network |
CN112712857A (en) * | 2020-12-08 | 2021-04-27 | 北京信息科技大学 | Method for generating biological Raman spectrum data based on WGAN (WGAN) antagonistic generation network |
CN113378680A (en) * | 2021-06-01 | 2021-09-10 | 厦门大学 | Intelligent database building method for Raman spectrum data |
CN113378680B (en) * | 2021-06-01 | 2022-06-28 | 厦门大学 | Intelligent database building method for Raman spectrum data |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109508647A (en) | A kind of spectra database extended method based on generation confrontation network | |
Muro et al. | Forensic body fluid identification and differentiation by Raman spectroscopy | |
Chen et al. | Exploration research on the fusion of multimodal spectrum technology to improve performance of rapid diagnosis scheme for thyroid dysfunction | |
JP7528087B2 (en) | Method and system for spectrally resolving fluorescent materials in a sample - Patents.com | |
US8452716B2 (en) | Kernel-based method and apparatus for classifying materials or chemicals and for quantifying the properties of materials or chemicals in mixtures using spectroscopic data | |
Maltsev | Scanning flow cytometry for individual particle analysis | |
Zimmerman et al. | Analysis of allergenic pollen by FTIR microspectroscopy | |
CN111707656B (en) | Cerebrospinal fluid cell detection method and system based on Raman scattering spectrum | |
JP2010520471A (en) | Ensemble method and apparatus for classification of materials and quantification of components of mixtures | |
JP2015534078A (en) | System, apparatus and method employing angular dispersive scattering and spectrally resolved measurements for object classification | |
JP2007315941A (en) | Plant variety determination system, method, and program | |
EP3290908A1 (en) | Unknown sample determining method, unknown sample determining instrument, and unknown sample determining program | |
CN109187443A (en) | Water body bacterial micro-organism based on multi-wavelength transmitted spectrum accurately identifies method | |
Teng et al. | Predictive data clustering of laser-induced breakdown spectroscopy for brain tumor analysis | |
CN107402192A (en) | A kind of method of quick analysis essence and flavoring agent quality stability | |
CN109470686A (en) | A kind of laser induced breakdown spectroscopy classifying identification method based on semi-supervised learning | |
CN105069462A (en) | Qualitative determination method for organic chemical based on spectral characteristic refinement and classifier cascading | |
CN104374737A (en) | Near-infrared quantitative identification method | |
Qi et al. | Composition analysis of white mineral pigment based on convolutional neural network and Raman spectrum | |
US20220091017A1 (en) | Methods for continuous measurement of baseline noise in a flow cytometer and systems for same | |
Amirvaresi et al. | Multiway data analysis applied to miniaturized spectroscopy | |
CN110110842A (en) | A kind of Portable fluorescence spectrum quickly detects the method for building up of artificial neural network intelligent distinguishing model | |
EP4012386B1 (en) | Identifying 3d objects | |
Wang et al. | A Virtual Impactor-Based Portable Optical Sensor With Probabilistic Neural Network: For Particle Property Identification | |
Wang et al. | Determination of the carmine content based on spectrum fluorescence spectral and PSO-SVM |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190322 |