CN101839858B - Nonlinear laser fluorescence spectrum real-time identification method - Google Patents

Nonlinear laser fluorescence spectrum real-time identification method Download PDF

Info

Publication number
CN101839858B
CN101839858B CN2010101734358A CN201010173435A CN101839858B CN 101839858 B CN101839858 B CN 101839858B CN 2010101734358 A CN2010101734358 A CN 2010101734358A CN 201010173435 A CN201010173435 A CN 201010173435A CN 101839858 B CN101839858 B CN 101839858B
Authority
CN
China
Prior art keywords
spectrum
fluorescence
data
wave
sample
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN2010101734358A
Other languages
Chinese (zh)
Other versions
CN101839858A (en
Inventor
李颖
陈澎
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Dalian Maritime University
Original Assignee
Dalian Maritime University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Dalian Maritime University filed Critical Dalian Maritime University
Priority to CN2010101734358A priority Critical patent/CN101839858B/en
Publication of CN101839858A publication Critical patent/CN101839858A/en
Application granted granted Critical
Publication of CN101839858B publication Critical patent/CN101839858B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Landscapes

  • Investigating, Analyzing Materials By Fluorescence Or Luminescence (AREA)

Abstract

The invention discloses a nonlinear laser fluorescence spectrum real-time identification method, which comprises the following steps: learning a sample spectrum, testing sample spectral classification, extracting ROI in an interested region, preprocessing the spectrum, extracting the fluorescence spectrum characteristics by discrete curvelet transform, forming feature vectors, constructing i classes of support vector machines, and distinguishing the test results by classes. The invention adopts the classification method of the support vector machines, and does not depend on large sample training, the input vector is the low-frequency coefficient part after curvelet decomposition, the number of training samples is small, the number of the support vectors is greatly reduced, so the operation time is shortened and the method has instantaneity. The second-generation curvelet transform adopted by the invention is based on a new support frame, and can provide high-efficient, stable and nearly-optimal sparse representation for the curve function with strangeness. Compared with the traditional method, the method is more effective and has higher identification rate. The invention can identify the spectrum samples with data format and image format, and has better adaptability.

Description

A kind of nonlinear laser fluorescence spectrum real-time identification method
Technical field
The present invention relates to a kind of real time recognition of fluorescence spectrum, particularly a kind of nonlinear laser fluorescence spectrum real-time identification method.
Background technology
Airborne laser radar produces fluorescence through the ultraviolet band laser excitation offshore spilled oil oil film of emission fixed wave length, collects excited fluorescent through telescope and processes fluorescence spectrum.Owing to contain the kind of fluorescent base matter and the ratio difference of various kinds of substrates in the different petroleum productss; Every kind of matrix all can be launched its distinctive fluorescence spectrum; The fluorescence Spectra of being launched down by the Ultra-Violet Laser excitation of certain wavelength has different intensity and shape usually, therefore can discern the kind of oil spilling according to spectral signature.But, need rich experience, and it is longer to expend time in through human eye identification.
More known fluorescence spectrum real-time identification methods have maintenance sciagraphy (Locality Preserving Projection; LPP), PCA (PCA), core principle component analysis (KPCA) and two-dimentional principal component analytical method (2DPCA) etc.; But a common feature of these methods needs the large sample support exactly; And the training time is long, and identification effect is not very good yet.
Wavelet transformation be a kind of very outstanding, have when strong, the non-stationary signal analytical approach of the function of partial analysis frequently, be widely used in many research fields at present, and obtained effect preferably.Wavelet transformation is an optimal base when expression has the objective function of a singularity, and when representing on the edge of, wavelet transformation and Fourier transform all are not optimal bases.Bent wave conversion not only keeps the multiple dimensioned characteristics of small wave converting method, but also has the anisotropy characteristics, can approach singular curve well, and its some ability in Flame Image Process slightly is superior to wavelet transformation.
The support vector base is the optimal classification face development under the linear separability situation and coming, and it reveals special advantage to solving issue table such as small sample study, non-linear and higher-dimension pattern-recognition.
But do not see public reported about the support vector based non-linear laser fluorescence spectrum real-time identification method that utilizes bent wave conversion at present.
Summary of the invention
For solving the problems referred to above that prior art exists, the objective of the invention is to propose a kind of support vector based non-linear laser fluorescence spectrum real-time identification method that utilizes bent wave conversion, to improve its real-time, discrimination and adaptability.
To achieve these goals; Technical scheme of the present invention is following: a kind of nonlinear laser fluorescence spectrum real-time identification method; Through computing machine the fluorescence spectrum of the ultraviolet band laser excitation offshore spilled oil oil film generation of airborne laser radar emission fixed wave length is handled; To discern the oil product of fluorescence spectrum representative fast and accurately, specifically may further comprise the steps:
A, the oil spilling oil film fluorescence data that the airborne laser-fluorescence radar is obtained are divided into two types: one type is the fluorescence data of the oil product known, and as the learning sample spectrum that sign is arranged; Another kind of is the fluorescence data of unknown oil product, and as the test sample book spectrum that does not have sign;
B, region of interest ROI are extracted: the fluorescence wave band of intercepting spectroscopic data, the equal elimination of all the other wave bands; Its method is to begin to line by line scan from data head, as runs into saturating capacity and empty data are then deleted this part wave band, is only contained the spectroscopic data of fluorescence information at last; The spectroscopic data that extracts region of interest ROI comprises learning sample and test sample book, and to establish the sample class number be i, and the learning sample total quantity is n, and the test sample book total quantity is m;
C, spectrum pre-service: learning sample spectrum and test sample book spectroscopic data that region of interest ROI is extracted, with 0～4 passage totally five passages represent the wave band that extracts, its intensity distributions scope converts in [0,1]; If fluorescence data is a picture format, be bianry image with image format conversion earlier, background is 0, spectral signature is 1, then the unification of image size dimension is transformed to p * p, and carries out spectral position and correct;
D, feature extraction: utilize discrete bent wave conversion to carry out the fluorescence Spectra feature extraction, learning sample spectrum and test sample book spectrum that step C is obtained adopt the fast discrete song wave conversion USFFT based on Unequispaced to decompose, and obtain bent wave conversion coefficient c Jlk, it is 3 layers that Qu Bofen separates yardstick, its process is following:
D1, pre-service is obtained fluorescence data carry out two-dimensional Fourier transform, obtain frequency array
f ^ [ n 1 , n 2 ] - - - ( 1 )
Wherein ,-n/2≤n 1, n 2＜n/2, n are hits;
D2, to arbitrary scale parameter j and direction parameter l, insert
Figure GDA0000142696550000022
to obtain
f ^ [ n 1 , n 2 - n 1 tan θ l ] - - - ( 2 )
In the formula, θ lBe the anglec of rotation
(n 1，n 2)∈P j＝{(n 1，n 2)：n 1，0≤n 1＜n 1，0+L 1j，n 2，0≤n 2＜n 2，0+L 2j}
Wherein, 0≤n 1＜L 1j, 0≤n 2＜L 2j,-π/4≤θ l≤π/4, L 1, L 2Be respectively parallelogram P jThe length of side;
D3, with a parabolic window multiplied by the interpolation object
Figure GDA0000142696550000032
get
f ~ jl [ n 1 , n 2 ] = W ( U ~ ij f ^ ) [ n 1 , n 2 ] - - - ( 3 )
D4, each
Figure GDA0000142696550000034
carried out two-dimentional reverse Fourier transform after, obtain discrete bent wave system number
c jlk = Σ n 1 , n 2 ∈ P j f ^ [ n 1 , n 2 - n 1 tan θ l ] U ~ [ n 1 , n 2 ] e i 2 π ( k 1 n 1 / L 1 j + k 2 n 2 / L 2 j ) - - - ( 4 )
In the formula, k=(k 1, k 2) ∈ Z 2Be the locus
E, formation proper vector: the low frequency coefficient in the bent wave conversion coefficient has comprised important characteristic information, it is made up laggard line translation form proper vector
Figure GDA0000142696550000036
The SVMs of F, i classification of structure carries out learning training respectively to i SVMs: the proper vector that forms after all learning sample conversion is input to respectively in i the SVMs trains separately;
G, Classification and Identification test: the characteristics of low-frequency vector that test spectral is carried out forming behind the bent wave conversion of two generations is input to respectively in i the SVMs that trains, and the result can obtain spectrum discriminator result according to output.
The described spectral position of step C of the present invention is corrected and may further comprise the steps: confirm the position of the current curve of spectrum and the deviation of standard spectrum curve location by the frame scan entire image; Whether deviation is excessive to confirm present image according to preset deviation threshold then, then corrects as surpassing threshold value.
Compared with prior art, the present invention has following beneficial effect:
1, the present invention adopts the sorting technique of SVMs, does not rely on the large sample training, and input vector is the low frequency coefficient part after Qu Bofen separates; Training sample is few, and support vector quantity significantly reduces, thereby shortens operation time; Make method have real-time; Adopt the Pentium D 3.40GHz of Intel processor, the 1G internal memory, its training time is no more than 20ms.
2, the present invention adopt two generations bent wave conversion based on new supporting frame, to having the curvilinear function of singularity, rarefaction representation efficient, that stablize, approach " optimum " can be provided, with classic method relatively, more effective, discrimination is higher.
But 3, the spectrum samples of the present invention's recognition data form and picture format is carried out location determination for view data, effectively solves the problem of image recognition with certain position skew, rotation, has adaptability preferably.
4, the present invention provides a kind of Real time identification laser fluorescence spectrum method, and which kind of material can utilize small sample to discern the curve of spectrum that the airborne laser-fluorescence radar obtains fast and accurately automatically is, like algae, former wet goods.Problems such as poor in timeliness that solution is brought by artificial cognition and accuracy are low are that relevant departments such as China's maritime affairs monitoring marine site situation, especially marine oil spill pollute, and in time formulate the handled scheme, and strong assistance is provided.
Description of drawings
6 in the total accompanying drawing of the present invention, wherein:
Fig. 1 is the FB(flow block) of nonlinear laser fluorescence spectrum real-time identification method.
Fig. 2 is a sample position offset detection algorithm block diagram.
Fig. 3 is the initial fluorescence spectrogram with light diesel fuel sign.
Fig. 4 is the initial fluorescence spectrogram with lubricating oil sign
Fig. 5 is pretreated light diesel fuel fluorescence spectrum figure.
Fig. 6 is pretreated lubricating oil fluorescence spectrum figure.
Embodiment
Below in conjunction with accompanying drawing the present invention is described further.
Embodiment one:
Fig. 1 is a process flow diagram of the present invention, and totally 5 types of fluorescence datas that have the oil product sign are as learning sample spectrum at first from the oil product storehouse, to choose 30 groups, and the laser fluorescence spectrum data of surveying 3 groups of unknown oil products are as test sample book spectrum.
Fig. 3, Fig. 4 have the initial fluorescence spectrogram that has light diesel fuel sign and lubricating oil sign in 30 learning sample spectrum respectively; Extract effective fluorescence wave band as region of interest ROI; The process of extracting area-of-interest is to begin to line by line scan from data head; As run into saturating capacity and empty data are then deleted this part wave band, only contained the spectroscopic data of fluorescence information at last.And with its intensity normalization in [0,1] scope, the spectrum of whole extraction with 0～4 wave band totally five wave bands characterize, as pre-service learning sample spectrum, like Fig. 5, shown in Figure 6.
The gained spectroscopic data is carried out the bent wave conversion of fast discrete according to the step of step D, if adopt 3 layers of decomposition scale, its coefficient is:
c j，k＝0，j＝1；
c j , l , k = 0 , c jlk 2 - c j 2 < 0 , c jlk c jlk 2 - c j 2 &GreaterEqual; 0 , , j = 2,3 ;
To obtain coefficient of dissociation c 1l, c 2lBe combined as proper vector C.
Characteristic proper vector C as proper vector, is sent into the spectrum samples library storage.30 learning samples with all tape identification all carry out the as above processing of step like this.The proper vector that obtains is sent into the spectrum samples library storage, and therefrom sample drawn collects as study.
3 groups of test sample book spectrum are obtained proper vector as test set according to above-mentioned learning sample spectral manipulation method processing.
Make up the SVMs of 5 classifications, aforementioned gained study is collected these 5 SVMs training studies that are input to structure, adopt Sigmoid kernel function K (x here i, x j(α (xy)+β) is as the SVMs kernel function for)=tanh.
Then; 3 samples to be tested that do not have sign are input to 5 SVMs that train with the characteristics of low-frequency vector behind the bent wave conversion according to the method described above; The present invention adopts 1-a-r sorter (One-against-rest classifiers); Promptly between i class and other i-1 class, make up lineoid, each SVMs identifies its mask data separately from other grouped data.
At last, obtain the classification results of fluorescence spectrum according to the output result.
Embodiment two:
Embodiment two adopts view data to classify, and the sample image after region of interest ROI is chosen carries out pre-service: image transitions is a bianry image, and with background be made as 0, spectral signature is made as 1, then with the unified p * p that is transformed into of image size.Shown in Figure 2 is the process that the spectrum picture offset is corrected, and by the pretreated image sequence of frame scan, travels through all sub-pieces; Whether the check image position exceeds off-set value; Again carry out region of interest ROI if the picture position skew is excessive and choose continuation inspection again,, proceed bent wave conversion up to meeting offset conditions; The training of input support vector base finally obtains classification results.

Claims (2)

1. nonlinear laser fluorescence spectrum real-time identification method; Through computing machine the fluorescence spectrum of the ultraviolet band laser excitation offshore spilled oil oil film generation of airborne laser radar emission fixed wave length is handled; To discern the oil product of fluorescence spectrum representative fast and accurately, it is characterized in that: specifically may further comprise the steps:
A, the oil spilling oil film fluorescence data that the airborne laser-fluorescence radar is obtained are divided into two types: one type is the fluorescence data of the oil product known, and as the learning sample spectrum that sign is arranged; Another kind of is the fluorescence data of unknown oil product, and as the test sample book spectrum that does not have sign;
B, region of interest ROI are extracted: the fluorescence wave band of intercepting spectroscopic data, the equal elimination of all the other wave bands; Its method is to begin to line by line scan from data head, as runs into saturating capacity and empty data are then deleted this part wave band, is only contained the spectroscopic data of fluorescence information at last; The spectroscopic data that extracts region of interest ROI comprises learning sample and test sample book, and to establish the sample class number be i, and the learning sample total quantity is n, and the test sample book total quantity is m;
C, spectrum pre-service: learning sample spectrum and test sample book spectroscopic data that region of interest ROI is extracted, with 0～4 passage totally five passages represent the wave band that extracts, its intensity distributions scope converts in [0,1]; If fluorescence data is a picture format, be bianry image with image format conversion earlier, background is 0, spectral signature is 1, then the unification of image size dimension is transformed to p * p, and carries out spectral position and correct;
D, feature extraction: utilize discrete bent wave conversion to carry out the fluorescence Spectra feature extraction, learning sample spectrum and test sample book spectrum that step C is obtained adopt the fast discrete song wave conversion USFFT based on Unequispaced to decompose, and obtain bent wave conversion coefficient c Jlk, it is 3 layers that Qu Bofen separates yardstick, its process is following:
D1, pre-service is obtained fluorescence data carry out two-dimensional Fourier transform, obtain frequency array
Figure FDA0000142696540000011
Wherein ,-n/2≤n 1, n 2＜n/2, n are hits;
D2, to arbitrary scale parameter j and direction parameter l, insert
Figure FDA0000142696540000012
to obtain
In the formula, θ lBe the anglec of rotation
(n 1，n 2)∈P j＝{(n 1，n 2)：n 1，0≤n 1＜n 1，0+L 1j，n 2，0≤n 2＜n 2，0+L 2j}
Wherein, 0≤n 1＜L 1j, 0≤n 2＜L 2j,-π/4≤θ l≤π/4, L 1, L 2Be respectively parallelogram P jThe length of side;
D3, with a parabolic window? multiplied by the interpolation object? get
Figure FDA0000142696540000021
D4, each
Figure FDA0000142696540000022
carried out two-dimentional reverse Fourier transform after, obtain discrete bent wave system number
Figure FDA0000142696540000023
In the formula, k=(k 1, k 2) ∈ Z 2Be the locus
E, formation proper vector: the low frequency coefficient in the bent wave conversion coefficient has comprised important characteristic information, it is made up laggard line translation form proper vector
Figure FDA0000142696540000024
The SVMs of F, i classification of structure carries out learning training respectively to i SVMs: the proper vector that forms after all learning sample conversion is input to respectively in i the SVMs trains separately;
G, Classification and Identification test: the characteristics of low-frequency vector that test spectral is carried out forming behind the bent wave conversion of two generations is input to respectively in i the SVMs that trains, and the result can obtain spectrum discriminator result according to output.
2. a kind of nonlinear laser fluorescence spectrum real-time identification method according to claim 1; It is characterized in that: the described spectral position of step C is corrected and may further comprise the steps: confirm the position of the current curve of spectrum and the deviation of standard spectrum curve location by the frame scan entire image; Whether deviation is excessive to confirm present image according to preset deviation threshold then, then corrects as surpassing threshold value.
CN2010101734358A 2010-05-13 2010-05-13 Nonlinear laser fluorescence spectrum real-time identification method Active CN101839858B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN2010101734358A CN101839858B (en) 2010-05-13 2010-05-13 Nonlinear laser fluorescence spectrum real-time identification method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN2010101734358A CN101839858B (en) 2010-05-13 2010-05-13 Nonlinear laser fluorescence spectrum real-time identification method

Publications (2)

Publication Number Publication Date
CN101839858A CN101839858A (en) 2010-09-22
CN101839858B true CN101839858B (en) 2012-11-07

Family

ID=42743364

Family Applications (1)

Application Number Title Priority Date Filing Date
CN2010101734358A Active CN101839858B (en) 2010-05-13 2010-05-13 Nonlinear laser fluorescence spectrum real-time identification method

Country Status (1)

Country Link
CN (1) CN101839858B (en)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102262081A (en) * 2011-07-07 2011-11-30 公安部第一研究所 Fluorescence spectrum unscrambling method based on matrix analysis
CN102323241B (en) * 2011-07-27 2013-07-03 上海交通大学 One-dimensional spectroscopic data characteristic detection method oriented to optical sensor
US10527546B2 (en) * 2015-06-10 2020-01-07 Saudi Arabian Oil Company Characterizing crude oil using laser induced ultraviolet fluorescence spectroscopy

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004132921A (en) * 2002-10-15 2004-04-30 Yukihiro Toyoda Method of predicting property
US20040206914A1 (en) * 2003-04-18 2004-10-21 Medispectra, Inc. Methods and apparatus for calibrating spectral data
CN1598537A (en) * 2004-08-06 2005-03-23 山东省科学院海洋仪器仪表研究所 Field optical investigating device and detecting method for multiple organic pollutant in sea water
CN101398898A (en) * 2008-10-20 2009-04-01 中国科学院合肥物质科学研究院 Plant leaf identification method based on manifold learning
CN101458213A (en) * 2008-12-23 2009-06-17 潍坊学院 Oil species identification method by sea oil spill concentration auxiliary auxiliary parameter fluorescence spectrum
WO2009080049A1 (en) * 2007-12-21 2009-07-02 Dma Sorption Aps Monitoring oil condition and/or quality, on-line or at-line, based on chemometric data analysis of flourescence and/or near infrared spectra

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2004132921A (en) * 2002-10-15 2004-04-30 Yukihiro Toyoda Method of predicting property
US20040206914A1 (en) * 2003-04-18 2004-10-21 Medispectra, Inc. Methods and apparatus for calibrating spectral data
CN1598537A (en) * 2004-08-06 2005-03-23 山东省科学院海洋仪器仪表研究所 Field optical investigating device and detecting method for multiple organic pollutant in sea water
WO2009080049A1 (en) * 2007-12-21 2009-07-02 Dma Sorption Aps Monitoring oil condition and/or quality, on-line or at-line, based on chemometric data analysis of flourescence and/or near infrared spectra
CN101398898A (en) * 2008-10-20 2009-04-01 中国科学院合肥物质科学研究院 Plant leaf identification method based on manifold learning
CN101458213A (en) * 2008-12-23 2009-06-17 潍坊学院 Oil species identification method by sea oil spill concentration auxiliary auxiliary parameter fluorescence spectrum

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
宋成文 等.同步荧光谱法鉴别海上溢油研究.《船海工程》.2009,第38卷(第3期),160-163. *
陈海菊 等.基于SVM的激光诱导荧光遥感识别海面溢油.《应用能源技术》.2008,(第2期),6-9. *

Also Published As

Publication number Publication date
CN101839858A (en) 2010-09-22

Similar Documents

Publication Publication Date Title
Priyadarshani et al. Automated birdsong recognition in complex acoustic environments: a review
Filho et al. Forest species recognition using macroscopic images
US10360900B1 (en) System and method for taxonomically distinguishing sample data captured from sources
US10699719B1 (en) System and method for taxonomically distinguishing unconstrained signal data segments
CN106653032B (en) Based on the animal sounds detection method of multiband Energy distribution under low signal-to-noise ratio environment
US20030236661A1 (en) System and method for noise-robust feature extraction
CN101551864B (en) Image classification method based on feature correlation of frequency domain direction
CN101299237B (en) High spectroscopic data supervision classifying method based on information quantity dimensionality sequence
CN102509120B (en) Supervised image segmentation method for hyperspectral image based migration dictionary learning
CN103839078A (en) Hyperspectral image classifying method based on active learning
CN104392242A (en) Method for cooperatively classifying perceived solid wood panel surface textures and defects by feature extraction and compressive sensing based on dual-tree complex wavlet
CN112289326B (en) Noise removal method using bird identification integrated management system with noise removal function
CN107219510B (en) Radar target identification method based on unlimited largest interval linear discriminant projection model
CN112750442B (en) Crested mill population ecological system monitoring system with wavelet transformation and method thereof
Rasmussen et al. Automatic detection and classification of baleen whale social calls using convolutional neural networks
CN101839858B (en) Nonlinear laser fluorescence spectrum real-time identification method
CN105512622A (en) Visible remote-sensing image sea-land segmentation method based on image segmentation and supervised learning
CN115661649B (en) BP neural network-based shipborne microwave radar image oil spill detection method and system
US20220390374A1 (en) Method for extracting raman characteristic peaks employing improved principal component analysis
CN102902984B (en) Remote-sensing image semi-supervised projection dimension reducing method based on local consistency
Song et al. Red blood cell classification based on attention residual feature pyramid network
Nieto-Mora et al. Systematic review of machine learning methods applied to ecoacoustics and soundscape monitoring
CN104331711A (en) Multi-scale fuzzy measure and semi-supervised learning based SAR (Synthetic Aperture Radar) image identification method
CN111488851A (en) Traceability detection method, device, equipment and medium for fruit production place
CN116682015A (en) Feature decoupling-based cross-domain small sample radar one-dimensional image target recognition 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
C14 Grant of patent or utility model
GR01 Patent grant
C53 Correction of patent for invention or patent application
CB03 Change of inventor or designer information

Inventor after: Li Ying

Inventor after: Liu Yu

Inventor after: Chen Peng

Inventor before: Li Ying

Inventor before: Chen Peng

COR Change of bibliographic data

Free format text: CORRECT: INVENTOR; FROM: LI YING CHEN PENG TO: LI YING LIU YU CHEN PENG