CN108805187A - Celestial spectrum sequence automatic classification system and method - Google Patents
Celestial spectrum sequence automatic classification system and method Download PDFInfo
- Publication number
- CN108805187A CN108805187A CN201810531003.6A CN201810531003A CN108805187A CN 108805187 A CN108805187 A CN 108805187A CN 201810531003 A CN201810531003 A CN 201810531003A CN 108805187 A CN108805187 A CN 108805187A
- Authority
- CN
- China
- Prior art keywords
- spectrum sequence
- neural network
- celestial
- celestial spectrum
- network model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/192—Recognition using electronic means using simultaneous comparisons or correlations of the image signals with a plurality of references
- G06V30/194—References adjustable by an adaptive method, e.g. learning
Abstract
The invention discloses a kind of celestial spectrum sequence automatic classification system and methods, wherein celestial spectrum sequence automatic classification system includes:First unit obtains history celestial spectrum sequence as training sample;Second unit builds neural network model;Third unit obtains celestial spectrum sequence to be sorted;Wherein, the second unit is trained neural network model with the training sample, obtains disaggregated model;The second unit is by disaggregated model described in the celestial spectrum sequence inputting to be sorted, to obtain the classification results of the disaggregated model output.Classification speed can be improved in the present invention.
Description
Technical field
The present invention relates to astronomy technical field, more particularly to a kind of celestial spectrum sequence automatic classification system and method.
Background technology
The present invention belongs to the relevant technologies related to the present invention for the description of background technology, be only used for explanation and just
In the invention content for understanding the present invention, it should not be construed as applicant and be specifically identified to or estimate applicant being considered of the invention for the first time
The prior art for the applying date filed an application.
Astronomy is a time-honored observation science, is a most basic task of astronomy to celestial body classification.With
The development of science and technology, observation device constantly upgrades, and the mankind gradually extend the understanding in universe from the near to the remote, from the earth to too
Sun system from fixed star to the milky way galaxy, then arrives extragalactic system.Advanced observation device enables us to hope to universe deeper inside, simultaneously
Also chronometer data explosive growth is brought.Such as LAMOST telescopes each observe night and can acquire more than ten thousand spectrum, make
It is traditional it is artificial or semi-artificial cannot be coped with very well in the way of template matches, need efficient and accurate celestial body spectrum
Intelligent recognition sorting algorithm.
Invention content
In view of this, a kind of celestial spectrum sequence automatic classification system of offer of the embodiment of the present invention and method, main purpose
It is to improve celestial spectrum sequence classification speed.
In order to achieve the above objectives, present invention generally provides following technical solutions:
In a first aspect, an embodiment of the present invention provides a kind of celestial spectrum sequence automatic classification systems, including:
First unit obtains history celestial spectrum sequence as training sample;
Second unit builds neural network model;
Third unit obtains celestial spectrum sequence to be sorted;
Wherein, the second unit is trained neural network model with the training sample, obtains disaggregated model;
The second unit is by disaggregated model described in the celestial spectrum sequence inputting to be sorted, to obtain the classification mould
The classification results of type output.
Preferably, the first unit includes:
Data acquisition module obtains the history celestial spectrum sequence;
Preprocessing module pre-processes the history celestial spectrum sequence of acquisition, and the pretreatment is, will be astronomical
Spectral sequence is standardized, the signal after being standardized, and the Gaussian noise that variance is particular size is superimposed upon standardization
On signal afterwards.
Preferably, the neural network model include multiple Inception-ResNet composite layers for setting gradually and
One resnet layers, wherein core (kernel) size of Inception modules is successively decreased as network is deepened.
Preferably, being added in each layer of the neural network model has Dropout.
Preferably, when the neural network model training, all kinds of weight of the celestial spectrum sequence is called together with all kinds of
The rate of returning is inversely proportional.
Preferably, further including amending unit;The second unit carries out neural network model with the training sample
Training, obtains multiple disaggregated models, and the second unit is by the multiple classification moulds of celestial spectrum sequence inputting to be sorted
Type obtains multiple classification results, and the amending unit is modified multiple classification results using Kalman filtering thought, corrects
Classification results afterwards are as final classification result.
Preferably, the amendment is specific as follows:An infrastructure is selected first in multiple classification results, with
Initial infrastructure is modified according to other classification results afterwards, the modified formula is as follows,
Xhat=x+ry/rx*y, wherein xhat is correction result, based on x as a result, based on rx result recall rate,
Y is another classification results, and ry is the recall rate of another classification results.
Second aspect, an embodiment of the present invention provides a kind of celestial spectrum sequence automatic classification methods, include the following steps:
History celestial spectrum sequence is obtained as training sample;
Neural network model is trained with the training sample, obtains disaggregated model;
Obtain celestial spectrum sequence to be sorted;
By disaggregated model described in the celestial spectrum sequence inputting to be sorted, to obtain the classification of the disaggregated model output
As a result.
Preferably, be as training sample, the pretreatment after the history celestial spectrum sequence is preprocessed, it will
Celestial spectrum sequence is standardized, the signal after being standardized, and the Gaussian noise that variance is particular size is superimposed upon mark
On signal after standardization.
Preferably, the neural network model include multiple Inception-ResNet composite layers for setting gradually and
One resnet layers, the wherein core size of Inception modules is successively decreased as network is deepened.
Preferably, being added in each layer of the neural network model has Dropout.
Preferably, when the neural network model training, all kinds of weight of the celestial spectrum sequence is called together with all kinds of
The rate of returning is inversely proportional.
Preferably, being trained to neural network model with the training sample, multiple disaggregated models are obtained, it will be described
The multiple disaggregated models of celestial spectrum sequence inputting to be sorted, obtain multiple classification results, to multiple classification results using card
Kalman Filtering thought is modified, and revised classification results are as final classification result.
Preferably, the amendment is specific as follows:An infrastructure is selected first in multiple classification results, with
Initial infrastructure is modified according to other classification results afterwards, the modified formula is as follows,
Xhat=x+ry/rx*y, wherein xhat is correction result, based on x as a result, based on rx result recall rate,
Y is another classification results, and ry is the recall rate of another classification results.
The third aspect, an embodiment of the present invention provides a kind of computer readable storage mediums, are stored thereon with computer journey
The step of sequence, which realizes above-mentioned method when being executed by processor.
Fourth aspect an embodiment of the present invention provides a kind of computer equipment, including memory, processor and is stored in
On reservoir and the computer program that can run on a processor, the processor realize above-mentioned method when executing described program
Step.
Compared with prior art, the advantageous effect of the embodiment of the present invention is:
Celestial spectrum sequence automatic classification system provided in an embodiment of the present invention and method are based on deep learning, in conjunction with history
Grouped data establishes model and has successfully trained disaggregated model, can sort out possible valuable spectral information automatically, carry
High-class precision.Assorting process need not have professional knowledge of the present invention and correlation experience personnel.
Description of the drawings
Fig. 1 shows the schematic diagram of an embodiment of celestial spectrum sequence automatic classification system of the present invention.
Fig. 2 shows the schematic diagrames of another embodiment of celestial spectrum sequence automatic classification system of the present invention.
Fig. 3 shows the neural network structure schematic diagram of one embodiment of the invention.
Fig. 4 shows the flow chart of an embodiment of celestial spectrum sequence automatic classification method of the present invention.
Fig. 5 shows the flow chart of another embodiment of celestial spectrum sequence automatic classification method of the present invention.
Specific implementation mode
With reference to specific embodiment, present invention is further described in detail, but not as a limitation of the invention.?
In following the description, what different " embodiment " or " embodiment " referred to is not necessarily the same embodiment.In addition, one or more are implemented
Special characteristic, structure or feature in example can be combined by any suitable form.
Fig. 1 shows the schematic diagram of an embodiment of celestial spectrum sequence automatic classification system of the present invention.Fig. 2 shows this
The schematic diagram of another embodiment of invention celestial spectrum sequence automatic classification system.Referring to Fig. 1 and Fig. 2, the embodiment of the present invention
Celestial spectrum sequence automatic classification system, including:
First unit 10 obtains history celestial spectrum sequence as training sample;
Second unit 20, builds neural network model;
Third unit 30 obtains celestial spectrum sequence to be sorted;
Wherein, second unit 20 is trained acquisition disaggregated model with training sample to neural network model;
Second unit 20 is by celestial spectrum sequence inputting disaggregated model to be sorted, to obtain the classification knot of disaggregated model output
Fruit.
Celestial spectrum sequence automatic classification system provided in an embodiment of the present invention is based on deep learning, in conjunction with history classification number
According to, establish model and successfully trained disaggregated model, can sort out automatically may valuable spectral information, improve classification
Precision.Assorting process need not have professional knowledge of the present invention and correlation experience personnel.
In one embodiment of the invention, first unit includes data acquisition module 11 and preprocessing module 12, and wherein data obtain
Modulus block 11 obtains history celestial spectrum sequence;Preprocessing module 12 locates the history celestial spectrum sequence of acquisition in advance
Reason, it is to be standardized celestial spectrum sequence to pre-process, the signal after being standardized, by the height that variance is particular size
This noise is superimposed upon on the signal after standardization.In the present embodiment, for sequence data, it is controllable to devise a kind of new signal-to-noise ratio
Data enhancement method, sequence data is subjected to conventional Mean-Variance first and is standardized, the mean value of fixed single sample sequence
It is zero, variance 1 can generate the Gaussian noise signal that variance is particular size and be superimposed upon on the signal after standardization at this time.
The variance D (X+Y) of two mutually independent random variables X and Y=D (X)+D (Y) in principle.Signal to Noise Ratio (SNR)=D (X)/D
(Y), it is assumed that Y is noise, since we are by the normalized square mean of original signal to 1, so, by the variance for controlling noise
Size can control signal-to-noise ratio, can effectively enhance in the much noises of addition and give quantitative target in data volume.Pass through processing
Front and back all kinds of spectrograms compare the ordinate generation significant change it is found that spectrogram, each value such as shape of original spectrum
Between relative information retain.
Fig. 3 shows the neural network structure schematic diagram of one embodiment of the invention.Referring to Fig. 3, one embodiment of the invention
In, neural network model includes the multiple Inception-ResNet composite layers set gradually and one resnet layers, wherein
The core size of Inception modules is successively decreased as network is deepened.Multiple Inception-ResNet composite layers are formed
Structure stacked Inception and ResNet.In the embodiment of the present invention, the sorter network of NN Model Reference classics
The thought of ResNet and Inception devises the neural network classified dedicated for celestial spectrum.The present embodiment ties the two
It is specifically applied to spectral classification altogether, the powerful feature extraction ability of neural network model is utilized, compares traditional support
Vector machine scheduling algorithm is not required to manual features extraction, is classified by learning characteristic.The core size of Inception modules is with net
Network is deepened and successively decreases and be conducive to model convergence.Inception-ResNet composite layers can be 3-6.Referring to Fig. 3, such as god
Can be 4 Inception-ResNet composite layers and one ResNet layers through network model.
Being added in one embodiment of the invention, in each layer of neural network model has Dropout.The present embodiment passes through analysis
Data cases adjust model parameter and reach ideal classifying quality meticulously.The generalization ability of model is improved, training result is more steady
It is fixed, solve overfitting problem.
In one embodiment of the invention, when neural network model is trained, all kinds of weight of celestial spectrum sequence is called together with all kinds of
The rate of returning is inversely proportional.The institute in training of the sample according to each classification of performance adjustment on verification collection is needed during model training
The weight ratio accounted for.Celestial spectrum sequence can be divided into for fixed star, galaxy, quasar and other four major class, and Various types of data is uneven,
Such as fixed star class data are when having 100,000, quasar class data about 10,000 or so.Therefore, the present embodiment is according to celestial spectrum number
According to the characteristics of, all kinds of weights used by using and verification concentrate mode that the recall rate of each classification is inversely proportional targetedly
Training pattern so that model reaches more excellent result.
In one embodiment of the present of invention, referring to Fig. 2, the categorizing system of the present embodiment further includes amending unit 40;Second
Unit 20 is trained neural network model with training sample, obtains multiple disaggregated models, and second unit 20 is by day to be sorted
Literary spectral sequence inputs multiple disaggregated models, obtains multiple classification results, and amending unit 40 uses karr to multiple classification results
Graceful filter thought is modified, and revised classification results are as final classification result.The present embodiment is thought using Kalman filtering
Want to carry out modified result, an infrastructure is selected first in multiple classification results, then according to point of other models output
Class result is modified initial infrastructure, can more accurately obtain final classification result in this way.The present invention is implemented
In example, it is assumed that multiple prediction results have respectively different distribution, by probability space weighted superposition can obtain it is more smart
True model.Specific correction formula is as follows:Two prediction results x, y are chosen, they have respective recall rate rx, ry.Selected x
Based on as a result, result x_hat=x+ry/rx*y after being so modified to x with y.So there is one group of prediction result X=
{ x1, x2 ..., xn } and their corresponding recall rate R={ r1, r2 ..., rn }.In this way, using method said before into
Row, which is corrected, can obtain better final classification result.
Second aspect, an embodiment of the present invention provides a kind of celestial spectrum sequence automatic classification method,
Fig. 4 shows the flow chart of an embodiment of celestial spectrum sequence automatic classification method of the present invention.
Fig. 5 shows the flow chart of another embodiment of celestial spectrum sequence automatic classification method of the present invention.Referring to Fig. 4 and
Fig. 5, the celestial spectrum sequence automatic classification method include the following steps:
History celestial spectrum sequence is obtained as training sample;
Neural network model is trained with training sample, obtains disaggregated model;
Obtain celestial spectrum sequence to be sorted;
By celestial spectrum sequence inputting disaggregated model to be sorted, to obtain the classification results of disaggregated model output.
Celestial spectrum sequence automatic classification system provided in an embodiment of the present invention is based on deep learning, in conjunction with history classification number
According to establishing model and successfully trained disaggregated model, using the powerful calculation power of GPU, can sort out automatically may be valuable
Spectral information, improve nicety of grading.The method of the present invention can complete the classification of 100,000 celestial spectrum information, nothing in several seconds
Any artificial participation is needed, the burden of astronomer is greatly mitigated.Reach simultaneously all kinds of average recall rates 0.8 or more it is excellent
Elegant performance.
In one embodiment of the invention, it is used as training sample after history celestial spectrum sequence is preprocessed, it is to incite somebody to action to pre-process
Celestial spectrum sequence is standardized, the signal after being standardized, and the Gaussian noise that variance is particular size is superimposed upon mark
On signal after standardization.In the present embodiment, for sequence data, a kind of data enhancing side that new signal-to-noise ratio is controllable is devised
Sequence data is carried out conventional Mean-Variance first and standardized by formula, and the mean value of fixed single sample sequence is zero, and variance is
1, the Gaussian noise signal that variance is particular size can be generated at this time to be superimposed upon on the signal after standardization.Two phases in principle
Variance D (X+Y)=D (X)+D (Y) of mutual independent stochastic variable X and Y.Signal to Noise Ratio (SNR)=D (X)/D (Y), it is assumed that Y is to make an uproar
Sound, since we are by the normalized square mean of original signal to 1, so, the variance size by controlling noise can control
Signal-to-noise ratio can effectively enhance in the much noises of addition and give quantitative target in data volume.Pass through all kinds of light before and after the processing
Spectrogram compares it is found that significant change, the relative information between each value such as shape of original spectrum occur for the ordinate of spectrogram
Retain.
Fig. 3 shows the neural network structure schematic diagram of one embodiment of the invention.Referring to Fig. 3, one embodiment of the invention
In, neural network model includes the multiple Inception-ResNet composite layers set gradually and one resnet layers, wherein
The core size of Inception modules is successively decreased as network is deepened.Multiple Inception-ResNet composite layers are formed
Structure stacked Inception and ResNet.In the embodiment of the present invention, the sorter network of NN Model Reference classics
The thought of ResNet and Inception devises the neural network classified dedicated for celestial spectrum.The present embodiment ties the two
It is specifically applied to spectral classification altogether, the powerful feature extraction ability of neural network model is utilized, compares traditional support
Vector machine scheduling algorithm is not required to manual features extraction, is classified by learning characteristic.The core size of Inception modules is with net
Network is deepened and successively decreases and be conducive to model convergence.Inception-ResNet composite layers can be 3-6.Referring to Fig. 3, such as god
Can be 4 Inception-ResNet composite layers and one ResNet layers through network model.
Being added in one embodiment of the invention, in each layer of neural network model has Dropout.The present embodiment passes through analysis
Data cases adjust model parameter and reach ideal classifying quality meticulously.The generalization ability of model is improved, training result is more steady
It is fixed, solve overfitting problem.
In one embodiment of the invention, when neural network model is trained, all kinds of weight of celestial spectrum sequence is called together with all kinds of
The rate of returning is inversely proportional.The institute in training of the sample according to each classification of performance adjustment on verification collection is needed during model training
The weight ratio accounted for.Celestial spectrum sequence can be divided into for fixed star, galaxy, quasar and other four major class, and Various types of data is uneven,
Such as fixed star class data are when having 100,000, quasar class data about 10,000 or so.Therefore, the present embodiment is according to celestial spectrum number
According to the characteristics of, all kinds of weights used by using and verification concentrate mode that the recall rate of each classification is inversely proportional targetedly
Training pattern so that model reaches more excellent result.
Referring to Fig. 5, in one embodiment of the present of invention, neural network model is trained with training sample, is obtained more
The multiple disaggregated models of celestial spectrum sequence inputting to be sorted are obtained multiple classification results by a disaggregated model, are tied to multiple classification
Fruit is modified using Kalman filtering thought, and revised classification results are as final classification result.The present embodiment is using card
Kalman Filtering thought carries out modified result, an infrastructure is selected first in multiple classification results, then according to other moulds
The classification results of type output are modified initial infrastructure, can more accurately obtain final classification result in this way.
In the embodiment of the present invention, it is assumed that multiple prediction results have respectively different distribution, can be with by the weighted superposition in probability space
Obtain more accurate model.Specific correction formula is as follows:Two prediction results x, y are chosen, they have respective recall rate
rx,ry.As a result, result x_hat=x+ry/rx*y after being so modified to x with y based on selected x.So there is one group
Prediction result X={ x1, x2 ..., xn } and their corresponding recall rate R={ r1, r2 ..., rn }.In this way, utilizing front institute
The method said, which is modified, can obtain better final classification result.
The third aspect, an embodiment of the present invention provides a kind of computer readable storage mediums, are stored thereon with computer journey
The step of sequence, which realizes above-mentioned method when being executed by processor.
Fourth aspect an embodiment of the present invention provides a kind of computer equipment, including memory, processor and is stored in
On reservoir and the computer program that can run on a processor, the processor realize above-mentioned method when executing described program
Step.
Those skilled in the art can be understood that the embodiment of the present invention technical solution can by software and/or
Hardware is realized." unit " or " unit " in this specification is to refer to complete independently or completed with other component cooperation specific
The software and/or hardware of function, wherein hardware for example can be that (Field-Programmable Gate Array show FPGA
Field programmable gate array), IC (Integrated Circuit, integrated circuit) etc..
The embodiment of the present invention additionally provides a kind of computer readable storage medium, is stored thereon with computer program, the journey
The step of any of the above-described embodiment method is realized when sequence is executed by processor.Wherein, computer readable storage medium may include
But be not limited to any kind of disk, including floppy disk, CD, DVD, CD-ROM, mini drive and magneto-optic disk, ROM, RAM,
EPROM, EEPROM, DRAM, VRAM, flash memory device, magnetic or optical card, nanosystems (including molecular memory IC),
Or it is suitable for any kind of medium or equipment of store instruction and/or data.
The embodiment of the present invention additionally provides a kind of computer equipment, including memory, processor and storage are on a memory
And the computer program that can be run on a processor, the step of realizing any of the above-described embodiment method when processor executes program.
In embodiments of the present invention, processor is the control centre of computer system, can be the processor of physical machine, can also be void
The processor of quasi- machine.
In the present invention, term " first ", " second " etc. are only used for the purpose of description, are not understood to indicate or imply
Relative importance or sequence;Term " multiple " then refers to two or more, unless otherwise restricted clearly.Term " installation ",
The terms such as " connected ", " connection ", " fixation " shall be understood in a broad sense, for example, " connection " may be a fixed connection, can also be can
Dismantling connection, or be integrally connected;" connected " can be directly connected, can also be indirectly connected through an intermediary.For this
For the those of ordinary skill in field, the specific meanings of the above terms in the present invention can be understood according to specific conditions.
In description of the invention, it is to be understood that the orientation or positional relationship of the instructions such as term "upper", "lower" be based on
Orientation or positional relationship shown in the drawings, is merely for convenience of description of the present invention and simplification of the description, rather than indicates or imply institute
The device or unit of finger must have specific direction, with specific azimuth configuration and operation, it is thus impossible to be interpreted as to this hair
Bright limitation.
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can easily think of the change or the replacement, and should all contain
Lid is within protection scope of the present invention.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (16)
1. celestial spectrum sequence automatic classification system, including:
First unit obtains history celestial spectrum sequence as training sample;
Second unit builds neural network model;
Third unit obtains celestial spectrum sequence to be sorted;
Wherein, the second unit is trained neural network model with the training sample, obtains disaggregated model;
The second unit is defeated to obtain the disaggregated model by disaggregated model described in the celestial spectrum sequence inputting to be sorted
The classification results gone out.
2. system according to claim 1, which is characterized in that the first unit includes:
Data acquisition module obtains the history celestial spectrum sequence;
Preprocessing module pre-processes the history celestial spectrum sequence of acquisition, and the pretreatment is, by celestial spectrum
Sequence is standardized, the signal after being standardized, after the Gaussian noise that variance is particular size is superimposed upon standardization
On signal.
3. system according to claim 1, which is characterized in that the neural network model include set gradually it is multiple
Inception-ResNet composite layers and one resnet layers, wherein the core size of Inception modules deepen with network and
Successively decrease.
4. system according to claim 3, which is characterized in that being added in each layer of the neural network model has
Dropout。
5. system according to claim 1, which is characterized in that when the neural network model training, the celestial spectrum
All kinds of weight of sequence is inversely proportional with all kinds of recall rates.
6. system according to claim 1, which is characterized in that further include amending unit;The second unit is with the instruction
Practice sample to be trained neural network model, obtain multiple disaggregated models, the second unit will the astronomy light to be sorted
Spectral sequence inputs multiple disaggregated models, obtains multiple classification results, the amending unit adopts the multiple classification results
It is modified with Kalman filtering thought, revised classification results are as final classification result.
7. system according to claim 6, which is characterized in that the amendment is specific as follows:First in multiple classification
As a result an infrastructure is selected in, then initial infrastructure is modified according to other classification results, the amendment
Formula it is as follows,
Xhat=x+ry/rx*y, wherein xhat is correction result, based on x as a result, based on rx result recall rate, y is
Another classification results, ry are the recall rate of another classification results.
8. celestial spectrum sequence automatic classification method, includes the following steps:
History celestial spectrum sequence is obtained as training sample;
Neural network model is trained with the training sample, obtains disaggregated model;
Obtain celestial spectrum sequence to be sorted;
By disaggregated model described in the celestial spectrum sequence inputting to be sorted, to obtain the classification knot of the disaggregated model output
Fruit.
9. according to the method described in claim 8, it is characterized in that, conduct after the history celestial spectrum sequence is preprocessed
Training sample, the pretreatment are to be standardized celestial spectrum sequence, the signal after being standardized, and are spy by variance
The Gaussian noise for determining size is superimposed upon on the signal after standardization.
10. according to the method described in claim 8, it is characterized in that, the neural network model include set gradually it is multiple
Inception-ResNet composite layers and one resnet layers, wherein the core size of Inception modules deepen with network and
Successively decrease.
11. according to the method described in claim 10, it is characterized in that, being added in each layer of the neural network model has
Dropout。
12. according to the method described in claim 8, it is characterized in that, the neural network model train when, the celestial spectrum
All kinds of weight of sequence is inversely proportional with all kinds of recall rates.
13. according to the method described in claim 8, it is characterized in that, being instructed to neural network model with the training sample
Practice, obtains multiple disaggregated models, by the multiple disaggregated models of celestial spectrum sequence inputting to be sorted, obtain multiple classification
As a result, being modified using Kalman filtering thought to the multiple classification results, revised classification results are as final point
Class result.
14. according to the method for claim 13, which is characterized in that the amendment is specific as follows:First at multiple described points
An infrastructure is selected in class result, and then initial infrastructure is modified according to other classification results, it is described to repair
Positive formula is as follows,
Xhat=x+ry/rx*y, wherein xhat is correction result, based on x as a result, based on rx result recall rate, y is
Another classification results, ry are the recall rate of another classification results.
15. a kind of computer readable storage medium, is stored thereon with computer program, power is realized when which is executed by processor
Profit requires the step of any one of 8-14 the methods.
16. a kind of computer equipment, including memory, processor and storage are on a memory and the meter that can run on a processor
The step of calculation machine program, the processor realizes any one of claim 8-14 the methods when executing described program.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810531003.6A CN108805187B (en) | 2018-05-29 | 2018-05-29 | Astronomical spectrum sequence automatic classification system and method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810531003.6A CN108805187B (en) | 2018-05-29 | 2018-05-29 | Astronomical spectrum sequence automatic classification system and method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108805187A true CN108805187A (en) | 2018-11-13 |
CN108805187B CN108805187B (en) | 2022-07-19 |
Family
ID=64090676
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810531003.6A Active CN108805187B (en) | 2018-05-29 | 2018-05-29 | Astronomical spectrum sequence automatic classification system and method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108805187B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110543361A (en) * | 2019-07-29 | 2019-12-06 | 中国科学院国家天文台 | Astronomical data parallel processing device and method |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104978570A (en) * | 2015-06-25 | 2015-10-14 | 西北工业大学 | Incremental learning based method for detecting and identifying traffic sign in traveling video |
US20160290940A1 (en) * | 2015-04-01 | 2016-10-06 | Indiana University Research And Technology Corporation | System and method for chemically authenticating items |
CN106779079A (en) * | 2016-11-23 | 2017-05-31 | 北京师范大学 | A kind of forecasting system and method that state is grasped based on the knowledge point that multimodal data drives |
-
2018
- 2018-05-29 CN CN201810531003.6A patent/CN108805187B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20160290940A1 (en) * | 2015-04-01 | 2016-10-06 | Indiana University Research And Technology Corporation | System and method for chemically authenticating items |
CN104978570A (en) * | 2015-06-25 | 2015-10-14 | 西北工业大学 | Incremental learning based method for detecting and identifying traffic sign in traveling video |
CN106779079A (en) * | 2016-11-23 | 2017-05-31 | 北京师范大学 | A kind of forecasting system and method that state is grasped based on the knowledge point that multimodal data drives |
Non-Patent Citations (6)
Title |
---|
BC.PAVEL HÁLA: "Spectral classification using convolutional neural networks", 《COMPUTER VISION AND PATTERN RECOGNITION》 * |
G.S.马达拉等: "《金融中的统计方法》", 31 August 2008 * |
任光等: "并行多参考模型卡尔曼滤波系统仿真研究", 《系统仿真学报》 * |
伍雪冬: "《非线性时间序列在线预测建模与仿真》", 30 November 2015 * |
曹东旭: "基于神经网络的人脸识别系统设计与实现", 《中国优秀硕士学位论文全文数据库-信息科技辑》 * |
董志荣: "《舰船信息融合与目标运动分析》", 30 June 2016 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110543361A (en) * | 2019-07-29 | 2019-12-06 | 中国科学院国家天文台 | Astronomical data parallel processing device and method |
CN110543361B (en) * | 2019-07-29 | 2023-06-13 | 中国科学院国家天文台 | Astronomical data parallel processing device and astronomical data parallel processing method |
Also Published As
Publication number | Publication date |
---|---|
CN108805187B (en) | 2022-07-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102201236B (en) | Speaker recognition method combining Gaussian mixture model and quantum neural network | |
CN104239858B (en) | A kind of method and apparatus of face characteristic checking | |
CN111929748B (en) | Meteorological element forecasting method and system | |
CN104281858B (en) | Three dimensional convolution neural network training method, video accident detection method and device | |
CN110503187B (en) | Implementation method for generating countermeasure network model for generating functional nuclear magnetic resonance imaging data | |
CN106952649A (en) | Method for distinguishing speek person based on convolutional neural networks and spectrogram | |
CN112465040B (en) | Software defect prediction method based on class unbalance learning algorithm | |
CN102136024B (en) | Biometric feature identification performance assessment and diagnosis optimizing system | |
CN110298663A (en) | Based on the wide fraudulent trading detection method learnt deeply of sequence | |
CN108478216A (en) | A kind of epileptic seizure intelligent Forecasting early period based on convolutional neural networks | |
CN110245711A (en) | The SAR target identification method for generating network is rotated based on angle | |
CN108446214B (en) | DBN-based test case evolution generation method | |
CN109934269A (en) | A kind of opener recognition methods of electromagnetic signal and device | |
CN108345904A (en) | A kind of Ensemble Learning Algorithms of the unbalanced data based on the sampling of random susceptibility | |
CN109491914A (en) | Defect report prediction technique is influenced based on uneven learning strategy height | |
CN109784288A (en) | A kind of pedestrian's recognition methods again based on differentiation perception fusion | |
CN110289081A (en) | The epilepsy detection method of depth network stack model adaptation weighted feature fusion | |
CN106874959A (en) | A kind of multiple dimensioned scanning cascade forestry practises the training method of machine | |
CN104463221A (en) | Imbalance sample weighting method suitable for training of support vector machine | |
CN110490267A (en) | A kind of bill method for sorting based on deep learning | |
CN115859142A (en) | Small sample rolling bearing fault diagnosis method based on convolution transformer generation countermeasure network | |
CN109284662A (en) | A kind of transfer learning method towards the classification of underwater voice signal | |
CN108805187A (en) | Celestial spectrum sequence automatic classification system and method | |
CN109214413A (en) | A kind of determination method and system of data distribution balance | |
CN104463207A (en) | Knowledge self-encoding network and polarization SAR image terrain classification method thereof |
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 |