CN106203462A - Astronomicalc optics transition source quick automatic identification method based on machine learning and system - Google Patents
Astronomicalc optics transition source quick automatic identification method based on machine learning and system Download PDFInfo
- Publication number
- CN106203462A CN106203462A CN201610465133.5A CN201610465133A CN106203462A CN 106203462 A CN106203462 A CN 106203462A CN 201610465133 A CN201610465133 A CN 201610465133A CN 106203462 A CN106203462 A CN 106203462A
- Authority
- CN
- China
- Prior art keywords
- image
- transition source
- characteristic parameter
- star
- simulation observation
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
Astronomicalc optics transition source quick automatic identification method based on machine learning and system, the method includes: (1) builds transition source simulation observation image training sample by the method for emulation；(2) complete the extraction of characteristic parameter by transition source simulation observation image training sample and be trained automatic categorizer；(3) observed image and template image are subtracted each other obtain residual image；(4) residual image is carried out the extraction of point source and characteristic parameter；(5) automatic categorizer generated by step (2) obtains transition source candidate's body according to the characteristic parameter processed by step (4), the classification automatically using Algorithms for Automatic Classification based on random forest to carry out target.
Description
Technical field
The present invention relates to the quick automatic identification method in a kind of astronomicalc optics transition source, be applied to the astronomicalc optics of big visual field
Transition source search project.In the residual image that can obtain from image subtraction method, automatic distinguishing goes out noise and transition source, from
And realize from substantial amounts of noise spot, do not know out transition source.Image subtraction method is i.e. to be carried out with template image by observed image
Carry out subtracting each other obtaining residual image after flow and location matches.
Background technology
Transition source refers to a kind of accidental of short duration acyclic celestial body or chronometer phenomenon.Transition source is for research space
The origin of cosmos, the physical phenomenon of research extreme environment have great significance.The discovery (2011 of universe acceleration expansion phenomenon
Year Nobel Prize in physics) it is exactly based on the observational study to large sample transition source supernova and finds.Meanwhile, up-to-date report
Gravitational wave event, the corresponding body of its electromagnetic wave is also a class transition source (the most upon the look).
Owing to transition source is accidental astronomical events, it is desirable to the scope of transition source search has big visual field (i.e. unit
Geng great Tian district can be observed in time) and the feature of high Temporal sampling (i.e. the return visit observing frequency to district on the same day).I
Equipment ground wide angle camera battle array is searched in transition source during state builds, is made up of, relatively 36 a diameter of wide angle camera of 18 centimetres
Search for equipment in the transition source of international mainstream, it is individual that ground wide angle camera battle array equipment all will improve 1-2 on visual field and sample frequency
The order of magnitude.Big visual field and high Temporal sampling mean that the generation rate of data is bigger, in order to find the transition source requirement of short time scale
The speed that data process has requirement of real time.Therefore, data are processed and brings challenges, i.e. need to realize quick to big data
Process in real time.
The cardinal principle of classical transition source search is by being subtracted each other with template image by observed image, if one
Individual transition source (the most emerging source) is exactly an image being similar to complete point source in the residual image after having subtracted, and other
Incomplete image patch is then the noise in subtraction.Traditional method be by the method for eye recognition by transition source from residual plot
In pick out, for Modern Astronomical transition source search big generating date require be difficult to meet.
Summary of the invention
The problem that the technology of the present invention solves is: overcome the deficiencies in the prior art, it is provided that a kind of astronomy based on machine learning
Optics transition source quick automatic identification method.
The technical solution of the present invention is: fast automatic identification side, a kind of astronomicalc optics transition source based on machine learning
Method, the method includes:
(1) transition source simulation observation image training sample is built by the method for emulation；
(2) complete the extraction of characteristic parameter by transition source simulation observation image training sample and be trained automatically classifying
Device；
(3) observed image and template image are subtracted each other obtain residual image；
(4) residual image is carried out the extraction of point source and characteristic parameter；
(5) automatic categorizer generated by step (2) is according to the characteristic parameter processed by step (4), uses based at random
The Algorithms for Automatic Classification of forest carries out the classification automatically of target and obtains transition source candidate's body.
Algorithms for Automatic Classification based on random forest is being used to complete on observed image after the fast automatic identification in transition source, then
Perform step (6) and recognition result is carried out numerical value filtration, transition source candidate's body that output category is good.
The process step that numerical value filters is as follows:
(6.1) on the residual image that step (5) obtains centered by the candidate's body of transition source, 15 × 15 pixels are intercepted respectively
With the video in window of 8 × 8 pixels, the response of its pixel is designated as Flux15x,yAnd Flux8x,y；
(6.2) following judgement is done in the response to each pixel of video in window, meets the either condition in below equation, then
It is filtered out from possible transition source candidate's body；
Formula is 1.: len (Flux8x,y==1e-30) > 3
Formula is 2.: len (Flux15x,y==1e-30) > 10
Formula is 3.: len (Flux15x,y<-6σ+median(Flux15x,y))>5
Formula is 4.: len (Flux8x,y<-4σ+median(Flux8x,y))>3
Wherein, len () is pixel number statistical operator, and median () is median calculation operator, and σ is that video in window pixel rings
The standard variance that should be worth.
Build what transition source simulation observation image training sample realized in the following way by the method for emulation:
Original image is carried out background subtraction process；
Image after background subtraction processes selects an isolated star, as star image contour mould, by star image is taken turns
The flow of wide template carries out emulation reconstruct, and the mode being added on original image afterwards constructs the simulation observation containing transition source
Image；
Image after background subtraction processes is selected a collection of bright dark star not waited, as star image contour mould, by this template
The mode being added on original image constructs the simulation observation image containing transition source；
Above two is utilized to build former in selected a period of time of the method for the simulation observation image containing transition source
In beginning image, construct transition source simulation observation image training sample.
The implementation of training automatic categorizer is as follows:
Transition source simulation observation image training sample and template image are subtracted each other and obtains residual image；Residual image is carried out
Point source and the extraction of characteristic parameter；The all characteristic parameters extracted are input to random forest grader be trained, obtain
Automatic categorizer.
The characteristic parameter extracted includes following 25 parameters, referring specifically to following table:
A kind of astronomicalc optics transition source based on machine learning fast automatic recognition system, this system includes:
Image pre-processing module, for building transition source simulation observation image training sample by the method for emulation；
Characteristic parameter extraction module, subtracts each other transition source simulation observation image training sample and template image and obtains residual plot
Picture；Afterwards residual image is carried out the extraction of point source and characteristic parameter；
Automatically all characteristic parameters extracted are input to random forest grader and are trained, obtain by sort module
Automatic categorizer；
Automatically identification module, subtracts each other observed image and template image and obtains residual image, carry out point source and feature afterwards
The extraction of parameter, utilizes the characteristic parameter and automatic categorizer obtained above extracted, and uses based on random forest automatic point
Class algorithm obtains transition source candidate's body.
Also including numerical filter, numerical filter carries out numerical value filtration to the result of automatic identification module, output
The transition source candidate's body classified.
On the residual image that observed image and template image obtain centered by the candidate's body of transition source, intercept 15 respectively ×
15 pixels and the video in window of 8 × 8 pixels, the response of its pixel is designated as Flux15x,yAnd Flux8x,y；Each to video in window
Following judgement is done in the response of pixel, meets the either condition in below equation, then by its mistake from possible transition source candidate's body
Filtering, remaining transition source candidate's body is the transition source candidate's body classified；
Formula is 1.: len (Flux8x,y==1e-30) > 3
Formula is 2.: len (Flux15x,y==1e-30) > 10
Formula is 3.: len (Flux15x,y<-6σ+median(Flux15x,y))>5
Formula is 4.: len (Flux8x,y<-4σ+median(Flux8x,y))>3
Wherein, len () is pixel number statistical operator, and median () is median calculation operator, and σ is that video in window pixel rings
The standard variance that should be worth.
Image pre-processing module builds transition source simulation observation image training sample by such as lower section by the method for emulation
Formula realizes:
Original image is carried out background subtraction process；
Image after background subtraction processes selects an isolated star, as star image contour mould, by star image is taken turns
The flow of wide template carries out emulation reconstruct, and the mode being added on original image afterwards constructs the simulation observation containing transition source
Image；
Image after background subtraction processes is selected a collection of bright dark star not waited, as star image contour mould, by this template
The mode being added on original image constructs the simulation observation image containing transition source；
Above two is utilized to build former in selected a period of time of the method for the simulation observation image containing transition source
In beginning image, construct transition source simulation observation image training sample.
The present invention compared with prior art provides the benefit that:
(1) present invention proposes a kind of automatic identifying method based on machine learning, by research astronomical observation data
Feature, propose, based on the series of optimum characteristic parameter waiting profile measurement, and to utilize the profile of actual star image to build
The screening and filtering device of simulation training sample and numerical quantization, finally realizes the fast automatic identification side in a kind of astronomicalc optics transition source
Method.In the case of classification accuracy rate is consistent, the most similar processing method of processing speed improves about 1 magnitude.This
Bright it be also applied for other similar astronomical transitions and be derived from dynamic identification project.
(2) present invention research and introduce new optimization characteristic parameter, such as the parameter 1-13 in characteristic parameter table as indicated,
I.e. based on the contour feature waiting profile photometry and relevant auxiliary parameter to judge transition source.By fixed model, (two dimension is high
This) contour fitting the profile measurement of luminosity such as be changed into so that this recognition methods is ensureing same recognition accuracy automatically
In the case of, processing speed has the lifting of a magnitude.
(3) present invention uses when training grader simulation sample method and the earth ensure that emulation sample tool out
There is the similarity with true transition source of height, solve and be difficult to obtain a large amount of hands-on sample owing to actual transition source is rare
Problem, also improve the accuracy of automatic recognition classification simultaneously.
Accompanying drawing explanation
Fig. 1 astronomicalc optics transition source speed automatic identifying method process chart；
Fig. 2 bootstrapped training data builds schematic flow sheet
Detailed description of the invention
The present invention utilizes machine learning algorithm based on random forest, is built by the characteristic parameter optimized and simulation sample
The method of training aids, it is achieved the fast automatic of transition source accurately identifies.Implementation process is mainly: by observed image and template image
Execution alignment is subtracted each other, and the residual image after subtracting each other carries out the extraction of point source and characteristic parameter.Then, the characteristic parameter that will extract
It is input to automatic categorizer, Algorithms for Automatic Classification based on random forest carries out automatic recognition classification.Then after by fixed
The filter of system filters, and finally exports possible transition source candidate's body.Its specific implementation process is as shown in Figure 1:
(1) training sample is built by the method for emulation, as in figure 2 it is shown,
1. original image is carried out background subtraction process: first image is divided into the subelement lattice of 64 × 64 pixels, every height list
Unit's lattice calculate and filter through 3 σ that (that is, calculating variance is σ, removes the discrete point outside 3 σ, and continuous cycle calculations is until without outside 3 σ
Till discrete) after intermediate value.Then binary cubic polynomial is utilized to carry out plus positional information the intermediate value of single for all of son lattice
Matching, the value after matching is background value.Finally, background is removed from image.
2. from the image removing background, select a more isolated star, with 10 times of full width at half maximum (full width at half maximum FWHM:
Two-dimensional Gaussian function the Fitting Calculation) cage from image cutting out, as star image contour mould.Then, to this template
Flow carry out emulation reconstruct.The reconstruction calculations formula of flow is: Inew=K ∑ Fi,j, wherein, Fi,jFor in template (i, j) as
Flow at Yuan.Between limiting magnitude and saturated magnitude when 3 σ, with every 0.1 magnitude for interval, by adjusting K, reconstruct
Go out to emulate star image.Then these are emulated star image, be sprinkled into, by random site or regular arrangement, the original observed image that is added to
(each identical magnitude emulates the emulation star image reconstructing 10-15) constructs the simulation observation image containing transition source.
3. from the image removing background, a collection of (about 20) are selected from bright to the dark star not waited, as star image profile die
Plate.Should be noted that when choosing star image contour mould that these star images are not disturbed by surrounding star, the most isolated.Then by this
A little star image contour moulds, are sprinkled into the original observed image that is added to (each identical magnitude weight by random site or regular arrangement
Structure 10-15) construct the simulation observation image containing transition source.
4. two kinds of transition source reconstructing methods described in above the most 2. and 3. step are applied to the observed image in an evening (about
1000 width) in, two kinds of methods amount to emulation and reconstruct about 2000 width transition source simulation observation image training samples.
(2) complete the extraction of characteristic parameter by the training sample of emulation and be trained automatic categorizer；
1. transition source simulation observation image training sample and template image are subtracted each other and obtain residual image.Then to residual plot
As carrying out the extraction of point source and characteristic parameter.
2. all characteristic parameters extracted are input to random forest grader be trained, obtain automatic categorizer.
Major parameter during training describes and chooses as follows:
(3) observed image and template image subtract each other and obtain residual image.During actually used more than general selection three width
Observation as figure carry out intermediate value merging (participate in merge image correspondence pixel intermediate value be last value), so process can remove with
Machine noise spot, such as ultra rays etc..Template image typically choose some skies before the preferable identical width number image of picture quality carry out with
The merging method merging that observed image is same.The process of image subtraction mainly performs: the position pair between observation and template image
Neat coupling, observation and the flow between template image and star image outline, then perform the flow of corresponding pixel between two figures
Subtract each other (specific algorithm sees document " Image subtraction using a space-varying kernel " C.Alard,
Astron.Astrophys.Suppl.Ser.Volume 144,Number 2,June I2000)。
(4) residual image is carried out the extraction of point source and characteristic parameter；
1. use point source extraction algorithm conventional in astronomy (can be found in SEXtractor point source extraction algorithm: http: //
Www.astromatic.net/software/sextractor), residual image is measured, calculate point source position, stream
Value etc..
2. according to the above point source position attribution measured, the characteristic parameter of 25 dimensions is measured in corresponding star image position
Attribute, concrete dimension describes and see table.
R during matrix R (d) and B (d) are calculation in detail below in tablex,yAnd Bx,y.I (d) refers in residual image every
The two-dimensional matrix that individual pixel and pixel response quautity are constituted.
Concrete calculation is:
The calculating of Group I characteristic parameter (parameter 1-13) in upper table: light-metering radius is r0The flow at placeCalculating formula is:Fx,yFor pixel, (x, y) place's traffic intensity response quautity, total flow I is the flow at r=∞ in theory, but actual
Due to the attribute of function itself during calculating, r takes 10FWHM can be approximately equal to total flow, and wherein FWHM is Gauss mentioned above
Full width at half maximum after Function Fitting.The parameter 1. of upper table and parameter 2. can be calculated total flow 10% He according to this formula
Light-metering radius at 20%.It is light-metering flow during 2 pixels that fixing light-metering radius (aperture) is then chosen in the calculating of parameter 3..Deng
The light-metering flow rate calculation formula of high contour line is:Wherein start is the 5 of background fluctuation standard variance σ
I.e. take 5 σ, I againpFor pixel stream value maximum in star image.According to this formula difference level of response i, take i=0～4,
Obtain waiting contour line, calculate the area in different contour contour respectively, be the value of parameter 4.-8..Being calculated as of parameter 9.
Peak response flow (Ip) and constant aperture (at 2 pixels) flowRatio:Parameter 10. be calculated as aperture
(at 2 pixels) light-metering flowWith calculate with first-class profile in i=4 time wait the total flow (I in contour area4)
Ratio:Parameter 11. be calculated as aperture photometry flowWith profile total flow (I such as corrections4corr) ratio:Wherein I4corrCalculating be above contour contour inner area equivalence to the circular radius under same area, then
Calculate the total flow in this radius light-metering aperture, be I4corr.The calculating of parameter 12. light-metering error, is the light-metering error of star image,
Concrete calculating sees the astronomical algorithmic descriptions processed in software I RAF.Parameter 13. is i.e. therein ellipse according to calculating of star image profile
Rate, value is (0～1).
Being calculated as of Group II characteristic parameter (14.-25.) in upper table: first the residual image after subtracting each other is carried out pre-place
Reason, the process of pretreatment is: centered by star image target, intercept out the video in window of (2k+1) × (2k+1) pixel, correspondence
Two-dimensional matrix is that table is calculated as I, and merging matrix is C, then Cx,yCalculating statement formula:
Then two matrix amounts R are calculatedx,yAnd Bx,y, its expression formula is as follows: Wherein, median () is that intermediate value is calculated
Symbol, max () takes maximum operator.Following parameter can be calculated according to above formula.
Parameter 14.
Wherein (xc,ycThe star image centre coordinate extracted for point source)
Parameter 15.
Parameter 16.
Wherein count () is counting operator.
Parameter 17-20, specific algorithm sees astronomical process software SEXtractor.
Parameter 21: for the light-metering flow that constant aperture is 2 pixels, be converted into magnitude unit value.
Parameter 22:
Parameter 23:
Parameter 24:
Parameter 25:
(5) grader generated by step (2) carries out the automatic classification of target according to the characteristic parameter processed by step (4)
(python module based on random forests algorithm: http://neuro.debian.net/pkgs/python-
sklearn.html)。
In the case of camera hardware and weather conditions do not have special change, substantially it is not required to do automatic categorizer more more
Newly.As long as and the process of observed image performs step (3)-(5) and even performs following step (6).
(6) recognition result is carried out numerical value filtration, transition source candidate's body that output category is good.Data filter processes step
As follows:
1., on residual image centered by candidate's transition source, the window figure of 15 × 15 pixels and 8 × 8 pixels is intercepted respectively
Picture, the response of its pixel is designated as Flux15x,yAnd Flux8x,y。
2. following judgement is done in the response to each pixel of video in window, meets the either condition in below equation, then will
It filters out from possible transition source candidate's body.
Formula is 1.: len (Flux8x,y==1e-30) > 3
Formula is 2.: len (Flux15x,y==1e-30) > 10
Formula is 3.: len (Flux15x,y<-6σ+median(Flux15x,y))>5
Formula is 4.: len (Flux8x,y<-4σ+median(Flux8x,y))>3
Wherein, len () is pixel number statistical operator, and median () is median calculation operator, and σ is that video in window pixel rings
The standard variance that should be worth.
Transition source candidate's body that last output category is good.
The present invention also provides for a kind of astronomicalc optics transition source based on machine learning fast automatic recognition system, this system bag
Include:
Image pre-processing module, for building transition source simulation observation image training sample by the method for emulation；
Characteristic parameter extraction module, subtracts each other transition source simulation observation image training sample and template image and obtains residual plot
Picture；Afterwards residual image is carried out the extraction of point source and characteristic parameter；
Automatically all characteristic parameters extracted are input to random forest grader and are trained, obtain by sort module
Automatic categorizer；
Automatically identification module, subtracts each other observed image and template image and obtains residual image, carry out point source and feature afterwards
The extraction of parameter, utilizes the characteristic parameter and automatic categorizer obtained above extracted, and uses based on random forest automatic point
Class algorithm obtains transition source candidate's body.
Numerical filter carries out numerical value filtration to the result of automatic identification module, the transition source candidate that output category is good
Body.
Related content in system is corresponding with method, the most too much describes, referring in particular to the correspondence in method
Introduce.
The present invention is unspecified partly belongs to general knowledge as well known to those skilled in the art.
Claims (10)
1. astronomicalc optics transition source based on a machine learning quick automatic identification method, it is characterised in that the method includes:
(1) transition source simulation observation image training sample is built by the method for emulation；
(2) complete the extraction of characteristic parameter by transition source simulation observation image training sample and be trained automatic categorizer；
(3) observed image and template image are subtracted each other obtain residual image；
(4) residual image is carried out the extraction of point source and characteristic parameter；
(5) automatic categorizer generated by step (2) is according to the characteristic parameter processed by step (4), uses based on random forest
Algorithms for Automatic Classification carry out the automatically classification of target and obtain transition source candidate's body.
Method the most according to claim 1, it is characterised in that: complete using Algorithms for Automatic Classification based on random forest
On observed image after the fast automatic identification in transition source, then performing step (6) recognition result is carried out numerical value filtration, output category is good
Transition source candidate's body.
Method the most according to claim 2, it is characterised in that: the process step that numerical value filters is as follows:
(6.1) on the residual image that step (5) obtains centered by the candidate's body of transition source, 15 × 15 pixels and 8 are intercepted respectively
The video in window of × 8 pixels, the response of its pixel is designated as Flux15x,yAnd Flux8x,y；
(6.2) following judgement is done in the response to each pixel of video in window, meets the either condition in below equation, then by it
Filter out from possible transition source candidate's body；
Formula is 1.: len (Flux8x,y==1e-30) > 3
Formula is 2.: len (Flux15x,y==1e-30) > 10
Formula is 3.: len (Flux15x,y<-6σ+median(Flux15x,y))>5
Formula is 4.: len (Flux8x,y<-4σ+median(Flux8x,y))>3
Wherein, len () is pixel number statistical operator, and median () is median calculation operator, and σ is video in window pixel response value
Standard variance.
Method the most according to claim 1 and 2, it is characterised in that: build transition source simulation observation by the method for emulation
Image training sample realizes in the following way:
Original image is carried out background subtraction process；
Image after background subtraction processes selects an isolated star, as star image contour mould, by star image profile die
The flow of plate carries out emulation reconstruct, and the mode being added on original image afterwards constructs the simulation observation figure containing transition source
Picture；
Image after background subtraction processes is selected a collection of bright dark star not waited, as star image contour mould, by this template superposition
Mode on original image constructs the simulation observation image containing transition source；
Above two is utilized to build the method for the simulation observation image containing transition source to the original graph in selected a period of time
In Xiang, construct transition source simulation observation image training sample.
Method the most according to claim 4, it is characterised in that: the implementation of training automatic categorizer is as follows:
Transition source simulation observation image training sample and template image are subtracted each other and obtains residual image；Residual image is carried out point source
Extraction with characteristic parameter；The all characteristic parameters extracted are input to random forest grader be trained, obtain automatically
Grader.
Method the most according to claim 1 and 2, it is characterised in that: the characteristic parameter of extraction includes following 25 parameters, tool
Body sees following table:
7. astronomicalc optics transition source based on a machine learning fast automatic recognition system, it is characterised in that this system includes:
Image pre-processing module, for building transition source simulation observation image training sample by the method for emulation；
Characteristic parameter extraction module, subtracts each other transition source simulation observation image training sample and template image and obtains residual image；
Afterwards residual image is carried out the extraction of point source and characteristic parameter；
Automatically all characteristic parameters extracted are input to random forest grader and are trained by sort module, obtain automatically
Grader；
Automatically identification module, subtracts each other observed image and template image and obtains residual image, carry out point source and characteristic parameter afterwards
Extraction, utilize extract characteristic parameter and automatic categorizer obtained above, use calculation of automatically classifying based on random forest
Method obtains transition source candidate's body.
System the most according to claim 7, it is characterised in that: also including numerical filter, numerical filter is to automatically knowing
The result of other module carries out numerical value filtration, transition source candidate's body that output category is good.
System the most according to claim 8, it is characterised in that: the implementation of numerical filter is as follows: at observed image
On the residual image obtained with template image centered by the candidate's body of transition source, intercept 15 × 15 pixels and 8 × 8 pixels respectively
Video in window, the response of its pixel is designated as Flux15x,yAnd Flux8x,y；The response of each pixel of video in window is done and sentences as follows
Disconnected, meet the either condition in below equation, then it is filtered out from possible transition source candidate's body, remaining transition source is waited
Selecting body is the transition source candidate's body classified；
Formula is 1.: len (Flux8x,y==1e-30) > 3
Formula is 2.: len (Flux15x,y==1e-30) > 10
Formula is 3.: len (Flux15x,y<-6σ+median(Flux15x,y))>5
Formula is 4.: len (Flux8x,y<-4σ+median(Flux8x,y))>3
Wherein, len () is pixel number statistical operator, and median () is median calculation operator, and σ is video in window pixel response value
Standard variance.
10. according to the system described in claim 7 or 8, it is characterised in that: image pre-processing module is built by the method for emulation
Transition source simulation observation image training sample realizes in the following way:
Original image is carried out background subtraction process；
Image after background subtraction processes selects an isolated star, as star image contour mould, by star image profile die
The flow of plate carries out emulation reconstruct, and the mode being added on original image afterwards constructs the simulation observation figure containing transition source
Picture；
Image after background subtraction processes is selected a collection of bright dark star not waited, as star image contour mould, by this template superposition
Mode on original image constructs the simulation observation image containing transition source；
Above two is utilized to build the method for the simulation observation image containing transition source to the original graph in selected a period of time
In Xiang, construct transition source simulation observation image training sample.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610465133.5A CN106203462B (en) | 2016-06-23 | 2016-06-23 | Astronomicalc optics transition source quick automatic identification method and system based on machine learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610465133.5A CN106203462B (en) | 2016-06-23 | 2016-06-23 | Astronomicalc optics transition source quick automatic identification method and system based on machine learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106203462A true CN106203462A (en) | 2016-12-07 |
| CN106203462B CN106203462B (en) | 2019-05-21 |
Family
ID=57460969
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610465133.5A Expired - Fee Related CN106203462B (en) | 2016-06-23 | 2016-06-23 | Astronomicalc optics transition source quick automatic identification method and system based on machine learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106203462B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113865518A (en) * | 2021-10-09 | 2021-12-31 | 中国科学院新疆天文台 | Astronomical detection method for included angle between focal plane of large-field telescope and CCD target plane |
| CN117786417A (en) * | 2024-02-28 | 2024-03-29 | 之江实验室 | Model training method, transient source identification method and device and electronic equipment |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103124291A (en) * | 2012-12-11 | 2013-05-29 | 中国科学院国家天文台 | Intelligent supernova alarm and observation system |
| CN104574269A (en) * | 2015-01-08 | 2015-04-29 | 中国科学院国家天文台 | Method for correcting line-direction dragging of image observed by lunar-based astronomical telescope |
-
2016
- 2016-06-23 CN CN201610465133.5A patent/CN106203462B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103124291A (en) * | 2012-12-11 | 2013-05-29 | 中国科学院国家天文台 | Intelligent supernova alarm and observation system |
| CN104574269A (en) * | 2015-01-08 | 2015-04-29 | 中国科学院国家天文台 | Method for correcting line-direction dragging of image observed by lunar-based astronomical telescope |
Non-Patent Citations (6)
| Title |
|---|
| D.A.GOLDSTEIN 等: "Automated Transient Identi cation in the Dark Energy Survey", 《ARXIV:1504.02936V2》 * |
| HENRIK BRINK 等: "Using Machine Learning for Discovery in Synoptic Survey Imaging", 《ARXIV:1209.3775V1》 * |
| S. BAILEY等: "How to Find More Supernovae with Less Work: Object Classification Techniques for Difference Imaging", 《ARXIV:0705.0493V1》 * |
| YAN ZHAO 等: "Accelerating Astronomical Image Subtraction on Heterogeneous Processors", 《2013 IEEE 9TH INTERNATIONAL CONFERENCE ON E-SCIENCE》 * |
| 姜斌 等: "基于随机森林的激变变星候选体的数据挖掘", 《光谱学与光谱分析》 * |
| 马冬 等: "大视场光学瞬变源认证系统的设计与实现", 《天文研究与技术》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113865518A (en) * | 2021-10-09 | 2021-12-31 | 中国科学院新疆天文台 | Astronomical detection method for included angle between focal plane of large-field telescope and CCD target plane |
| CN113865518B (en) * | 2021-10-09 | 2024-02-06 | 中国科学院新疆天文台 | Astronomical detection method for included angle between focal plane of large-view-field telescope and CCD target surface |
| CN117786417A (en) * | 2024-02-28 | 2024-03-29 | 之江实验室 | Model training method, transient source identification method and device and electronic equipment |
| CN117786417B (en) * | 2024-02-28 | 2024-05-10 | 之江实验室 | Model training method, transient source identification method and device and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106203462B (en) | 2019-05-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jayakody et al. | Microscope image based fully automated stomata detection and pore measurement method for grapevines | |
| CN110084165A (en) | The intelligent recognition and method for early warning of anomalous event under the open scene of power domain based on edge calculations | |
| CN103870816A (en) | Plant identification method and device with high identification rate | |
| CN108198147A (en) | A kind of method based on the multi-source image fusion denoising for differentiating dictionary learning | |
| CN104156734A (en) | Fully-autonomous on-line study method based on random fern classifier | |
| CN110222773A (en) | Based on the asymmetric high spectrum image small sample classification method for decomposing convolutional network | |
| CN105842173A (en) | Method for identifying hyperspectral material | |
| CN113191452B (en) | Coal ash content online detection system based on deep learning and detection method thereof | |
| CN106204642A (en) | A kind of cell tracker method based on deep neural network | |
| CN106910188A (en) | The detection method of airfield runway in remote sensing image based on deep learning | |
| CN110287882A (en) | A kind of big chrysanthemum kind image-recognizing method based on deep learning | |
| CN110232334A (en) | A kind of steel construction corrosion recognition methods based on convolutional neural networks | |
| CN104504391B (en) | A kind of hyperspectral image classification method based on sparse features and markov random file | |
| CN108960413A (en) | A kind of depth convolutional neural networks method applied to screw surface defects detection | |
| CN108932474A (en) | A kind of remote sensing image cloud based on full convolutional neural networks compound characteristics sentences method | |
| CN106203462A (en) | Astronomicalc optics transition source quick automatic identification method based on machine learning and system | |
| CN110119739B (en) | Automatic classification method of ice crystal pictures | |
| CN113673618A (en) | Tobacco insect target detection method fused with attention model | |
| Martinazzo et al. | Deep learning for astronomical object classification: a case study | |
| CN109145955A (en) | A kind of Wood Identification Method and system | |
| CN114913523B (en) | Yolox-based multifunctional real-time intelligent plant stomata recognition system | |
| Awalgaonkar et al. | DEEVA: a deep learning and IoT based computer vision system to address safety and security of production sites in energy industry | |
| CN115273009A (en) | Road crack detection method and system based on deep learning | |
| Tan et al. | Automatic extraction of built-up area based on deep convolution neural network | |
| CN113920378A (en) | Attention mechanism-based radix bupleuri seed identification method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20190521 Termination date: 20200623 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |