CN102346129B - Gas radiation spectrum invariant characteristic extraction method suitable for temperature pressure change - Google Patents
Gas radiation spectrum invariant characteristic extraction method suitable for temperature pressure change Download PDFInfo
- Publication number
- CN102346129B CN102346129B CN 201110181499 CN201110181499A CN102346129B CN 102346129 B CN102346129 B CN 102346129B CN 201110181499 CN201110181499 CN 201110181499 CN 201110181499 A CN201110181499 A CN 201110181499A CN 102346129 B CN102346129 B CN 102346129B
- Authority
- CN
- China
- Prior art keywords
- spectrum
- fingerprint image
- arch
- target
- fingerprint
- 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.)
- Expired - Fee Related
Links
Images
Abstract
The invention provides a target identification method of gas extracted based on radiation spectrum characteristics. A fingerprint image is generated on a gas target spectrum by using a one-dimensional curvature scale space description method; fingerprint characteristics of the fingerprint image, namely the position of an arched peak, are extracted; and the closest position of the fingerprint characteristics is as the reference to carry out target matching identification. According to the method disclosed by the invention, steady radiation spectrum intrinsic characteristics which are not influenced by temperature and pressure are extracted; the method has robustness resisting condition change and good differentiating capability; and the support is provided for the future identification based on the spectrum under a complex condition.
Description
Technical field
The invention belongs to based on spectrum signature the time become the identification field, be specifically related to a kind of gas width of cloth and penetrate the spectrum signature extracting method.
Background technology
Conventional detection means normal operation simple spectrum section imaging detection method, gather to be identified/background energy at wider spectral coverage, this moment is to be identified often be submerged among the complicated background clutter or disturbed, camouflage etc. hidden, it is very faint to show as signal to be identified, signal to noise ratio (S/N ratio), signal to noise ratio are very low.Yet intrinsic physics, chemical characteristic and the structure etc. of object to be identified have than big difference with background, interference, camouflage etc., therefore on some spectral line or narrower spectral coverage, its feature can be apparent in view, namely on these spectral lines or narrower spectral coverage, signal to noise ratio (S/N ratio), the signal to noise ratio of the relative background of object to be identified are higher.By the extraction of spectrum fingerprint, utilize the spectrum signature of the uniqueness on these spectral lines to be identified, can greatly improve detectivity to be identified.
Multispectral information to be identified, background is effectively selected, extracts, analyzed, and spectrum information, image information and time-varying information merged, can realize the simplification system, distribute rationally, have complementary advantages, greatly improve the reliability of complex condition detection to be identified, identification.
Along with increasing of wave band quantity, can select as required or extract the outstanding feature to be identified of specific wave band during detection to be identified, the selection of spectral signature becomes more flexibly and is various.Simultaneously, contain abundant spectral to be identified in the spectroscopic data, these spectrals can be converted into different features by different performances with array mode, can provide widely signature analysis space for detecting and identifying to be identified.In addition, all kinds of atural objects have spectral reflectance and the radiation characteristic of oneself, utilize the spectral signature of different atural objects, can realize classification and the identification of atural object.Therefore, the spectrum fingerprint characteristic extracts the important content that becomes in the Study of recognition.
From the nineties in last century, the seminar of high spectrum image detection algorithm theoretical research to be identified has abroad appearred carrying out, such as the Chang group of Univ Maryland-Coll Park USA Baltimore branch school remote sensing signal and image treatment of laboratory, the Lincoln laboratory Manolakis of Massachusetts Institute Technology group etc.So far, use spectrum to identify and still exist a difficult problem (as shown in Figure 3): spectrum can change with the change of the physical conditions such as body temperature to be identified, pressure. therefore, unique invariant features that extracts spectrum become based on the multispectral figure of spectrum fingerprint (as) the important research content with detecting to be identified.
Summary of the invention
The object of the present invention is to provide a kind of gas width of cloth of resisting temperature pressure change to penetrate the spectrum signature extracting method, for identification and the detection of complex condition provides excellent support.
A kind ofly penetrate the gas target identification method that spectrum signature is extracted based on the width of cloth, relate to the fingerprint picture library in order to coupling, spectrum fingerprint picture library comprises the fingerprint image that adopts respectively one dimension curvature scale space describing method to generate to a plurality of reference spectrums, and the method is specially:
(1) adopt one dimension curvature scale space describing method to generate fingerprint image to gas target spectrum;
(2) extract respectively the fingerprint characteristic that target is composed all reference spectrum fingerprint images in fingerprint image and the fingerprint picture library, described fingerprint characteristic is the two-dimensional coordinate value on arch summit in the fingerprint image;
(3) upgrade reference spectrum fingerprint image R
iThe Y-axis coordinate figure on all arch summits be itself and height correction parameter k
iProduct, wherein,
I=1,2 ..., N, N are the reference spectrum sum,
Be Y-axis coordinate figure maximum in the arch summit of i reference spectrum fingerprint image,
Be maximum Y-axis coordinate figure in the arch summit of target spectrum fingerprint image;
(4) in the reference spectrum fingerprint image after step (3) is upgraded, be close to the standard search most with the arch vertex position and compose the reference spectrum fingerprint image R that fingerprint image mates most with target
m, and record R
mEach arch summit different with its two-dimensional coordinate value difference on arch summit corresponding to target spectrum fingerprint image respectively, the different summation of two-dimensional coordinate value difference that obtains is obtained mating cost:
(5) if, then judging the gas target less than discrimination threshold, the coupling cost that step (4) calculates is gas corresponding to R.
Further, described step (4) is specially:
(41) all arch summit two-dimensional coordinate values of the reference spectrum fingerprint image after target spectrum fingerprint image and step (3) renewal are carried out respectively normalized;
(42) after the normalized, the arch summit T of maximum Y-axis coordinate figure in the target spectrum fingerprint image
oWith reference spectrum fingerprint image R
iThe arch summit of middle maximum Y-axis coordinate figure
Consist of start node
I=1,2 ..., N;
(43) calculate T in the start node
oWith
The two-dimensional coordinate value difference different
Mark arch summit T
oWith
(44) from
I=1,2 ..., select reckling among the N
Its corresponding reference spectrum fingerprint image is designated as R
mIf target spectrum and R
mAll there is unmarked arch summit, enters step (45), if the target spectrum is only arranged or reference spectrum fingerprint image R is only arranged
mThere is unmarked arch summit, enters step (46), if target spectrum and reference spectrum fingerprint image R
mAll there is not unmarked arch summit, enters step (47);
(45) respectively at target spectrum and reference spectrum fingerprint image R
mSeek the arch summit configuration node of Y-axis coordinate figure maximum in the interior unlabelled arch summit, calculate the two-dimensional coordinate value discrepancy delta on two summits in this node, upgrade
Step (44) is returned on two arch summits in this node of mark;
(46) calculate target spectrum or reference spectrum fingerprint image R
mIn the two-dimensional coordinate value sum Δ on unmarked arch summit, upgrade
Step (44) is returned on these unmarked arch summits of mark;
(47) note R
mFor target is composed the reference spectrum fingerprint image that fingerprint image mates most,
For the coupling cost, finish.
Technique effect of the present invention is embodied in:
Can effectively obtain the invariant features that accumulates in the spectrum after the change on condition changes caused stave sight, to the condition influence of sample collection, not establish good basis for adapting to complex condition identification with detection when not set up the spectrum database.The present invention proposes to compose the New Policy that fingerprint characteristic extracts, and solves the problem of extracting invariant features, for the complex condition identification based on spectrum signature provides support.
Description of drawings
Fig. 1 is flow chart of steps of the present invention;
Fig. 2 is effect process flow diagram of the present invention;
Fig. 3 is the variation of spectral line under different physical conditions, and Fig. 3 (a) penetrates spectrum for the width of cloth of atmospheric carbon dioxide under different temperatures; Fig. 3 (b) penetrates spectrum for the width of cloth of atmospheric carbon dioxide under different pressure;
Fig. 4 is that the spectrum signature of atmospheric carbon dioxide under temperature variation extracted the result. Fig. 4 (a) is the radiation spectrum under the atmospheric carbon dioxide different temperatures; Fig. 4 (b) for this reason material penetrates the spectrum fingerprint image of spectrum at the solid line width of cloth; Fig. 4 (c) for this reason material represents the spectrum fingerprint image that the width of cloth is penetrated spectrum at dotted line; Fig. 4 (d) distributes for initial characteristics; Fig. 4 (e) distributes for feature after revising.
Fig. 5 is that the spectrum signature of allied substances under pressure change extracted the result. Fig. 5 (a) is the radiation spectrum under the different pressure of atmospheric carbon dioxide; Fig. 5 (b) for this reason width of cloth that represents at dotted line of material penetrates the spectrum fingerprint image of spectrum; Fig. 5 (c) for this reason width of cloth that represents at solid line of material penetrates the spectrum fingerprint image of spectrum; Fig. 5 (d) distributes for initial characteristics; Fig. 5 (e) distributes for feature after revising.
Fig. 6 is that allied substances extracts the result in temperature with the spectrum signature under pressure changes simultaneously. the radiation spectrum under the different physical states of Fig. 6 (a) carbon dioxide (temperature, pressure all change); The width of cloth that Fig. 6 (b) dotted line represents is penetrated the fingerprint image of spectrum; The width of cloth that Fig. 6 (c) solid line represents is penetrated the fingerprint image of spectrum; Fig. 6 (d) initial characteristics distributes; Feature distributed after Fig. 6 (e) revised.
Fig. 7 extracts the result for the spectrum signature of foreign peoples's material. and Fig. 7 (a) is the radiation spectrum of two kinds of different materials. and Fig. 7 (b) is dotted line representative spectrum fingerprint image; Fig. 7 (c) is solid line representative spectrum fingerprint image; Fig. 7 (d) distributes for initial characteristics; Fig. 7 (e) distributes for feature after revising.
Embodiment
Invariant features of the present invention that refer under multispectral and super spectrum acquisition sensor condition, to obtain with spectrum signature relevant uniqueness to be identified, and this spectrum signature remains unchanged under different environmental baselines.
Treatment scheme of the present invention is specially as illustrated in fig. 1 and 2:
(1) use respectively one dimension curvature scale space (CSS) to generate fingerprint image to reference spectrum and spectrum to be identified.
1. create the fingerprint image of a sky, set up the two-dimensional coordinate axle, wherein X-axis represents wave number, and Y-axis is to rear sight degree parameter;
2. select a less scale parameter value as current scale parameter;
3. generate the one dimension gaussian kernel according to current scale parameter, and the spectrum data be considered as the one dimension vector data corresponding with current scale parameter the one dimension gaussian kernel carry out convolution algorithm;
4. the result who obtains after the convolution is asked second derivative, the positional information (comprising wave number and scale parameter) that record second derivative zero crossing occurs, step 1. in the fingerprint image correspondence position of establishment mark;
5. obtain larger scale parameter, repeat step 3. to 4., until no longer produce new second derivative zero crossing.
(2) extract the initial fingerprint feature
Fingerprint image apparent be some arches, take the height (being the Y-axis coordinate) of arch and position (being the X-axis coordinate) as the initial fingerprint feature.
(3) initial characteristics correction
In the reference spectrum fingerprint image the apex height of high arch as standard, in the spectrum fingerprint image to be identified the summit of high arch carry out high elongation, it is overlapped with the former. the computed altitude corrected parameter:
All dome height of spectrum fingerprint to be identified multiply by NormRatio and carry out normalization.
(4) the coupling cost is calculated
In the reference spectrum fingerprint image after step (3) is upgraded, be close to the standard search most with the arch vertex position and compose the reference spectrum fingerprint image R that fingerprint image mates most with target
m, and record R
mEach arch summit different with its two-dimensional coordinate value difference on arch summit corresponding to target spectrum fingerprint image respectively, the different summation of two-dimensional coordinate value difference that obtains is obtained mating cost.
The simplest a kind of direct method is exactly to calculate respectively the coupling cost of all fingerprint images in spectrum fingerprint image to be identified and the target spectrum fingerprint picture library, and the target spectrum fingerprint image of the minimum correspondence of coupling cost is final coupling cost.
In practical operation, in order to improve computing velocity, the implementation algorithm performance optimization can adopt the strategy of following depth-first search to finish.
The coupling cost of two fingerprint images of the present invention is to obtain after the balanced cost algorithms of special case (Uniform Cost Algorithm) of employing A* algorithm is improved.The function of this algorithm is that all or part of arch in the whole arches in the fingerprint image (template figure) and another fingerprint image (target figure) is carried out optimum matching.Algorithm steps is as follows:
If fingerprint image to be identified is O, the template base fingerprint image is R
i, i=1,2 ..., the N matching task is specially at i=1, and 2 ..., the fingerprint image the most close with the arch summit of O sought on the arch summit on the whole among the N.
(41) to all fingerprint images (be O, i=1,2 ..., XY axle N) is normalized to respectively 0 to 1, and matching task is from higher arch; In practical operation, can ignore highly too small arch, its objective is the interference of removing the factors such as noise. it is 5% (desirable scope 1%-10%) that arch filtration threshold value can be set, and namely height will be left in the basket less than the highest arch 5%.
(42) create initial node.The arch summit T of maximum Y-axis coordinate figure in the target spectrum fingerprint image
oWith reference spectrum fingerprint image R
iThe arch summit of middle maximum Y-axis coordinate figure
Consist of start node
I=1,2 ..., N.Because 3 fingerprint images are arranged in the template base, so O and i=1,2 ..., the N symbiosis becomes N initial node.
(43) the coupling cost of calculating initial node. what mate cost is calculated as two pairing arch summits respectively at X, the difference sum of Y-axis. the arch of having matched is carried out mark.
An example is described below.
If fingerprint image to be identified is A, reference fingerprint figure is B
1, B
2, B
3The arch apex coordinate is (0.2,0.4) among the A, (0.6,0.9) and (0.9,0.2) .B
1Middle arch apex coordinate is (0.1,0.4) and (0.8,0.6), B
2Middle arch apex coordinate is (0.2,0.5), (0.2,0.7) and (0.8,0.2), B
3Middle arch apex coordinate is (0.4,0.7), (0.5,0.3), (0.8,0.4) and (0.9,0.5).
The highest arch apex coordinate is (0.6,0.9) among the A, B
1, B
2, B
3In the highest arch summit sit and to be respectively (0.8,0.6), (0.4,0.7) and (0.2,0.7), among the A the highest arch summit respectively with B
1, B
2, B
3In the highest arch summit create initial node, calculating is mated on the above arch of mark summit.If two arch apex coordinates are respectively (x
1, y
1) and (x
2, y
2), coupling cost computing formula is
MC＝abs(x
2-x
1)+abs(y
2-y
1)
Abs () expression takes absolute value, and calculates node AB according to following formula
1The coupling cost is 0.5, AB
2The coupling cost is 0.6, AB
3The coupling cost is 0.4.Node AB
3The coupling cost is minimum, it is expanded, at A and B
3In seek in the unlabelled arch summit the highest a pair ofly, carry out step in claims (44) to (47), finally obtain AB
2The coupling cost is 0.6.
(5) identification is differentiated
When mating cost less than discrimination threshold, object to be identified is judged to target; Otherwise, object to be identified be judged to non-target .. wherein threshold value be empirical value, general way is that all samples of getting in a target to be identified and the target fingerprint picture library mate calculating, will obtain the maximal value of a coupling cost, and threshold value is got the 1%-10% of maximum coupling cost.
The below enumerates several examples.
Example one
Carry out in this example the fingerprint image matching test of radiation spectrum under the allied substances different temperatures.
Choose the width of cloth of atmospheric carbon dioxide under different temperatures penetrate the spectrum data test. the spectrum data shown in Fig. 4 (a).Fig. 4 (a) is the radiation spectrum under the atmospheric carbon dioxide different temperatures, and wherein to represent atmospheric carbon dioxide be that the width of cloth under the 296K is penetrated spectrum in temperature to solid line, and it is that the width of cloth under the 460K is penetrated spectrum in temperature that dotted line represents atmospheric carbon dioxide.
The first step: generate the CSS fingerprint image.
1. create the fingerprint image of a sky, X-axis is wave number, and Y-axis is scale parameter;
2. setting initial gaussian kernel scale parameter is 1, and the increment step-length of scale parameter is 0.5;
3. generate the one dimension gaussian kernel according to current scale parameter;
4. the one dimension gaussian kernel that obtains in the spectrum data vector and 3 is carried out convolution algorithm, the result who obtains after the convolution is asked second derivative, the position that record second derivative zero crossing occurs. the fingerprint image Y-axis (being scale parameter) that creates in 1 is the beam location that 1 local mark second derivative zero crossing occurs.
5. treat the scale parameter of computing according to the increment step size computation next one of scale parameter, repeat 3 and 4, until no longer produce new second derivative zero crossing.
6. fingerprint image generates. and the result is shown in Fig. 4 (b) and 4 (c).
Second step: extract the initial fingerprint feature.
Fingerprint image apparent be some arches, take the height (being the Y-axis coordinate) of arch and position (being the X-axis coordinate) as the initial fingerprint feature. the formation of arch is the process that is faded away along with the increase of scale parameter by a pair of second derivative zero crossing.
Result of calculation as shown in Tables 1 and 2, initial fingerprint feature such as Fig. 4 (d), round dot represents spectrum to be identified, Fang Dian represents target spectrum.Table 1 reference spectrum fingerprint characteristic
| X (wave number) | 0.4571 | 0.2857 | 0.8266 | 0.1357 |
| Y (yardstick) | 0.8400 | 0.2800 | 0.3600 | 0.1000 |
Table 2 spectrum fingerprint beginning feature to be identified:
| X (wave number) | 0.5000 | 0.2786 | 0.8429 | 0.12869 |
| Y (yardstick) | 0.5000 | 0.2786 | 0.8429 | 0.1286 |
The 3rd step: initial characteristics correction.
In the reference spectrum fingerprint image the apex height of high arch as standard, in the spectrum fingerprint image to be identified the summit of high arch carry out high elongation, it is overlapped with the former.Calculate normalized parameter:
All dome height of spectrum fingerprint to be identified multiply by NormRatio and carry out normalization.
Result of calculation as shown in Tables 3 and 4, initial characteristics distributes such as Fig. 4 (e), round dot represents spectrum to be identified, Fang Dian represents target spectrum.
The revised reference spectrum fingerprint characteristic of table 3
| X (wave number) | 0.4571 | 0.2857 | 0.8286 | 0.1357 |
| Y (yardstick) | 0.8400 | 0.2800 | 0.3600 | 0.1000 |
Feature after the table 4 spectrum correction to be identified:
| X (wave number) | 0.3571 | 0.2786 | 0.8429 | 0.1286 |
| Y (yardstick) | 0.8400 | 0.2809 | 0.3326 | 0.1800 |
The 4th step: calculate the coupling cost.
Calculating the coupling cost according to the balanced cost algorithms of the special case of A* algorithm (Uniform Cost Algorithm) is 0.40.
The 5th step: identification is differentiated.
The coupling cost is less than threshold value 1, and spectrum to be identified is judged to carbon dioxide.
Example two
Carry out in this example the fingerprint image matching test of radiation spectrum under the different pressure of allied substances.
Choose the width of cloth of atmospheric carbon dioxide under different pressure penetrate the spectrum data test. the spectrum data shown in Fig. 5 (a).Fig. 5 (a) is the radiation spectrum under the different pressure of atmospheric carbon dioxide, and wherein to represent atmospheric carbon dioxide be that the width of cloth under the 0.1MPa is penetrated spectrum at pressure to solid line, and it is that the width of cloth under the 0.2MPa is penetrated spectrum at pressure that dotted line represents atmospheric carbon dioxide.
The first step: generate the CSS fingerprint image.
Reference spectrum and spectrum to be identified are carried out respectively step 1 to 6.
1. create the fingerprint image of a sky, X-axis is wave number, and Y-axis is scale parameter;
2. setting initial gaussian kernel scale parameter is 1, and the increment step-length of scale parameter is 0.5;
3. generate the one dimension gaussian kernel according to current scale parameter;
4. the one dimension gaussian kernel that obtains in the spectrum data vector and 3 is carried out convolution algorithm, the result who obtains after the convolution is asked second derivative, the position that record second derivative zero crossing occurs. the fingerprint image Y-axis (being scale parameter) that creates in 1 is the beam location that 1 local mark second derivative zero crossing occurs.
5. treat the scale parameter of computing according to the increment step size computation next one of scale parameter, repeat 3 and 4, until no longer produce new second derivative zero crossing.
6. fingerprint image generates. and the result is shown in Fig. 5 (b) and 5 (c).
Second step: extract the initial fingerprint feature.
Fingerprint image apparent be some arches, take the height (being the Y-axis coordinate) of arch and position (being the X-axis coordinate) as the initial fingerprint feature. the formation of arch is the process that is faded away along with the increase of scale parameter by a pair of second derivative zero crossing.
Result of calculation is shown in table 6 and 7, and initial characteristics distributes and represents spectrum to be identified such as Fig. 5 (d) round dot, and Fang Dian represents the target spectrum.
Table 5 reference spectrum fingerprint characteristic
| X (wave number) | 0.4571 | 0.2929 | 0.8286 |
| Y (yardstick) | 0.8240 | 0.3600 | 0.3000 |
Table 6 spectrum fingerprint characteristic to be identified
| X (wave number) | 0.4714 | 0.2857 | 0.8214 | 0.1371 |
| Y (yardstick) | 0.9080 | 0.3240 | 0.3040 | 0.0880 |
The 3rd step: initial characteristics correction.
In the reference spectrum fingerprint image the apex height of high arch as standard, in the spectrum fingerprint image to be identified the summit of high arch carry out high elongation, it is overlapped with the former. calculate normalized parameter:
All dome height of spectrum fingerprint to be identified multiply by NormRatio and carry out normalization.
Result of calculation shown in table 8 and 9, initial characteristics such as Fig. 5 (e), round dot represents spectrum to be identified, Fang Dian represents target spectrum.
The revised reference spectrum fingerprint characteristic of table 7
| X (wave number) | 0.4571 | 0.2929 | 0.8543 |
| Y (yardstick) | 0.8240 | 0.3600 | 0.3000 |
The revised spectrum fingerprint characteristic to be identified of table 8:
| X (wave number) | 0.4714 | 0.2857 | 0.8214 | 0.1371 |
| Y (yardstick) | 0.8240 | 0.2940 | 0.2758 | 0.0798 |
The 4th step: calculate the coupling cost.
Calculating the coupling cost according to the balanced cost algorithms of the special case of A* algorithm (Uniform Cost Algorithm) is 0.45.
The 5th step: identification is differentiated.
The coupling cost is less than threshold value 1, and spectrum to be identified is judged to carbon dioxide.
Example three
Carry out in this example allied substances temperature and pressure and change simultaneously the fingerprint image matching test of lower radiation spectrum.Choose the width of cloth of atmospheric carbon dioxide when different temperatures and pressure penetrate the spectrum data test. the spectrum data shown in Fig. 6 (a).Radiation spectrum under the different physical states of Fig. 6 (a) carbon dioxide (temperature, pressure all change). wherein to represent carbon dioxide be 172K in temperature to solid line, and pressure is the radiation spectrum under the 0.1MPa; It is 296K in temperature that dotted line represents carbon dioxide, and pressure is the radiation spectrum under the 0.2MPa.
The first step: generate the CSS fingerprint image.
Reference spectrum and spectrum to be identified are carried out respectively step 1 to 6.
1. create the fingerprint image of a sky, X-axis is wave number, and Y-axis is scale parameter;
2. setting initial gaussian kernel scale parameter is 1, and the increment step-length of scale parameter is 0.5;
3. generate the one dimension gaussian kernel according to current scale parameter;
4. the one dimension gaussian kernel that obtains in the spectrum data vector and 3 is carried out convolution algorithm, the result who obtains after the convolution is asked second derivative, the position that record second derivative zero crossing occurs. the fingerprint image Y-axis (being scale parameter) that creates in 1 is the beam location that 1 local mark second derivative zero crossing occurs.
5. treat the scale parameter of computing according to the increment step size computation next one of scale parameter, repeat 3 and 4, until no longer produce new second derivative zero crossing.
6. fingerprint image generates. and the result is shown in Fig. 6 (b) and 6 (c).
Second step: extract the initial fingerprint feature.
Fingerprint image apparent be some arches, take the height (being the Y-axis coordinate) of arch and position (being the X-axis coordinate) as the initial fingerprint feature. the formation of arch is the process that is faded away along with the increase of scale parameter by a pair of second derivative zero crossing.
Result of calculation as shown in Tables 9 and 10, initial characteristics distributes such as Fig. 6 (d), red representative spectrum to be identified, the blue target that represent is composed.
Table 9 reference spectrum initial characteristics
| X (wave number) | 0.48 | 0.27 | 0.84 | 0.13 |
| Y (yardstick) | 0.83 | 0.23 | 0.21 | 0.06 |
Table 10 spectrum initial characteristics to be identified:
| X (wave number) | 0.49 | 0.28 | 0.82 |
| Y (yardstick) | 0.91 | 0.23 | 0.20 |
The 3rd step: initial characteristics correction.
In the reference spectrum fingerprint image the apex height of high arch as standard, in the spectrum fingerprint image to be identified the summit of high arch carry out high elongation, it is overlapped with the former. calculate normalized parameter:
All dome height of spectrum fingerprint to be identified multiply by NormRatio and carry out normalization.
Result of calculation shown in table 11 and 12, initial characteristics such as Fig. 6 (e), red representative spectrum to be identified, the blue target that represent is composed.
Table 11 reference spectrum correction feature
| X (wave number) | 0.4857 | 0.2786 | 0.8429 | 0.1357 |
| Y (yardstick) | 0.8333 | 0.2333 | 0.2167 | 0.0667 |
Table 12 spectrum correction to be identified feature:
| X (wave number) | 0.4929 | 0.2857 | 0.8286 |
| Y (yardstick) | 0.8333 | 0.2030 | 0.1848 |
The 4th step: calculate the coupling cost.
Calculating the coupling cost according to the balanced cost algorithms of the special case of A* algorithm (Uniform Cost Algorithm) is 2.6.
The 5th step: identification is differentiated.
The coupling cost is less than threshold value 1, and it is carbon dioxide that spectrum to be identified is judged to reference spectrum.
Example four
Carry out in this example the fingerprint image matching test of foreign peoples's material radiation spectrum.The width of cloth of choosing the mixed gas that atmospheric carbon dioxide and aviation kerosene JET A-1 burning produces is penetrated the spectrum data and is tested. and the spectrum data are shown in Fig. 7 (a).Fig. 7 (a) is the radiation spectrum of two kinds of different materials, and wherein solid line represents the radiation spectrum that aviation kerosene JETA-1 open burning produces mixed gas; It is 296K in temperature that dotted line represents carbon dioxide, and pressure is that the width of cloth under the 0.2MPa is penetrated spectrum.
The first step: generate the CSS fingerprint image.
Reference spectrum and spectrum to be identified are carried out respectively step 1 to 6.
1. create the fingerprint image of a sky, X-axis is wave number, and Y-axis is scale parameter;
2. setting initial gaussian kernel scale parameter is 1, and the increment step-length of scale parameter is 0.5;
3. generate the one dimension gaussian kernel according to current scale parameter;
4. the one dimension gaussian kernel that obtains in the spectrum data vector and 3 is carried out convolution algorithm, the result who obtains after the convolution is asked second derivative, the position that record second derivative zero crossing occurs. the fingerprint image Y-axis (being scale parameter) that creates in 1 is the beam location that 1 local mark second derivative zero crossing occurs.
5. treat the scale parameter of computing according to the increment step size computation next one of scale parameter, repeat 3 and 4, until no longer produce new second derivative zero crossing.
6. fingerprint image generates. and the result is shown in Fig. 7 (b) and 7 (c).
Second step: extract the initial fingerprint feature.
Fingerprint image apparent be some arches, take the height (being the Y-axis coordinate) of arch and position (being the X-axis coordinate) as the initial fingerprint feature. the formation of arch is the process that is faded away along with the increase of scale parameter by a pair of second derivative zero crossing.
Result of calculation is shown in table 13 and 14, and initial characteristics distributes such as Fig. 7 (d), and round dot represents spectrum to be identified, and Fang Dian represents the target spectrum.
Table 13 reference spectrum fingerprint characteristic
| X (wave number) | 0.4571 | 0.2857 | 0.8286 | 0.1357 |
| Y (yardstick) | 0.7000 | 0.2333 | 0.3000 | 0.0833 |
Table 14 spectrum fingerprint beginning feature to be identified:
| X (wave number) | 0.5429 | 0.2571 | 0.8786 | 0.2286 |
| Y (yardstick) | 0.9000 | 0.2333 | 0.0833 | 0.0667 |
The 3rd step: initial characteristics correction.
In the reference spectrum fingerprint image the apex height of high arch as standard, in the spectrum fingerprint image to be identified the summit of high arch carry out high elongation, it is overlapped with the former. calculate normalized parameter:
All dome height of spectrum fingerprint to be identified multiply by NormRatio and carry out normalization.
Result of calculation shown in table 15 and 16, initial characteristics such as Fig. 7 (e), round dot represents spectrum to be identified, Fang Dian represents target spectrum.
The revised reference spectrum fingerprint characteristic of table 15
| X (wave number) | 0.4571 | 0.2857 | 0.8286 | 0.1357 |
| Y (yardstick) | 0.7000 | 0.2333 | 0.3000 | 0.0833 |
The revised spectrum fingerprint characteristic to be identified of table 16:
| X (wave number) | 0.5429 | 0.2571 | 0.8786 | 0.2286 |
| Y (yardstick) | 0.7000 | 0.1815 | 0.0648 | 0.0519 |
The 4th step: calculate the coupling cost.
Calculating the coupling cost according to the balanced cost algorithms of the special case of A* algorithm (Uniform Cost Algorithm) is 3.6.The 5th step: identification is differentiated.
The coupling cost is greater than threshold value 1, and it is foreign peoples's material that spectrum to be identified is judged to reference spectrum.
To sum up example one to four as seen, the feature extraction that the present invention is used for the gas radiation spectrum all is not subjected to the impact of temperature and pressure change, and has good material and distinguish and separating capacity.
Claims (1)
1. penetrate the gas target identification method that spectrum signature is extracted based on the width of cloth for one kind, relate to the fingerprint picture library in order to coupling, the fingerprint picture library comprises the fingerprint image that adopts respectively one dimension curvature scale space describing method to generate to a plurality of reference spectrums, and this gas target identification method is specially:
(1) adopt one dimension curvature scale space describing method to generate fingerprint image to gas target spectrum;
(2) extract respectively the fingerprint characteristic that target is composed all reference spectrum fingerprint images in fingerprint image and the fingerprint picture library, described fingerprint characteristic is the two-dimensional coordinate value on arch summit in the fingerprint image;
(3) upgrade reference spectrum fingerprint image R
iThe Y-axis coordinate figure on all arch summits be the product of itself and height correction parameter ki, wherein,
I=1,2 ..., N, N are the reference spectrum sum,
Be Y-axis coordinate figure maximum in the arch summit of i reference spectrum fingerprint image,
Be maximum Y-axis coordinate figure in the arch summit of target spectrum fingerprint image;
(4) in the reference spectrum fingerprint image after step (3) is upgraded, be close to the standard search most with the arch vertex position and compose the reference spectrum fingerprint image R that fingerprint image mates most with target
m, and record R
mEach arch summit different with its two-dimensional coordinate value difference on arch summit corresponding to target spectrum fingerprint image respectively, the different summation of two-dimensional coordinate value difference that obtains is obtained mating cost:
(5) if the coupling cost that step (4) calculates, judges then that the gas target is the gas of correspondence less than discrimination threshold;
Described step (4) is specially:
(41) all arch summit two-dimensional coordinate values of the reference spectrum fingerprint image after target spectrum fingerprint image and step (3) renewal are carried out respectively normalized;
(42) after the normalized, the arch summit T of maximum Y-axis coordinate figure in the target spectrum fingerprint image
oWith reference spectrum fingerprint image R
iThe arch summit of middle maximum Y-axis coordinate figure
Consist of start node
I=1,2 ..., N;
(43) calculate T in the start node
oWith
The two-dimensional coordinate value difference different
, mark arch summit T
oWith
(44) from
I=1,2 ..., select reckling among the N
Its corresponding reference spectrum fingerprint image is designated as R
mIf target spectrum and R
mAll there is unmarked arch summit, enters step (45), if the target spectrum is only arranged or reference spectrum fingerprint image R is only arranged
mThere is unmarked arch summit, enters step (46), if target spectrum and reference spectrum fingerprint image R
mAll there is not unmarked arch summit, enters step (47);
(45) respectively at target spectrum and reference spectrum fingerprint image R
mSeek the arch summit configuration node of Y-axis coordinate figure maximum in the interior unlabelled arch summit, calculate the two-dimensional coordinate value discrepancy delta on two summits in this node, upgrade
Step (44) is returned on two arch summits in this node of mark;
(46) calculate target spectrum or reference spectrum fingerprint image R
mIn unmarked arch summit two-dimensional coordinate value sum Δ ', upgrade
Step (44) is returned on these unmarked arch summits of mark;
(47) note R
mFor target is composed the reference spectrum fingerprint image that fingerprint image mates most,
For the coupling cost, finish.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110181499 CN102346129B (en) | 2011-06-30 | 2011-06-30 | Gas radiation spectrum invariant characteristic extraction method suitable for temperature pressure change |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110181499 CN102346129B (en) | 2011-06-30 | 2011-06-30 | Gas radiation spectrum invariant characteristic extraction method suitable for temperature pressure change |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102346129A CN102346129A (en) | 2012-02-08 |
| CN102346129B true CN102346129B (en) | 2013-03-27 |
Family
ID=45545004
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 201110181499 Expired - Fee Related CN102346129B (en) | 2011-06-30 | 2011-06-30 | Gas radiation spectrum invariant characteristic extraction method suitable for temperature pressure change |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102346129B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6424745B1 (en) * | 1998-05-19 | 2002-07-23 | Lucent Technologies Inc. | Method and apparatus for object recognition |
| CN101853503A (en) * | 2010-04-26 | 2010-10-06 | 华中科技大学 | Spectral line inflexion multi-scale optimizing segmentation method and application thereof |
| CN101976337A (en) * | 2010-10-27 | 2011-02-16 | 华中科技大学 | Method for extracting stable spectrum characteristics of solid matter |
-
2011
- 2011-06-30 CN CN 201110181499 patent/CN102346129B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6424745B1 (en) * | 1998-05-19 | 2002-07-23 | Lucent Technologies Inc. | Method and apparatus for object recognition |
| CN101853503A (en) * | 2010-04-26 | 2010-10-06 | 华中科技大学 | Spectral line inflexion multi-scale optimizing segmentation method and application thereof |
| CN101976337A (en) * | 2010-10-27 | 2011-02-16 | 华中科技大学 | Method for extracting stable spectrum characteristics of solid matter |
Non-Patent Citations (3)
| Title |
|---|
| 范璐 等.大豆油和花生油傅里叶变换红外吸收光谱识别分析.《河南工业大学学报（自然科学版）》.2007,第28卷(第1期),全文. * |
| 药品质量稳定性的傅立叶变换红外吸收光谱识别分析;赵小蓉 等;《分析测试学报》;20080831;第27卷(第8期);全文 * |
| 赵小蓉 等.药品质量稳定性的傅立叶变换红外吸收光谱识别分析.《分析测试学报》.2008,第27卷(第8期),全文. |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102346129A (en) | 2012-02-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108389188B (en) | Sparse hyperspectral abnormal target detection method | |
| Chang | Hyperspectral imaging: techniques for spectral detection and classification | |
| Chang | Hyperspectral target detection: Hypothesis testing, signal-to-noise ratio, and spectral angle theories | |
| Xie et al. | Background learning based on target suppression constraint for hyperspectral target detection | |
| CN109858477A (en) | The Raman spectrum analysis method of object is identified in complex environment with depth forest | |
| CN103065135A (en) | License number matching algorithm based on digital image processing | |
| Plourde et al. | Estimating species abundance in a northern temperate forest using spectral mixture analysis | |
| CN102521605A (en) | Wave band selection method for hyperspectral remote-sensing image | |
| CN110443139B (en) | Hyperspectral remote sensing image noise band detection method for classification | |
| CN107292258A (en) | High spectrum image low-rank representation clustering method with filtering is modulated based on bilateral weighted | |
| CN107463895B (en) | Small and weak damage object detection method based on neighborhood vector PCA | |
| Toniol et al. | Potential of hyperspectral metrics and classifiers for mapping Brazilian savannas in the rainy and dry seasons | |
| CN102663740B (en) | SAR image change detection method based on image cutting | |
| Pu | Detecting and Mapping Invasive Plant Species Using Hyperspectral Data | |
| CN111027509A (en) | Hyperspectral image target detection method based on double-current convolution neural network | |
| Zhang et al. | Hyperspectral anomaly detection via dual collaborative representation | |
| Zhang et al. | An improved feature set for hyperspectral image classification: Harmonic analysis optimized by multiscale guided filter | |
| Zhang et al. | Isolation forest for anomaly detection in hyperspectral images | |
| Zhao et al. | Hyperspectral target detection method based on nonlocal self-similarity and rank-1 tensor | |
| Negri et al. | Exploring the capability of ALOS PALSAR L-band fully polarimetric data for land cover classification in tropical environments | |
| CN102346129B (en) | Gas radiation spectrum invariant characteristic extraction method suitable for temperature pressure change | |
| CN111767813B (en) | Multi-feature and long-time-sequence probability synergistic mangrove forest extraction method | |
| Hassaïne et al. | An online signature verification system for forgery and disguise detection | |
| CN107153807A (en) | A kind of non-greedy face identification method of two-dimensional principal component analysis | |
| CN101976337B (en) | Method for extracting stable spectrum characteristics of solid matter |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20130327 Termination date: 20180630 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |